6th WGNE workshop on systematic errors in weather and climate models

This study assesses the performance of the German Meteorological Services (DWD) Consortium for Small-scale Modelling (COSMO) model in predicting rainfall over West Africa for six high-impact rainfall events from 2015 to 2020 during the boreal summer monsoon season. The study aims to investigate the synoptic forcings modulating daily rainfall variability during this period. The results show that COSMO adequately simulates the spatiotemporal variability of rainfall distribution over West Africa, albeit with distinct inherent biases in different seasons and years. Although there is a strong correlation between 10 and 12oN, COSMO generally shows a decreasing skill in producing spatial rainfall distribution as rainfall amounts tend toward the 75th percentile and above. Furthermore, areas of heavy rainfall are mostly located about 100-250 km southwest of the core of the Africa easterly jet (AEJ), but with varying unique attributes from 700 to 600 hPa levels. On average, these areas frequently coincide with decreasing mean sea level pressure of at least 0.6 hPa and increasing convective available potential energy (CAPE) of at least 700 J Kg-1. The Africa Easterly Wave (AEW) trough is always located to the east of these areas, with less interaction with storms east of the prime meridian.

Results also show that incorporating the Fractions Skill Score metric could help operational forecasters decide at what scale to issue a severe weather alert using fraction forecasts. Irrespective of the year and season, COSMO can reproduce the atmospheric dynamics modulating the daily rainfall variability, in addition to capturing the daily propagation of the AEW trough, and the core of the AEJ. Finally, COSMO may not generally be adept at producing heavy rainfall, operational forecasters may be able to identify likely areas of heavy rainfall based on the position of the AEJ core, areas of the least falling pressure, and increasing CAPE from COSMO.

Winter precipitation (December to March) over the north Indian region comes primarily through eastward propagating synoptic scale mid-latitudinal cyclonic systems (Western Disturbances) embedded in large-scale sub-tropical westerlies. This precipitation is a crucial source for accumulation of snowfall over the Himalayan terrains feeding major north Indian rivers and an important irrigational source for Rabi crops. However, the sparsity of in-situ observations over this region in conjunction with complex topography highlights the uncertainties in available coarse resolution precipitation products and emphasizes the necessity of finer grid scaled regional climate models for accurate simulation of precipitation and associated spatio-temporal variability as well as regional dynamics. In the present study, we investigate the performance of a high-resolution Weather Research and Forecasting (WRF) model through the evaluation of its sensitivity to three convective physical parameterization schemes (CPSs), namely, Betts-Miller-Janjic (BMJ), Kain-Fritsch (KF) and Grell-Freitas (GF) for the simulation of seasonal winter (DJFM) precipitation over the north Indian region during the period 2001-2016. The sensitivity of winter seasonal precipitation to different CPSs, in the WRF model configured on two two-way nested domains (15km and 5km), has been validated using gauge-based (IMD) and satellite (TRMM) observations as well as with the recently released high-resolution reanalysis dataset, IMDAA. Furthermore, the elucidation of associated regional dynamical features and underlying processes simulated in the model has been carried out using the reanalyses, IMDAA and ERA5. Our findings underpin the sensitivity of model-simulated north Indian seasonal winter precipitation to different parameterizations. Detailed results will be discussed.

UGROW is an ECMWF cross-departmental project focused on Understanding systematic error GROWth from hours to seasons ahead. UGROW-JET, a UGROW sub-project, focuses on biases in the representation of the Pacific sub-tropical jet stream, which plays a key role in the tropical-extratropical teleconnections. A main issue in the representation of the Pacific sub-tropical jet stream in the ECMWF forecasts is a westward shift of the eastward extension of the jet stream. The bias increases with lead time up to week 4 and is largest in January. A series of sensitivity experiments, where part of the atmospheric circulation is nudged towards ERA5, suggests that there is not a unique source for this bias. However, model errors in the high latitudes seem to be a major contributor.

The Met Office has been running their coupled ocean-atmosphere NWP system since May 2016, alongside their atmosphere-only version of the model. Comparison between the models with Real-time Multivariate MJO (RMM) index shows that both models are similarly skilful in MJO prediction. While both models perform well, the coupled model predicts faster MJO propagation than atmosphere-only model and observations. In this paper, we examine potential sources of this erroneous propagation.

Observations show that SST and subsequently sea surface fluxes influence MJO convection in the tropics. Active MJO composites reveal SST biases in the coupled model in two regions within the first 24 hours of the forecast. When forecasts are initialized during active MJO convection over the Indian Ocean (RMM phase 1), the central Maritime Continent experiences warm SST anomalies of order 0.1°C, likely due to shortwave flux biases in the atmospheric model. Within the next 7 lead days, the coupled model produces excessive MJO convection in that region. When active MJO convection is over the Maritime Continent (RMM phase 4), the equatorial Indian Ocean region experiences cold SST anomalies at lead day 1, likely due to mixed-layer biases in the ocean model. Consistently, the coupled model produces excessive suppression of convection in that region within the next 7 lead days. Idealized case studies on the atmosphere-only version of the model are designed to test SST-convection feedback in these regions. Sources of initial SST biases are examined with biases in surface fluxes and upper ocean processes, e.g. vertical mixing.

It appears that, currently, systematic biases in the coupled model detract from any potential increases in MJO predictability due to the inclusion of coupled ocean-atmosphere processes. Hence, future focus should be on reducing these systematic biases.

The Korean Integrated Model (KIM) was developed for global weather forecasting at the first Phase (2011–2019) of the Korea Institute of Atmospheric Prediction Systems (KIAPS), and it has become the operational model of the Korea Meteorological Administration (KMA) since April 2020. To improve the predictability beyond 2 weeks, it is necessary to better represent the physical process and interaction between the atmosphere and surface. Therefore, the new KIAPS Phase 2 project (2021–2026) aims to advance the land surface model and to couple the ocean/seaice/wave/river-routing models to the operational KIM.
At the first stage (2020–2022), the KIM was newly coupled to ocean (the Nucleus for European Modelling of the Ocean; NEMO) and seaice (Sea Ice modelling Integrated Initiative; SI3) models by means of Model Coupling Toolkit (MCT) coupler. The evaluation result showed that the performance of the coupled KIM is promising on medium-range forecast as well as seasonal simulations when compared to the uncoupled version. The community Noah land surface model with multi-parameterization (Noah-MP) became optional to explicitly consider the geographical processes within snow and canopy layers. In addition, the river-routing (Catchment-based Macro-scale Floodplain; CaMa-Flood) and wave (Wave Watch III; WW3) models were coupled at a preliminary stage and their sensitivity are being explored. In the conference, the status and future plan of the coupled KIM will be presented in detail.

We use data assimilation statistics and Forecast Sensitivity Observation Impact (FSOI) of drifting buoy (drifter) surface pressure observations in the Northeastern Pacific to provide information on biases in the US Navy’s Global Environmental Model (NAVGEM) and hybrid 4D-VAR data assimilation (DA) system, and how these biases may prevent the effective use of these observations. To account for the fact that model biases may be different for e.g., high pressure systems vs. cyclones, we separate the observations into quartiles, with the lowest (first) quartile populated by the lowest surface pressure values and the highest (fourth) quartile populated by the highest surface pressure values. Innovation (observation minus background) statistics indicate that the model has a conditional bias such that it systematically underestimates high pressures. FSOI indicates that almost all the beneficial impact on forecast error reduction comes from observations in the lowest quartile, with near neutral impacts from observations in the other quartiles. Case studies indicate that these beneficial observations occur in dynamically active regions such as cyclones, fronts and atmospheric rivers. When considering observation impact as a function of the hour within the six-hour DA cycle, the net impact of observations in the lowest quartile is always beneficial, with the largest beneficial impacts coming from observations take late in the update cycle (similar to what was found for satellite observations in McNally 2019). For observations in the highest quartile, the observations taken at the beginning of the DA window are beneficial, but the observations taken in the second half of the DA window are non-beneficial. Taken together, the innovation statistics and FSOI results suggest that conditional model bias for high surface pressures is preventing the DA system (currently in “strong-constraint” mode) from using these observations effectively when taken at the end of the DA window (when model bias are most pronounced).

How subtropical marine low cloud cover (LCC) will respond to global warming is a major source of uncertainty in future climate change. While low clouds are maintained by a delicate balance of complex physical processes, a predictive index for LCC can be obtained from large-scale temperature and water vapor profiles, of which the responses to warming are more reliable. It is well known that the estimated inversion strength (EIS) is a good predictive index of LCC. However, EIS has a serious limitation when applied to evaluate LCC changes due to warming: the LCC decreases despite increases in EIS in future climate simulations of global climate models (GCMs). In this study, using state-of-the-art GCMs, we show that the recently proposed estimated cloud-top entrainment index (ECTEI) decreases consistently with LCC in warmer sea surface temperature (SST) climates. For the patterned SST warming predicted by coupled GCMs, ECTEI can constrain the subtropical marine LCC feedback to $-$0.41% $\pm$ 0.28% K$^{-1}$ (90% confidence interval), implying virtually certain positive feedback. ECTEI physically explains the heuristic model for LCC changes based on a linear combination of EIS and SST changes in previous studies in terms of cloud-top entrainment processes.

Climate simulations often need to be adjusted (i.e., corrected) before any climate change impacts studies. However usual bias correction approaches do not differentiate the bias from the different uncertainties of the climate simulations: scenario uncertainty, model uncertainty and internal variability. In particular, in the case of a multi-run ensemble of simulations (i.e., multiple runs of one model), correcting, as usual, each member separately, would mix up the model biases with its internal variability. In this study, two ensemble bias correction approaches preserving the internal variability of the initial ensemble are proposed. These “Ensemble bias correction” (EnsBC) approaches are assessed and compared to the approach where each ensemble member is corrected separately, using precipitation and temperature series at two locations in North America from a multi-member regional climate ensemble. The preservation of the internal variability is assessed in terms of monthly mean and hourly quantiles. Besides, the preservation of the internal variability in a changing climate is evaluated. Results show that, contrary to the usual approach, the proposed ensemble bias correction approaches adequately preserve the internal variability even in changing climate. Moreover, the climate change signal given by the original ensemble is also conserved by both approaches.

Tropical cyclones (TC) are one of the deadliest weather hazards which affect a large part of the coastal areas of the world. The North Indian Ocean (NIO) comprises two basins namely; Bay of Bengal and Arabian Sea on the eastern and western side of the Indian peninsula. The TC occurring in the NIO are only 10% of the world, but their effect is higher due to a larger population density around the coastal regions. Global Numerical Weather Prediction (NWP) models, including ensemble prediction systems (EPS), are now routinely used to predict tropical cyclone (TC) tracks and intensity. However, due to computational restrictions the EPSs are usually of a coarser resolution which results in lack of proper representation of air-sea interactions and underestimation of convective cloud process. These problems account for the systematic errors in the intensity forecasts for TC by EPSs and it is usually under predicted. Also, due to the The biases in the model predicted maximum sustained winds (MSW) and central pressure (CP) are large when TCs are very intense. Many statistical methods are available for the removal of biases in the TC intensity forecasts. Recently many machine learning (ML) methods are being used for bias correction (BC) for calibrating the TC parameters. The increased accuracy of intensity forecasts can lead to better disaster mitigation efforts and reduction in losses incurred due to TCs.
This paper describes the suitability of ML techniques to reduce error in mean TC intensity forecasts obtained from the NCMRWF Ensemble Prediction System (NEPS-G) over the NIO. Three different ML techniques namely: Random Forest (RF), Multivariate Liner Regression (MLR) and eXtreme Gradient Boost (XGB) have been tried for the BC of the ensemble mean forecast of MSW and CP. The study is based on 20 TC cases during 2018-21 and shows that ML based models using XGB and RF techniques are superior to the MLR method. The model was trained and verified against best track (BT) data received from India Meteorological Department (IMD). The reduction in RMSE in mean MWS and CP while using the best algorithm are 33 and 55 % respectively. The correlation coefficient increases from 0.58 to 0.93 for CP and from 0.60 to 0.75 for MSW. In addition the t-test showed that the reduction in the absolute error was significant at 95% CI for both XGB and RF.

The Global Precipitation Experiment (GPEX) is aimed at reducing model errors in global coupled models using an integrated observations and modeling strategy and targeting priority processes and phenomena that are key to enhancing precipitation predictability and improving precipitation predictions and projections. GPEX is envisioned to be a multi-year activity with U.S. and international collaboration.

GPEX major research components:
1) Predictability and Process Studies to advance understanding of precipitation predictability sources and limits; and improve processes key to global and regional precipitation distributions and variability, and the representations of the processes in coupled models. This will be achieved through process studies of intensive and innovative field campaigns and hierarchical model experiments.
2) Optimizing observations and datasets (including satellite, radar and in-situ) for prediction initialization, evaluation and process understanding. This includes enhancement of existing observation networks, applications of new observing technologies to fill in observation gaps in key regions, targeted observations focusing on specific high impact events, and integrated datasets for process-level understanding and model diagnosis.
3) Improving global coupled model systems: This includes improved physics, high-resolution modeling, and novel approaches to represent processes critical to precipitation (e.g., applications of AI/ML) in global coupled models, development of coupled data assimilation capability, and creative numerical configurations.

Agencies of USGCRP and USCLIVAR (e.g., DOE, NASA, NOAA, NSF) are interested in addressing key scientific gaps in reducing model errors to improve precipitation prediction through GPEX. The international WCRP adopted GPEX as a cross-WCRP activity, and set up a GPEX Tiger Team in May 2022 to develop a strategy as to how to advance the GPEX activity within WCRP. The discussions in this WGNE workshop will be helpful to inform the GPEX planning and coordination of ongoing and future research activities.

In recent years, studies have put forth various theories and findings on the role of equatorial Rossby waves (ERW) in the subseasonal-to-seasonal (S2S) predictability of the Indian Ocean (IO). While much of the scientific literature uses data from in-situ, satellite, and/or reanalysis datasets, this study focuses on reforecast fields from the European Centre for Medium-Range Weather Forecasting’s (ECMWF) S2S dataset in order to evaluate the model’s predictive skill in representing ERWs as well as the associated OHC variations. This work provides a unique and objective methodology to calculate and evaluate the predictability of ERWs from model forecast data, which, to the author’s knowledge, is the first of its kind to do so. Our results indicate that the model forecasts ERWs with relatively high skill (anomaly correlation > 0.5 for all lead times), indicating they are a key source of oceanic subseasonal predictability at extended lead times. Relationships between model forecasts of sea surface heights (SSHs) and OHC will be discussed, and details will be presented to determine implications for air-sea interactions and sources of model biases. Analysis of the wavenumber-frequency spectra for the IO indicates reduced power for equatorial Kelvin waves in forecasts, which might play a role in the reduced forecast skill of ERWs as a function of model lead time. Results on subseasonal wind scale biases that are hypothesized to be responsible for this behavior will also be presented.

Despite the importance of monsoon rainfall to over half of the world's population, many climate models of the current generation struggle to capture the major features of monsoon systems. Understanding the sources of such errors requires the combination of various modelling techniques and sensitivity experiments of varying complexity. Here, we demonstrate how such analysis can shed light on the way in which monsoon errors develop, their local and remote drivers and feedbacks.
We show that error patterns in circulation and rainfall over the Asian summer monsoon (ASM) region in Met Office models are similar between multidecadal climate simulations and seasonal hindcasts initialized in spring. Analysis of the development of these errors on both short-range and seasonal timescales following model initialization suggests that both the Maritime Continent and the oceans around the Philippines play a role in the development of East Asian summer monsoon errors, while errors over the Equatorial Indian Ocean (EIO) region are associated with circulation errors over India and the strengthening and extension of the westerly jet across Southeast Asia and the South China Sea into the West Pacific. We also find that the Maritime Continent plays an important role in the development of wind errors that force sea surface temperature (SST) errors around the Indonesian islands and in the eastern Indian Ocean. The EIO itself plays a key role, however, in the further development of an east-west SST dipole error, and we hypothesize that the atmosphere circulation errors in these two regions combine with an ocean circulation response in a coupled feedback that affects the ASM as a whole. Regional modelling with various lateral boundary locations helps to separate local and remote contributions to the errors, while regional relaxation experiments shed light on the influence of errors developing within particular areas on the region as a whole.

Lightning strikes are a major destructive factor in thunderstorms in the Indian subcontinent. Therefore, understanding the long-term variation of lightning Flashes (LF) and its relationship to meteorological variables that influence lightning occurrence is critical. A long-term lightning data from 1998 to 2013 has been utilized to study the relationship between LF with that of meteorological and climatic parameters. The statistical analysis shows that lightning flash rate follow the CAPE in pre-monsoon season. In pre-monsoon season high CAPE is found over eastern coastal region (1750-2250 J/Kg) and during monsoon season high over northern region (1250-1500 J/Kg) as compared to winter and post-monsoon season. Surface relative humidity follow more or less similar pattern as LF. The spatio-temporal distribution shows that most of the lightning flashes took place in the months of MAM (Pre-monsoon) (0.35-0.45 flashes/ sq. km /day) in northeast region covering states like Assam, Meghalaya and Tripura whereas in the month of JJA (Monsoon) LF area shift from northeast to northwest part of India covering states like Jammu & Kashmir, Himachal Pradesh and some parts of north Punjab. During post-monsoon, the entire eastern Ghats and Kerala (0.10-0.15 flashes/sq. km/day) experience high LF. In winter, the north western region got the highest lightning flashes (0.05-0.15 flashes/ sq. km /day). TCWV is found to be higher in eastern coast region of India in all the season majorly in the monsoon (JJA). During pre-monsoon, high TCWV value (50-70 Kg/m2) is found over the south-eastern coastal zone of India. Surface relative humidity is unevenly distributed over India showing higher value in the monsoon season (60-80%) and least in the winter season. Statistical analysis is performed to see seasonal correlation between lightning flashes and various climatic parameters. The PCA result show the CAPE, TCWV and RH are significant ally well corelated with LF over study domain.
Keywords: Lightning flashes, CAPE, RH, TCWV

Due to increasing the intensity, life period, and duration of maximum intensity of the intense tropical cyclones over the Bay of Bengal region (Singh et al., 2021). It is a need and scope to improve the forecast accuracy of intense tropical cyclones that made landfall. In this study, the performance of the WRF model was evaluated at different horizontal resolutions of 1.667 km, 3 km, and 5 km to forecast the intensity and structure of the super cyclone Amphan, which developed over the Bay of Bengal region in May 2020. The numerical experiments are carried out with the ARW-WRF model by using double nested domains with fine resolutions of 3 km and 5 km and three nested domains with a fine resolution of 1.667 km under the moving nested option. The initial and lateral boundary conditions for the simulations are derived from available high-resolution (25 km) NCEP operational Global Forecast System (GFS) analysis and forecasted datasets. The best-fit track datasets from the India Meteorological Department (IMD) are used to validate the predicted track, intensity, rapid intensification, and structures of the super cyclone Amphan. Results show that the track, landfall (position and time), and intensity in terms of minimum sea level pressure (MSLP) and maximum surface wind (MSW) of the storm are well predicted using the high-resolution WRF model. The structure of the storm is also compared to available observations in terms of relative humidity, water vapor, maximum reflectivity, and temperature anomalies. Finally, the results demonstrated that increasing horizontal resolution is not only sufficient to improve the forecast of maximum intensity and rapid intensification of the storms and hence needs to parameterize the physical processes at 1.667 km horizontal resolution. It is also expected that using proper data assimilation techniques and microphysical parametrization schemes with more number of cases across the region will provide a better forecast of the storms.

The simulated Northern Hemisphere winter stationary wave (SW) field is investigated in 11 Subseasonal-to-Seasonal (S2S) prediction project models. It is shown that while most models considered can well simulate the stationary wavenumbers 1 and 2 during the first 2 weeks of integration, they diverge from observations following week 3. Those models with a poor resolution in the stratosphere struggle to simulate the waves, in both the troposphere and the stratosphere, even during the first 2 weeks. Focusing on the tropospheric regions where SWs peak in amplitude reveals that the models generally do a better job in simulating the northwestern `Pacific stationary trough, while certain models struggle to simulate the stationary ridges in both western North America and the North Atlantic. In addition, a strong relationship is found between regional biases in the stationary height field and model errors in simulated upward propagation of planetary waves into the stratosphere. In the stratosphere, biases are mostly in wave 2 in those models with high stratospheric resolution, whereas in those models with low resolution in the stratosphere, a wave 1 bias is evident, which leads to a strong bias in the stratospheric mean zonal circulation due to the predominance of wave 1 there. Finally, biases in both amplitude and location of mean tropical convection and the subsequent subtropical downwelling are identified as possible contributors to biases in the regional SW field in the troposphere.

The Japan meteorological agency (JMA) has been improving the Global Spectral Model (GSM), which is in operation for short- and medium-range global forecasts. In this presentation, we will discuss the improvements of physics parameterizations of the GSM that we are working on to reduce the systematic error errors in energy budget, temperature, and precipitation, with a focus on the error compensation problems.

The GSM has the following characteristic systematic interrelated errors over almost all periods. The tropospheric atmosphere is too cold and too dry. The precipitation amount in the tropics is excessive. The cloud ice content is insufficient in the upper troposphere, particularly in areas of high convective activity. The top-of-the-atmosphere outgoing longwave radiation (OLR) is excessively emitted by up to 30 W/m2 in the monthly mean, mainly in areas of high convective activity. The sea surface latent and sensible heat fluxes from the sea to the atmosphere are excessive over the inactive convective regions.

We are simultaneously improving several physics schemes to reduce these systematic errors. This presentation will show one attempt, the improvement of the convection and the sea surface processes to solve the error compensation problem. The lack of cloud ice content is one of sources of the excess OLR and the cold bias, which is compensated by the excess precipitation and heat supply from the sea surface. This bias in cloud ice content is related to the detrainment and the auto-conversion process in the convection scheme, and adjusting these processes increased the cloud ice content. However, since the temperature error is worsened by those adjustments, it was necessary to modify the bulk formulations of the sea surface flux to reduce the latent heat flux bias.

Predicting the rapid intensification and structures of extremely severe cyclonic storms (ESCSs wind speed more than 90 knots) has become challenging for both researchers and the forecast community and is a key research area to improve the forecast of these parameters. It is expected that using a high resolution (horizontal and vertical) model with improved initial conditions will provide a better forecast of the storms. The assimilated observations are satellite radiances (AMSU-A, AMSU-B, HIRS, MHS) and PREBUFR observations in the WRF-3DVAR. In the study, regional background error statistics are used in data assimilation. The double nested domain of the WRF model with the finer resolution was considered about 9 km and 6 km and the vertical levels were about 73. Forecasted intensity, structure, rapid intensification, rainfall, reflectivity, and storm tracks were compared with available observations [India Meteorological Department (IMD) best-fit track, AWS, doppler weather radar (DWR)]. The results suggested that the WRF model is in good agreement to forecast the rapid intensification and structures of the storms with the available observations. The intensity in terms of the maximum wind speed of the cyclone was well captured. Even though RI and MSW were under-predicted by the model at 9 km horizontal resolution and improved the intensity and rapid intensification at 6 km horizontal resolution. The results show that increased horizontal resolution improves the forecast of intensity and rapid intensification of the extremely severe cyclonic storms.

Presence of large systematic errors in the simulation of basic state is one of the major drawbacks in monsoon modeling. Despite having the continuous efforts from modelling community, the errors still persist and needs better understanding. This study aims to understand the crucial physical processes responsible for precipitation variability at sub seasonal timescales during the boreal summer monsoon season over the Indian region. For this purpose, we have used newly generated long term (1981-2020) Indian Monsoon Data Assimilation and Analysis (IMDAA) reanalysis product. In this work, process-oriented diagnostics are employed onto the reanalysis data to examine the role of key physical processes in precipitation evolution. In addition, convective transition statistics are also employed to understand the sensitivity of moisture profiles in precipitation variability. Results are validated using the ERA5 reanalysis and the relative roles are quantified.
Our examination suggests that despite having systematic errors in key variables responsible for monsoon convection, several aspects (mean state, vertical structure, poleward propagation etc) of the sub-seasonal variability are captured well in the reanalysis data, which is encouraging. Nevertheless, diagnostics also reveal that moisture-convection feedback mechanism is relatively weaker in IMDAA reanalysis compared to ERA5. Our study not only highlights the need for process- based diagnostics in checking the fidelity of IMDAA reanalysis but also indicate the merit and demerits of IMDAA reanalysis for better understanding of the monsoon processes.

Many operational Numerical Weather Prediction (NWP) centers are now running models in the 2–5-km grid-size range that can steadily increase further in coming years. However, as the horizontal scale reduces and forcings become more complex at mesoscales. For instance, the divergence caused by the diabatic heating within convective systems generates additional kinetic energy through enhanced baroclinicity and thermally direct circulation. Further, the dynamics underlying the mesoscale spectrum and its -5/3 power-law dependence on the horizontal wavenumber are a matter of ongoing debate. Nevertheless, the correctness of model performance is assessed through the spectral slopes of horizontal fields. Hence, the motivation of this work lies in the existing need to evaluate energy spectra from high-resolution NWP model at National Centre for Medium Range Weather Forecasting (NCMRWF). A regional version of NCMRWF Unified Model (NCUM) is at 4 km resolution, covering entire Indian region, is run operationally and generates three-day forecasts based on 00 UTC and 12 UTC initial conditions. The horizontal velocities from the model forecasts are split into divergent and rotational components which are then compared at horizontal scales below 500 km and various depths in the troposphere. Furthermore, the relative contribution of divergent and rotational energy to total kinetic energy provides insights of main dynamical agent implicit in the mesoscale energy spectrum. Therefore, this study has strong implications for future high-resolution model development at NCMRWF

UGROW is an ECMWF cross-departmental project focused on Understanding systematic error GROWth from hours to seasons ahead. In this sub-topic of the UGROW project we have investigated biases in the lower to mid-tropospheric temperature during the northern hemisphere summers. The bias was investigated across different-time scales and with a range of diagnostic tools. The bias peaks around 700 hPa and grows fastest during the first days of the forecast. The bias mainly appears over land masses in early forecast ranges and has a maximum over eastern Asia. Despite being robust both in terms of day-to-day and year-to-year variability, the investigations so far have not pointed to a clear model error source. The aim of this report is to document the findings about this specific bias during the UGROW project, which will serve as a starting point for future investigations.

Korea Institute of Atmospheric Prediction Systems KIAPS has recently developed a global NWP model, Korean Integrated Model (KIM) and it has been in operation by Korea Meteorological Administration since April 2020. KIM showed comparable performance to other operational models for the last two years, but the verification is mainly focused on main forecast products or prognostic variables. In this study, we validate cloud simulation in KIM using various observation data in order to examine systematic bias in cloud prediction and evaluate cloud representation in KIM. Verification of cloud cover is one primary target. However, vertical distribution of clouds/precipitations and hydrometeors are also in great attention because the representation of cloud is still one of the most uncertain aspects of global NWP and climate models. Diurnal cycles, resolution sensitivities, and conservation issues are also discussed. In this evaluation, model experiments are not confined in medium-range forecast but also include long-term simulations to bring comprehensive diagnosis on cloud simulation and its impact on model uncertainties.

Abstract
Despite many significant improvements in numerical weather prediction models including major improvements in the model physics and resolution, these models are still having systematic biases. For increasing the reliability or accuracy of a forecast it is essential to remove these biases by using a process called post-processing. There are many methods available to remove these systematic errors from a model, for example by applying statistical post-processing algorithms. In the current study, we have attempted to correct the bias in the maximum temperature (Tmax) forecasts obtained from the NCMRWF’s deterministic and ensemble prediction system by applying Decaying Average Bias Correction (BCDA). This statistical post-processing method applies an adaptive [Kalman filter type (KF)] algorithm to accumulate the decaying averaging bias. One of the main advantages of both methods is that they do not require a large amount of past data for calibration and they take into account the most recent behaviour of the forecasting system. This maximum surface temperature forecast selected from MAM 2017-2022 was carried out in order to decide upon the method giving the best temperature forecast.
The bias correction method is applied to the ensemble forecast also. The mean of the TMAX forecasts from NEPS is bias-corrected (1st moment) using the technique of Decaying Average. This method does not lead to any correction in the spread of the EPS. Therefore the method of Variation Inflation is used to correct the spread (2nd moment) of NEPS. The forecasts of TMAX from the raw and the bias-corrected data are then compared for March to May 2019 using standard verification metrics for probabilistic forecasts like Brier Score (BS), Brier Skill Score (BSS), ROC and Reliability Diagrams as well as the Value score. A case by case comparison of several heatwave cases from March to May 2019 is also performed in order to assess the day-to-day performance of the model. This bias correction method shows improvement in heatwave forecasting.

Keywords: Post-processing, Heatwaves, Decaying average, Bias-Correction, Variation inflation

Anthropogenically-forced trends substantially impact variables like near surface temperature, even on seasonal timescales. This forcing can therefore be a source of predictive skill for seasonal forecasts, however incorrect representation of forced trends, could also be a source of systematic error in forecasts.

In this study, we evaluate trends in 2m temperature in a set of hindcasts performed with operational seasonal forecasting systems in the Copernicus Climate Change project. These forecasting systems in some sense bridge the gap between high-resolution numerical weather prediction and coarse-resolution climate model simulations and contain some representation of anthropogenic forcing. We show that the majority of the models overestimate trends with respect to reanalysis in global-mean warming, particularly in the tropics, hence, a source of model error. We also find that the trend biases tend to increase slightly with lead-time. Finally, we speculate on some of the causes of these discrepancies and differences across models.

The Madden-Julian oscillation (MJO) is the leading source of the intraseasonal variability in the tropics and is known to affect weather and climate not only in the tropics but also extra-tropics. Therefore, performances of MJO prediction and simulation has been considered one of the factors indicating the performance of the subseasonal climate prediction in numerical models. The Korean Institute of Atmospheric Prediction System is making various efforts to extend the forecast range to 30 days in the Korean Integrated Model (KIM), which is the operational global numerical weather prediction model of the Korean Meteorological Administration for the medium-range forecasting. Continuous model evaluation and diagnosis are required to provide appropriate feedback for the development and improvement of weather and climate prediction models. In this study, simulations of KIM with a low horizontal resolution are integrated for 30 days over several years in order to understand the characteristics of the MJO simulation in the current model. Initial conditions and boundary conditions of KIM are from ERA5 reanalysis data. First all, we estimate the simulation skill of the MJO using Real-time Multivariate MJO index, and then evaluate the propagating process of MJO and mean state bias, and examine the impact of systematic bias on the MJO simulation in the KIM.

This study compares the performance of three land surface models (LSMs) in the framework of the Korean Integrated Model (KIM). The operational KIM employs the Noah LSM, which is a relatively simple model but often hinders explicit computation of the additional land surface process. To better represent the land surface processes, two advanced LSMs are introduced to the KIM; 1) Noah with multiparameterization option (NoahMP) and 2) Community Land Model (CLM). Their energy balance and bio-geochemistry schemes were consistently updated by the LSM research community based on the latest theory. They also applied a dynamic/demography vegetation, lake/urban model, and more fine snow/soil layers. Replacing Noah with advanced LSMs may significantly affect the current KIM's performance, and a test is necessary.
The comparison analysis focuses on surface energy variables, and spatial and temporal variations are examined in long-term simulation. Compared to the climatological data in the KIM-Noah, the surface albedo is satisfactorily simulated by the CLM, while the NoahMP highly underestimates it from a global aspect. In particular, the NoahMP's albedo is about 0.05 lower at high latitudes in the cold season, thereby overestimating the surface temperature in the northern hemisphere. Despite similar surface albedo with the KIM-Noah, turbulent heat fluxes are apparently lower in the CLM, which implies that their land surface processes are systematically different. The consequent impacts of replacing LSMs on the atmosphere will also be discussed with the detailed feature of each LSM and coupling expectation.

The Global Ensemble Prediction System (GEPS) operated by the Japan Meteorological Agency (JMA) currently incorporates an atmospheric model with a two-tiered sea surface temperature (SST) approach for lower boundary conditions. This technique indirectly represents atmosphere-ocean interaction by combining SSTs prescribed as persisting anomalies from climatological SSTs and SSTs operationally precomputed using JMA’s seasonal EPS. In this study, as a next step, feasibility for atmosphere-ocean coupling of GEPS was assessed toward the incorporation of more directly representative atmosphere-ocean interaction. The atmosphere-ocean coupling experiments using an eddy-permitting ocean model showed improved forecast skills mainly in the tropics and summer hemisphere as expected, but also showed some deterioration due to initial shocks, insufficient ocean-model resolution in the mid-latitude and cold SST bias. Verification results of the experiments and our approach to tackle the issues will be shown in the poster presentation.

The global and ensemble models of ECMWF, UM(Unified Model), and KIM (Korean Integrated Model) were used to generate a multi-model ensemble prediction system. KIM is KMA’s new generation model which is based on the cubed sphere grid system and has 12km horizontal resolution and 91 vertical levels.
The weights which were given inversely proportional to the error of each model were used to calculate the ensemble weighted mean field.
The ensemble weighted mean field showed better performance than any other member model. As increasing the ensemble member model, it was found that the accuracy of the ensemble mean was more improved.
The performance improvement effect of the multi-model ensemble was more significant in temperature, wind, and humidity fields than in 500 hPa height and mean sea level pressure.
The bias of the multi-model ensemble was close to the intermediate value between the maximum bias and the minimum bias of member models and was smallest when the deviations of the member models were evenly distributed without being biased in one direction.

The term jet stream generally refers to a narrow region of intense winds near the top of the troposphere. Along the jet stream, instabilities and waves can develop into synoptic-scale systems, or midlatitude cyclones, which thus makes it critical for atmospheric development and predictability. Furthermore, their key role in cyclogenesis means they can be linked with atmospheric rivers and warm conveyor belts, yielding high-impact weather including extreme precipitation and flooding, severe winds, and ocean waves. Given the impacts associated with the jet stream, it is therefore important that numerical weather prediction (NWP) systems can accurately forecast the magnitude and the structure of jet streams.

Observations gathered during the Atmospheric River Reconnaissance (AR Recon) observational campaign provide a unique opportunity to investigate the structure of the North Pacific jet stream. In each winter season, AR Recon uses research aircraft to probe (with dropsondes) atmospheric rivers and other dynamically active regions, with their observations assimilated in real-time into global NWP systems. This helps to improve the initialization of the next forecast. In this research, we use a subset of dropsondes deployed by the National Oceanic and Atmospheric Administration Gulfstream IV-SP aircraft in the 2020, 2021, and 2022 seasons to evaluate the jet stream in the European Centre for Medium-Range Weather Forecasts Integrated Forecasting System. Results show that on average the model has a slow wind bias for winds ≥ 50 ms-1, but that this is largely corrected via data assimilation. The computation of the cross jet-axis potential vorticity (PV) gradient highlights model issues in resolving the sharp PV gradients across the jet in the tropopause region. A case study and composite analysis illustrate this finding. Potential model improvements that could be considered for better modelling of the jet stream will also be discussed.

Tropical Cyclones (TCs) are one of the most significant weather hazards in the tropical regions; with strong wind, storm surge, and extreme precipitation, TCs cause many human deaths and substantial property losses. Therefore, a significant reduction in systematic errors that can hinder accurate predictions of TC activity is an important and urgent task. During the past two decades, many studies have focused on improving the seasonal prediction skill of TC activity such as number, tracks, and intensity through both dynamical models and statistical methods. A new version of the Euro-Mediterranean Center on Climate Change (CMCC) Seasonal Prediction System (SPS3.5), which configures with a relatively high spatial resolution (0.5 degree) and a large number (40) of ensemble members, provides us a great opportunity to follow and possibly anticipate the characteristics of TC seasonal activities. This study aims to assess how good is SPS3.5 in representing the TC spatial and temporal variability compared to observations, including spatial distribution, seasonality, intensity, interannual variability, and their relationship with ENSO. We compare the SPS3.5 hindcasts to observational TC tracks from International Best Track Archive for Climate Stewardship (IBTrACS) over the period 1993 to 2016. To this aim, we use a Geophysical Fluid Dynamics Laboratory (GFDL) Tropical Cyclone tracking algorithm to identify storms and create cyclone trajectories. We find that the new CMCC SPS3.5 system captures well the number of TCs and their spatial distributions, although with some locations deviating from observations. Also, the performance of the model varies significantly from basin to basin. Specifically, the model captures TC activity best over the Pacific, while lower capability is shown over the Atlantic. We also discuss the characteristics of the simulated TCs frequency and its variability in time, and investigate possible root reasons of systematic errors and bias correction methods.

Increasingly, forecast emphasis is moving towards hazard- and impact-based forecasting, aiming to narrow the gap between science and decision-making for early action based on forecast information. For tropical cyclones, this means incorporating forecasts of not only track and intensity, but also the wider wind fields, precipitation and flooding, and another step further is to include risk information such as populations and infrastructure exposed to these hazards (Magnusson et al., 2021; Emerton et al., 2020). Traditionally, the focus of investigation into tropical cyclone predictability and predictive skill has been on track prediction, and has increasingly moved towards intensity and other characteristics as model resolution has increased. Now, as part of the shift towards hazard- and impact-based forecasting, it is imperative to understand the skill in forecasting tropical cyclone hazards and diagnose systematic errors associated with these.

We present a range of recently-developed, hazard-focussed tropical cyclone diagnostics, to better understand the ability of our forecasting systems to predict tropical cyclone winds, structure, rainfall and flooding. It is important to consider such diagnostics alongside existing verification of tropical cyclone tracks and intensity, particularly as forecasts of tropical cyclone hazards rely on accurate forecasts of the track, and, for example, it is known that rainfall amounts are not directly related to the intensity of TCs, but the translation speed and size of the TC alongside topography and geography of the landfall region are key. Therefore, predicting TC rainfall and flooding relies on several factors: track, intensity, size, structure, and interaction with land and the wider atmospheric circulation (Titley et al., 2021). We highlight case studies investigating the full chain of tropical cyclone forecasts, from the track and winds to the rainfall and flooding, using forecasts from ECMWF’s Integrated Forecasting System (IFS) and the Copernicus Emergency Management Service’s Global Flood Awareness System (GloFAS). We demonstrate the importance of evaluating tropical cyclone hazards, diagnosing systematic errors in these forecasts, and the ability of our forecasting system to predict these hazards and provide useful forecasts for decision-making.

References
Emerton, R. et al., (2020). Emergency flood bulletins for Cyclones Idai and Kenneth: A critical evaluation of the use of global flood forecasts for international humanitarian preparedness and response. International Journal of Disaster Risk Reduction, 50, 101811. https://doi.org/10.1016/j.ijdrr.2020.101811

Magnusson, L. et al. (2021): Tropical Cyclone Activities at ECMWF, ECMWF Tech Memo, 888, DOI: 10.21957/zzxxzzygwv

Titley, H. A., Cloke, H. L., Harrigan, S., Pappenberger, F., Prudhomme, C., Robbins, J. C., Stephens, E. M., & Zsoter, E. (2021). Key factors influencing the severity of fluvial flood hazard from tropical cyclones. Journal of Hydrometeorology, 1(aop). https://doi.org/10.1175/JHM-D-20-0250.1

We assess the magnitude of ENSO events in Met Office coupled models. In observations, the strongest warm El Niño events are stronger than the strongest cold La Niña events. However, in common with many climate models, Met Office coupled models show little difference in amplitude between the two phases. We use the same model in our seasonal and decadal prediction systems, and while on seasonal timescales the initialised prediction realistically captures the amplitude of extreme El Niño events, on longer timescales the predictions revert to the control behaviour, that is, there are no very large El Niño events. This may impact on our ability evaluate the risk of extreme regional events. Using a perturbed physics ensemble, this study explores the mechanisms responsible for ENSO amplitude asymmetry. Ocean nonlinear dynamic warming contributes to asymmetry as it acts to warm both La Niña and El Niño events. This is found to be underestimated in the model, with weak subsurface ocean current variability associated with ENSO likely playing a key role. Differences in westerly wind burst activity between El Niño and La Niña are also underestimated.

Modern forecast models simulate many storm-related measures on sub-seasonal timescales, including the location and intensities of atmospheric rivers: long, relatively thin plumes of poleward moisture transport associated with extreme precipitation over continents. Like other storm-related measures, these rivers are dependent on the planetary-scale circulation, which defines the background flow in which storms develop and propagate. The variability of the planetary circulation can be described by circulation regimes. Planetary scales are predictable for a longer time range than small spatial scales. Thus, forecast storminess measures are likely to be more accurate if their regime dependence is well reproduced in models.
This raises the question of the spatial resolution necessary to accurately simulate the circulation dependence of atmospheric rivers and storminess. This research investigates the impact of enhanced resolution on the regime dependence of atmospheric rivers using ensemble hindcasts made with the ECMWF coupled model at three resolutions (100km, 31km, 16km), verified against ERA5 reanalysis. We present the circulation regimes for boreal winter in the Pacific North America region using the simple machine learning algorithm k-means clustering, and the corresponding regime-dependent shifts in atmospheric rivers. We find that the medium 31km resolution produces the best overall correlation with reanalysis in the structure of the Pacific North-American regimes, with lower skill in the high 16km resolution. We also find that the model consistently undercounts atmospheric rivers. The dependence of the results on the criteria used to define atmospheric rivers will be discussed.

Modelers know very well that parameter tuning can drastically control the performance and the systematic errors of climate models. However, parameter tuning is not the only implementation detail that can drastically affect model performance and the representation of various phenomena in models. For example, lower limits (sometimes upper limits) of parameters often control the model performance critically. Not only lower limits but also thresholds of variables that control the enabling or disabling of a specific process sometimes exert large influence on the performance. In addition, whether two schemes can work together or only one scheme of them exclusively works affects significantly the results. The importance of these treatments is often overlooked and not discussed in the literature. However, the impacts of such minor-looking treatments are often even much larger than introducing advanced parameterizations based on theory or observation. We would like to show a number of examples of various minor-looking treatments that can considerably affect model performance and systematic errors. We discuss such minor-looking treatments especially in cloud related processes because clouds are crucially important for controlling radiation budget in GCMs. Many of them are important not only for GCMs but also for global numerical weather prediction models and even for regional scale models.

The Korea Institute of Atmospheric Prediction Systems (KIAPS) developed a new global numerical weather prediction for the medium-range forecast, Korean Integrated Model (KIM) and it began operation in 2020 at the Korea Meteorological Administration. The new project of KIAPS aims to extend the forecast length to 30 days. As the forecast time extends, prediction uncertainty increases. Therefore, it is important to diagnose systematic errors of the KIM and to understand error sources. In this study, we examine the prediction skill and simulated characteristics of East Asian monsoon shown in multi-year seasonal simulations of the KIM with low horizontal resolution by 100 km. In this experiment, 5 ensembles are initialized with a 24-hour time lag from 1st day of May and November from 2001 to 2020, and they are initialized with ERA5. Sea surface temperature and sea ice concentration are updated by observation at every 24-hour for a simulation range around 120 days. We examine the interannual variability of East Asian monsoon indices to investigate how well the KIM reproduces them. And related systematic errors and error source are investigated by evaluation of the KIM long-term simulation characteristics.

Climate models suffer from significant systematic errors (biases) that have proved difficult to reduce. Understanding how biases affect the simulated variability is difficult since biases, just like variability, likely have remote origins. We investigate bias-teleconnections using a dynamical approach. The bias patterns and the associated changes in spatio-temporal variability are studied using a set of century long (1900-2010) simulations with a general circulation model of intermediate complexity forced by a prescribed sea surface temperature (SST). The circulation bias originates from systematic errors in regional SST patterns. Our reference simulation is forced with the monthly SST from the ERA-20C reanalyses. Perturbation experiments are forced with the same SST but with addition of a perturbation in the tropical (Indian, Western Pacific, Central Pacific, Eastern Pacific and Atlantic) or extra-tropical (North Pacific and North Atlantic) oceans. The bias is computed as the time-averaged difference over 1930-2010 period between the perturbation and the reference experiments.
Bias analysis is performed in spectral space of the two main dynamical regimes: a balanced (quasi-geostrophic) regime and an unbalanced regime. The results show that systematic errors in tropical SST produce biases in global atmospheric circulation. In the tropics, biases are found in both balanced and unbalanced components of the circulation whereas the extratropical biases are confined to balanced flows. The tropical or near-field part of the biases in balanced flow shows a quadrupole structure typical for circulation response to diabatic forcing. The extratropical or far-field biases have a pattern of stationary Rossby waves. In contrast, the effect of systematic errors in extratropical SST is largely limited to the extratropical circulation (near field).
The effects of SST biases on wavenumber spectra of spatial and temporal variances show a close correspondence between the wavenumber spectrum of the bias and that of the spatial variance (energy). The exact shape of the spectra depend on the ocean basin with the SST bias.

Systematic errors in climate models can affect climate change simulations. Here we simulate recent climate change with an atmosphere-only model to investigate whether removing climatological mean biases in the large-scale atmospheric circulation can improve climate change simulations. To alleviate model biases, we apply a two-step nudging-based procedure. In the first step we nudge large-scale modes of simulated temperature, divergence, vorticity and surface pressure towards corresponding values from a reanalysis and record the nudging tendencies. In the second step we add smoothed, seasonally varying nudging tendencies from the first step to the otherwise freely running atmospheric model to alleviate model biases. Thereafter, we simulate the atmospheric response to the observed changes in sea surface temperature (SST) and sea ice concentrations (SIC) between 1980s and 2010s using both the original biased model and the model with alleviated large-scale atmospheric circulation biases. Since changes in the atmospheric circulation between these periods are due to both SST/SIC forcing and unforced climate variability, it is not expected that the models reproduce all observed features. Nevertheless, we find that the bias-corrected model, but not the original biased model, reproduced the observed shift of the wintertime stratospheric polar vortex towards Eurasia and a tendency towards more positive North Atlantic Oscillation phase between these periods. The result suggests that these changes likely represent a forced response and that their simulation is sensitive to model biases.

This study takes the first step to bridge the gap between the pressure drag of a shallow cloud ensemble and that of an individual cloud composed of rising thermals. Analysis of a large eddy simulation for BOMEX conditions reveals that the pressure drag for a cloud ensemble is primarily controlled by the dynamical component. The dominance of dynamical pressure drag and its increased magnitude with height are independent of cloud lifetime and are common features of individual clouds except that the total drag of a single cloud over life cycle presents vertical oscillations. These oscillations are associated with successive rising thermals but are further complicated by the evaporation-driven downdrafts outside the cloud. The horizontal vorticity associated with the vortical structure is amplified as the thermals rise to higher altitudes due to continuous baroclinic vorticity generation. This leads to the increased magnitude of local minima of dynamical pressure perturbation with height and consequently to increased dynamical pressure drag.

Tropical disturbances such as the Madden Julian Oscillation (MJO) and equatorial waves (including Kelvin waves, equatorial Rossby waves, mixed Rossby-gravity waves and African easterly waves) play a major role in organising tropical convection, and influence synoptic-scale features and high-impact weather events both within and beyond the tropics. They are also an important source of predictability, and it is therefore essential to understand the ability of numerical weather prediction models to represent these tropical waves.

At the European Centre for Medium-Range Weather Forecasts (ECMWF), we are undertaking evaluation and developing diagnostics to better understand tropical waves in the Integrated Forecasting System (IFS). This work aims to provide insights that will allow diagnosis of systematic errors and uncertainties in the forecasts of tropical waves themselves, alongside prediction of extreme precipitation in the tropics and teleconnections to the midlatitudes. We present a range of recently-developed diagnostics applying wave identification techniques to IFS forecasts at various timescales, producing a new ‘dashboard’ of products that can be used for both real-time monitoring and diagnostics of tropical waves themselves, and evaluation of tropical waves in the IFS and the link to predictability of high-impact weather and teleconnections.

The halo region, defined as the moist buffering region without cloud liquid water, is critical for the interplay between the cloud and the environment and also has non-negligible impact on radiation, but yet lacks enough attention. Previous studies found large uncertainty of halo size and suggested the dependency of halo size on cloud size but used coarser resolution simulations and limited sampling of observations. This study is designed to systematically investigate the halo region of shallow cumulus clouds using large eddy simulations.

The auto-correlation analyses suggest converged size of 200-300 m for moist patches outside clouds when model resolution is below 50 m but may have an overestimation. An "onion algorithm" is developed to examine the composite structure outside individual cloud. It is found that the distribution of relative humidity in the halo region does not depend on cloud size, but on the real distance away from the cloud boundary, regardless of model resolution. The relative humidity decays outward more quickly with finer horizontal resolution, which might result from the dependency of cloud spectrum on model resolution. There is no tendency for the relative humidity distribution within the halo region to converge when the model resolution is above 10 m. This poses some doubt on the fidelity of large eddy models to realistically capture the details of structures within halo region around natural clouds, at least across the resolutions (10-100 m) investigated in this study. A robust feature that halo size near cloud base is larger than that within cloud layer can be observed in all simulations. Lagrangian trajectory analyses further indicates that the formation of the halo region at different vertical levels may be results from different physical processes with different characteristic length scales. Possible length scales responsible for cloud halo size are discussed.

Ensemble forecasts aim to represent all sources of uncertainties including model uncertainties. Over more than two decades ECMWF has used a stochastic representation of model uncertainties in the ensembles known as SPPT: Stochastically Perturbed Parametrization Tendencies scheme (originally just stochastic physics). It perturbs the total physics tendencies. The scheme has been very successful in increasing the probabilistic skill. However, it also has limitations in its current form in the IFS (ECMWF's Integrated Forecast System) as it does not respect the physical consistency present in the real atmosphere and the unperturbed IFS.

It is not obvious how to overcome these limitations of SPPT and this has motivated work to develop a new scheme that aims to represent model uncertainties closer to the sources of the errors and maintains the physical consistency, e.g. is locally conserving energy and moisture. This scheme known as SPP (Stochastically Perturbed Parametrizations) has recently been revised and achieves similar probabilistic skill as SPPT in the medium-range.

The talk/poster will discuss recent evaluation of the SPP scheme from high-resolution medium-range ensembles to seasonal forecasts. Its impact on ensemble spread, systematic errors and probabilistic skill will be discussed.

Identifying and analysing Tropical Cyclones (TCs) in Global Climate Models (GCMs) is an important but challenging task. In this study, TC activity over the Bay of Bengal (BoB) using six multi-ensemble GCMs (both the atmosphere-only and coupled versions) in the PRIMAVERA project following the same protocol is examined in the present (1950-2014) and future (2015-2050) climates. The TCs have been identified and tracked in the model data using the TRACK algorithm. The International Best Track Archive for Climate Stewardship (IBTrACS) data and ERA5 reanalysis data are used for comparison with the model TCs. We use the Genesis Potential Index (GPI) to study the large-scale environmental conditions associated with the TC frequency in high resolution (~25 km) models. Although the models struggle to reproduce the observed frequency and intensity of TCs, all models can capture the bimodal characteristics of the seasonal cycle of cyclones over the BoB (with fewer TCs during the pre-monsoon [April-May] than the post-monsoon [October-November] season). We find that GPI is able to capture the seasonal variation of the TC frequency over the Bay of Bengal in both the observations and models. When comparing the atmosphere-only and coupled versions of the models, a reduction of 0.5°C in the Sea Surface Temperature (SST) and a lowering of TC frequency occur in almost all the coupled models compared to their atmosphere-only counterparts. We investigate the proportion of strong cyclones (after calibrating the maximum sustained windspeeds in the models with IBTrACS) in the pre-monsoon and post-monsoon seasons. We also evaluate the frequency and intensity changes over the BoB in future-climate runs.

Using machine learning to develop new parameterizations on data from high-resolution simulations is a growing research area. These studies have mostly used feedforward neural networks or random forests where the inputs and outputs correspond to atmospheric profiles of different variables. However, for physical processes which have vertical dependencies, should we not try to use machine learning models which can structurally incorporate such dependencies?

Such models are found in recurrent neural networks. Using recurrent neural networks, which process one model level at a time, also greatly reduces the dimensionality of the problem, which is important for generalization. To investigate this issue, some examples are presented where feedforward and recurrent neural networks are trained on sub-grid physical processes and compared.

National Centre for Medium Range Weather Forecasting (NCMRWF) runs operationally a lagged global ensemble prediction system (NEPS-G) of 23 ensemble members. The control and 11 perturbed ensemble members run from the initial conditions of 00 UTC of present day and other 11 perturbed members (i.e., lagged ensemble members) run from12 UTC of the previous day. The present study focuses on evaluating and comparing the skill of operational 23-member ensemble (E23) and 11-member ensemble starting from 00 UTC (E00_11) in two different seasons, namely April-May-June 2019 (hereafter, AMJ) and December 2019 - January 2020 (hereafter, DJ). We have also compared the skill of the configurations, E23 and E00_11 in these two seasons separately.
Verification of temperature at 850 hPa and geopotential height at 500 hPa pressure levels (T850 and Z500) are done over the northern hemisphere (NH). The verification of the zonal and meridional winds at 850 hPa and 200 hPa pressure levels (U850, V850, U200 and V200) are done over a tropical region that includes India. The metrics used for validation are ensemble-spread and root-mean-square error (RMSE) of the ensemble mean relationship, brier score (BS), brier skill score (BSS), reliability, outlier statistics, ranked probability score (RPS) and relative operating characteristics (ROC) skill score.
The results show that in both seasons, AMJ and DJ, the RMSE-spread relationship is better for E23 than for E00_11. For T850 and Z500 over NH, both E23 and E00_11 have higher RMSE and higher ensemble spread in DJ season as compared to AMJ season. In case of wind over Indian region, both RMSE and ensemble spread exhibit lower values in DJ season as compared to AMJ for both ensemble configurations. For both the seasons, BSS of E23 is better than that of E00_11. Again, BSS values of E23 are lower in DJ than AMJ season for T850 and Z500. ROC skill score also shows similar characteristics as BSS.
For both the seasons, we have also compared the forecast skill of E23 with that of the ensemble formed by 22 members, all running from 00 UTC i.e., E00_22 for the same set of variables as well as for precipitation with forecast data of both seasons i.e., June 2019 and January 2020.

The Model for Prediction Across Scales-Atmosphere (MPAS-A) is an open-source community model developed and supported by the National Center for Atmospheric Research (NCAR). As evidenced by the extensive suite of choices for the W\eather Research and Forecasting model (WRF-ARW), also developed at NCAR, an area of great interest from the community is exploring ideas for physical parameterizations. We present a design for and early experiments with a testing rig aimed at first-cut evaluations of physics schemes through cycling data assimilation with MPAS-A and the Joint Effort for Data assimilation Integration (JEDI). Because human resources for the evaluation task are very limited, a central goal is an evaluation that flags significant changes in the skill and systematic errors of the forecast system with minimal human intervention.

Annual tropical cyclone verification carried out for WGNE by the JMA has shown a persistent weak bias in the Canadian global NWP system. This bias continues to exist despite a major update to model physics in 2019 that dramatically reduced the number of false alarms generated by the model. As global models move towards higher resolution, the expectation is that they will become increasingly capable of providing useful tropical cyclone intensity guidance.

Recent tests in an idealized configuration suggests that solutions from the Canadian model (GEM) do not converge towards expected tropical cyclone intensity as the grid spacing is reduced. The storm remains well below its potential intensity, unlike the results of an equivalent simulation run with the WRF model. This suggests that changes to GEM will be required before the model is capable of accurately predicting tropical cyclone intensity at any grid spacing.

We discuss here a series of tests that attempt to identify the source of the tropical cyclone weak bias. Building on results from the DCMIP-2016 project in which several models suffered from excessively weak storms, both the GEM dynamical core and the physical parameterizations are investigated as possible sources. Although the focus of this investigation is on tropical cyclone intensity, we believe that increasing the accuracy of the model in the presence of strong gradients may also yield improvements in the prediction of other forms of high impact weather.

Absorbed solar irradiance is the primary energy source to the Earth system. Most incoming solar irradiance is absorbed at the surface. Here, particularly deviations in the snow reflectance can cause errors. We study the impact on the snow itself and the atmospheric surface fields. We use the generalized surface scheme SURFEX, in which different variants of the Crocus snow albedo scheme can be chosen. We test variables that determine the snow reflectance and show that the spectral coupling to the atmospheric radiation of is of particular importance. In particular, we show that the current default parametrization leads to a positive bias in the reflectance. Correcting this not only leads to a higher surface temperatures but also causes positive feedback effects that enhances the snow melt further.

Due to its maturity, artificial intelligence is already considered an intrinsic component of the development chain of numerous research products. This becomes even more evident when it comes to discovering trends in large datasets, such as climate forecast data, whose time series are non-stationary. However, due to the high degree of uncertainties that are intrinsic to these forecasts, it is important to know their degree of reliability at the time of their issue. The goal of this work is to apply a decision tree-based method to forecast the seasonal forecast skill map produced by the Brazilian Earth System Model - BESM. 30 years of twelve months of seasonal climate hindcasts for global precipitation and air temperature are used to automatically categorize the forthcoming seasonal predictions skill into three classe maps: low, average, and high at every model grid point. Based on the hindcast training period, the method predicts the likelihood of a prediction falling into one of the three classes of prediction skill, based uniquely on large scale anomalies in the forecast's atmospheric and oceanic initial conditions. The goal of the results aim to provide insights about how reliable the climate forecasts are in order to provide means to guide decision makers to enable efforts to be redirected to the most likely forecast events and, consequently, mitigate human lives and economical losses.

In the tropics, clouds and the large-scale circulation are intimately coupled. The large-scale circulation controls the location of clouds, which in turn feedback on the circulation though latent and radiative heating, and transport of air parcels. This complex coupling remains relatively poorly understood. Moreover, resolving this coupling requires higher horizontal resolutions than possible in climate models under current computational constraints. This contributes to the large uncertainty in the response of clouds to climate change.
In this study we introduce a simple new method for describing the relationship between clouds and circulation. We apply this method to a combination of reanalysis and satellite data to generate an estimate of the true relationships between clouds and circulation. We examine how these relationships change with season, ENSO phase and SST. We then evaluate the ability of climate models to represent these relationships.

Both atmospheric and land surface models affect the accuracy of predictions near surface. Therefore, in order to reduce biases in the predictions near surface, it is necessary to separate causes of the biases into origins from atmospheric and land surface models. It has been noticed that the JMA operational global model (GSM) has a cold bias in the screen level temperature in the daytime even though GSM overestimates surface downward shortwave radiative flux over the land. This suggests there are significant sources of the cold bias in other parts of GSM. Land surface processes can be candidates of the sources. This study investigates error characteristics of the improved Simple Biosphere Model (iSiB, Hirai et al. 2007; Yonehara et al. 2017, 2018, and 2020), the land surface model used in GSM, excluding the effect of coupling with the atmospheric part of GSM.
This study uses the offline version of iSiB to extract biases in iSiB alone. The offline iSiB is driven by the atmospheric forcing dataset of GSWP3 (Global Soil Wetness Project Phase 3), which has much lower errors in surface incident radiative fluxes than the atmospheric part of GSM. In-situ observation data of CEOP (Coordinated Energy and Water Cycle Observation Project) is used to verify the offline iSiB predictions. The verification results show that iSiB driven by appropriate atmospheric forcing underestimates surface upward shortwave radiative flux and overestimates surface upward longwave radiative flux. On the other hand, the offline iSiB predicts screen level temperature with no significant biases at most of the observation sites. This suggests there might be error compensation in surface heat budget of iSiB, therefore further verification of biases in elements related to surface heat budget (e.g., ground surface temperature, ground heat flux, and sensible and latent heat fluxes) will be carried out as the next step.

CanESM5, the Canadian Centre for Climate Modelling and Analysis (CCCma) Earth system model contributing to CMIP6, is among CMIP6 models having unusually high climate sensitivities. In addition, CanESM5 represents key aspects of ENSO variability, such as the amplitude and seasonal variation of equatorial Pacific SST anomalies, less realistically than the older CCCma model version CanCM4 that is currently used operationally for seasonal and decadal prediction. These model attributes may at least partially explain why CanESM5 skill is notably lower than CanCM4 for seasonal prediction of ENSO and decadal prediction of global temperature.

In an effort to improve the performance of CanESM5 for seasonal and decadal prediction, a run-time bias correction procedure has been developed whereby tendency corrections are applied to three dimensional atmospheric temperature, specific humidity and horizontal winds, along with oceanic temperature and salinity. These tendency corrections consist of seasonally varying climatologies of terms that relax these variables toward reanalysis values in separate data-constrained runs. Application of the corrections substantially reduces biases in CanESM5 runs, for example in SST and surface salinity, and improves the simulated seasonal cycles of equatorial Pacific SST and ENSO-related SST variance.

In addition to the bias correction method and its effects on CanESM5 simulations, this presentation will examine impacts on seasonal and decadal prediction skill for default and optimized implementations of the correction method, and for a perturbed parameter version of CanESM5 in which historical warming and ENSO amplitude are more realistic than in the CMIP6 version.

We identify a large bias in the simulated atmosphere-ocean variability over the subpolar North Atlantic in a number of models from HighResMIP, and then explore its causes. In general, it is found that the low-resolution (LR) simulations overestimate sea-surface temperature (SST) anomalies in the subpolar region compared to ERA5 reanalysis, and that this bias is reduced in the high-resolution (HR) simulations. Furthermore, in the HR simulations and ERA5, the SST anomalies project onto an NAO-like sea-level pressure (SLP) pattern, while in the LR simulations the SLP pattern is absent or weaker. It is argued that these discrepancies are related to the simulation of ocean dynamics in this region, rather than the atmospheric forcing. In particular, the variance of the subpolar SSTs is found to be correlated with the variance of the upper-ocean subpolar circulation across individual members of each model ensemble, with the LR simulations having much greater variance in both subpolar SSTs and the upper-ocean subpolar circulation compared to the HR simulations. Finally, it is shown that the discrepancies in the variance of the upper-ocean subpolar circulation are primarily related to the low-frequency (> 7 year timescale) component. This suggests that the biases in subpolar SST anomalies identified in the LR simulations may be related to biases in the delayed adjustment of the subpolar ocean gyre to the atmospheric forcing. Regardless of the mechanism, the results overall suggest that the LR simulations may overestimate the forcing of low-frequency SST variability in the subpolar North Atlantic by the ocean circulation. Since previous studies have demonstrated that initialized climate model simulations can skillfully predict SST anomalies in the subpolar region, our results suggest that this prediction skill may also depend on the model resolution.

Teleconnection patterns associated with the Madden-Julian oscillation (MJO) and El Nino Southern Oscillation (ENSO) impact weather and climate phenomena such as blocking and the Pacific-North American pattern, and therefore accurately simulating these teleconnections is of importance for seasonal and subseasonal forecasts.
Systematic biases in boreal mid-winter ENSO and MJO teleconnections are found in 8 subseasonal to seasonal (S2S) forecast models over the Pacific-North America (PNA) region. All models simulate an anomalous 500-hPa geopotential height response that is too weak. This overly weak response is associated with overly weak subtropical upper level convergence and a too-weak Rossby wave source in most models, and in many models also with a biased subtropical Pacific jet which affects the propagation of Rossby waves. In addition to this overly weak response, all models also simulate ENSO teleconnections that reach too far poleward towards Alaska and Northeastern Russia.

Surface-atmosphere interactions play a critical role in the development of the boundary layer and strongly control the diurnal evolution of near-surface temperature and humidity, important metrics in forecast verification. However, our knowledge of the dominant processes that control these interactions is still limited and further complicated by significant heterogeneity in the land surface. In semi-arid/arid regions water availability in the landscape can be altered by human management processes including irrigation. Irrigation processes are typically not represented in general circulation models (GCMs) and can give rise to systematic warm/dry biases.

The JULES irrigation code has been coupled to the Met Office Unified Model (UM). Convection-permitting regional simulations of the UM have been run for a study region in north-east Spain, incorporating the Ebro Basin, which has a significant irrigation presence. Simulations are evaluated using observations from the international LIAISE (Land surface Interactions with the Atmosphere over the Iberian Semi-arid Environment) project. We show the impact of representing irrigation processes on the development of the morning transition, including improvements to the heat and moisture budget, and demonstrate implications for land-atmosphere coupling using metrics from the GEWEX Local Land-Atmosphere Coupling (LoCo) project. A limitation in the use of diagnostic analysis and the availability of observations for advancing understanding of process level atmosphere–land interactions was a key outcome of the Fifth Workshop on Systematic Errors in Weather and Climate Models (WSE).

To advance our processes level understanding single-column model simulations are used to isolate feedbacks between land surface schemes and boundary layer schemes. Our methodology for assessing the impact of land/atmosphere feedbacks begins by assessing the individual components constrained by observational data and then identify changes due to coupling between schemes. This international modelling intercomparison will initially examine irrigated and natural semi-arid domains but will go on to explore how the regional spatial heterogeneity impacts on boundary layer development and evolution.

Precipitation and temperature biases from a set or Regional Climate Models from the CORDEX initiative have been analysed with the aim of assessing the extent to which the biases may impact on the climate change signal. The analysis has been performed for the South American CORDEX domain for the austral warm season (December-January-February). A large warm bias was found over central Argentina (CARG) for most of the models. Results indicate that the possible origin of this bias is an overestimation of the incoming shortwave radiation, in agreement with an underestimation of the relative humidity at 850 hPa, variable that could be used to diagnose cloudiness. Regarding precipitation, the largest biases were found over north east of Brazil (NEB), where most of the models overestimate the precipitation, leading to wet biases over that region. This wet bias agrees with models underestimation of both the moisture flux convergence and the relative humidity at lower levels of the atmosphere. This outcome suggests that the generation of more clouds in the models, may drive the wet bias over NEB. The climate change signal could be affected by these biases, considering that these biases may not be stationary. For both CARG and NEB regions, models with higher warm biases project higher warming levels. In addition, it was found that these relationships are statistically significant with a confidence level of 95%, pointing out that biases are linearly linked with the climate change signal. For precipitation, the relationship between the biases and the projected precipitation changes are only statistically significant for the NEB region, where models with larger wet biases present the highest positive precipitation changes. As in the case of biases, the analysis of the temperature and precipitation projections over CARG and NEB suggest that they are affected by clouds. This study point out that the analysis of bias behaviour could help in a better interpretation of the climate change signal.

We evaluated a short-range (0-75h) ensemble prediction system (EPS) running operationally in the NCMRWF, India at a convective scale (~4km) with 11 ensemble members and a control run. Verification has been conducted for the summer monsoon months of August and September 2019 over a domain centring over India and its neighbourhood. The performance of regional EPS is evaluated with respect to the global model running operationally at 12 km resolution. Standard verification matrices have been used to assess the EPS performance at a regional scale with respect to the observations/analysis and its global counterpart.

Comparing the output of general circulation models to observations is essential for assessing and improving the quality of models. While numerical weather prediction models are routinely assessed against a large array of observations, comparing climate models and observations usually requires long time series to build robust statistics.
Here, we show that by nudging the large-scale atmospheric circulation in coupled climate models, model output can be compared to local observations for individual days. Radiosondes, surface flux and cloud remote sensing observations from the MOSAiC expedition serve as reference observations. Our case study reveals model shortcomings in the representation of mixed-phase clouds and snow thermodynamics and confirms long-standing issues in the representation of stable boundary layers.
The comparison reveals that AWI-CM1/ECHAM and AWI-CM3/IFS miss the diurnal cycle of surface temperature in spring, likely because the snow pack on ice is assumed to have a uniform temperature in both models. In AWI-CM1/ECHAM, an unrealistic relationship between near-surface temperature gradients and surface fluxes can be attributed to a mismatch between the skin temperature representations within the atmospheric and sea ice models. During a cold and dry period with pervasive thin mixed-phase clouds over the MOSAiC site, AWI-CM1/ECHAM only produces partial cloud cover and overestimates downwelling shortwave radiation at the surface. AWI-CM3/IFS produces a closed cloud cover but misses cloud liquid water.
Our results show that nudging the large-scale circulation to the observed state allows to meaningfully compare climate model output even to short-term observational campaigns observing individual weather events.

CoMorph is a new mass-flux cumulus parameterization scheme designed for use within the UK Met Office Unified Model, and its successor model LFRic. It has various novel features including: its closure formulation, an assumed-pdf calculation of detrainment, the possibility for initation from an arbitrary source layer (or layers), and a careful numerical implementation that produces a smooth operation across timesteps. In global model tests, it has shown the ability to couple well to the large-scale circulation, improving the development of emergent features such as the MJO.

In this presentation we demonstrate the behaviour of its closure. Rather than considering some vertically-integrated summary measure of instability (e.g. CAPE or one of its variants), CoMorph considers the instability level-by-level. An initiating mass source is defined at each height from the local vertical instability (including contributions from dry-static instability in clear air and from moist-static instability within large-scale cloud). In a series of idealized tests (RCE, RCE with locally perturbed forcings, DGW couplings...) we compare CoMorph to equivalent simulations in which the CoMorph closure has been overwritten by a simple CAPE closure. This framework allows us to understand the behaviour of the CoMorph closure and to clearly and cleanly elucidate its advantages.

Based on the principle “learn from past errors to correct current forecasts”, statistical postprocessing consists in optimizing forecasts generated by numerical weather prediction (NWP) models. In this context, machine learning (ML) offers state-of-the-art tools for training statistical models and making predictions based on large datasets. In our application, ML-based solutions are developed to predict forecast errors of 2m temperature, and 10m wind speed of the ECMWF’s operational medium-range high-resolution forecasts produced with the Integrated Forecasting System (IFS). Statistical models of systematic and random errors are derived sequentially where the random error is defined as the residual error after bias correction. In terms of input, a variety of model variables and space/time indicators are used as predictors. In terms of output, bias correction and forecast uncertainty prediction are made available at any point-locations around the world. An ongoing topic of research is how ML interpretability tools can be leveraged in this context to help diagnose systematic and residual NWP forecast errors.

As part of a number of recent projects (including APPLICATE, INTERACT, YOPPsiteMIP, USURF) there have been developments in the use of supersite data both for evaluating of operational forecasts and to aid model development at ECMWF. In contrast to the synoptic network, which is relatively densely distributed geographically, but measures relatively few parameters, the supersites are a sparse network of observatories where detailed observations of many parameters in the atmosphere, snow, soil column are measured allowing physical processes to be investigated in detail. In this presentation we will describe the current practices in terms of tools for routine evaluation as well as presenting a couple of success stories, where diagnostics using supersite data have helped us understand the causes of forecast error and informed the model development process.

Land-atmosphere interactions play an important role in climate variability, however the detailed description on the evolution of these processes still requires further improvements from numerical weather prediction models. In the context of biogenic fluxes modelling, recent progress towards vegetation classification and land cover use in the Integrated Forecasting System (IFS), the global model from the European Centre for Medium-range Weather Forecasts (ECMWF), has been performed in the framework of the Copernicus Prototype System for CO2 Monitoring - CoCO2 project. In particular, this study focuses the developments made towards the IFS land surface model (ECLand, Boussetta et al. 2021) through an evaluation of the revised land cover and Leaf Area Index (LAI), focusing on the impact that these changes have over near-surface temperature forecasts. IFS online simulations are evaluated against ECLand default configuration and Global Historical Climatology Network (GHCN) observations for one year (2019). ECLand highest scores are found during spring in the north hemisphere, while systematic errors are generally found in the tropics, being more significant during the summer period (Figure 1). It was initially concluded that there is a general increase of temperature in the model due to changes in roughness lengths and vegetation cover, leading to a positive (negative) impact where there is pre-existing cold bias (warm bias). Further developments towards the revised land cover and LAI are currently in progress.

Given a set of imperfect weather or climate models, predictions and projections can be improved by combining the models dynamically into a so-called 'supermodel'. Supermodeling is a revolutionary step forward in the multi-model ensemble approach. Instead of combining model data after the simulations are completed, in a supermodel models exchange information during the simulation. Individual model errors can then be reduced at an early stage before they propagate into larger scales and affect other regions and variables. Effectively, a supermodel is a new dynamical system, in which the individual models synchronise, such that the strengths of the different models are optimally combined. The supermodel approach has the potential to rapidly improve current state-of-the-art weather forecasts and climate projections.
In this presentation we introduce supermodeling and demonstrate its potential in examples of various complexity. We discuss the current work to successfully apply supermodeling in the context of state-of-the-art models. We show the results of combining three different earth system models (NorESM, MPIESM and CESM) in the ocean component, by exchanging Sea Surface Temperature (SST) on a monthly timescale. The trained supermodel shows improvements in for example the climatology of tropical rainfall.

The Madden-Julian Oscillation (MJO) is the dominate mode of intraseasonal (30-90 days) variability in the tropics and has been shown to be an important factor for a wide range of weather and climate phenomena. The MJO is a key source of predictability in the subseasonal-to-seasonal (S2S) timescale (less than 60 days), however biases in MJO structure and behavior persist in coupled global models. In this study, we examine the MJO in the Navy Earth System Prediction Capability (Navy ESPC) forecasts performed for the Subseasonal eXperiment (SubX) using process-based diagnostics and a moisture budget analysis that uses wavenumber-frequency filtering to isolate the MJO.

The MJO in the Navy ESPC model is on average too strong and propagates too quickly. There is a strong bias that appears in the MJO’s amplitude at all forecast lead times. In the Navy ESPC, the MJO’s phase speed accelerates east of the Maritime Continent, which can be linked to an acceleration of moisture anomalies and anomalous moisture tendency in the model. We project the terms in the intraseasonal column-integrated moisture budget onto the maintenance and propagation of moisture anomalies to identify physical mechanisms that lead to the biases in the Navy ESPC MJO. Vertical moisture advection in the Navy ESPC is too strong and deep, driven by a more bottom-heavy vertical motion profile and steeper lower tropospheric vertical moisture gradient, which leads to stronger precipitation anomalies. In addition, the Navy ESPC has a faster convective moisture adjustment timescale than observed, suggesting a stronger precipitation response in the Navy ESPC for the same moisture anomaly. The too fast propagation of the MJO east of the Maritime Continent is present in both seasons, however the causes of this bias are different in each season. During the boreal winter, the too fast propagation of the MJO is driven by excess evaporation in the Western Pacific cause by stronger easterlies. During the boreal summer, the acceleration of the MJO is driven by the increase in the horizontal advection due to the mean state moisture bias.

The impact of tropical sea surface temperature (SST) biases on the deterministic skill of the Unified Forecast System (UFS) coupled model Prototype 5 is evaluated during weeks 1- 4 of the forecast. The evaluation is limited to the Contiguous U. S. (CONUS) and two seasons: boreal summer (June through September) and winter (December through March).

The tropical Pacific SST in the UFS model is warmer than in observations and bias patterns show seasonal dependence especially in the central and western Pacific. During boreal summer, the bias is located north of the equator whereas in winter, the bias is located in the Southern Hemisphere. A warm bias is also found in the tropical Atlantic. A regression analysis indicates a significant relationship between these SST biases and the biases in the surface temperature and precipitation along with mid-troposphere large-scale circulation and North Pacific storm track activity. The influence of tropical Pacific bias is through Rossby waves propagating from the tropics into the extra-tropics. In boreal summer, the Atlantic SST bias partly affects the pressure center of the North Atlantic Subtropical High (NASH), which in the forecast is weaker than in reanalysis. The weaker NASH favors an enhanced moisture transport from the Gulf of Mexico, leading to increased precipitation over the southeast US. During boreal summer, the NASH pressure center is also weaker and in addition, its position is displaced to the northeast, thus further affecting summer precipitation biases over the CONUS.

The SST biases affect the biases in other fields from week 1 of the forecast and the impact becomes stronger as the lead time increases to week 4. The impact of SST biases on the biases in other fields show a qualitative relationship to the patterns of forecast errors of the fields.

Early- and late-winter ENSO teleconnections to the Euro-Atlantic region in state-of-the-art seasonal forecasting systems

Franco Molteni and Anca Brookshaw
European Centre for Medium-range Weather Forecasts, Reading, U.K.

A number of recent studies have highlighted the differences in the northern extratropical response to ENSO during the early and late part of the boreal cold season, particularly over the North Atlantic/European (NAE) region. Diagnostic analyses of multi-decadal GCM simulations performed as a part of CMIP5 and CMIP6 projects have shown that early winter tropical teleconnections are usually simulated with lower fidelity than their late-winter equivalents. Although some results from individual seasonal forecasting systems have been published on this topic, it is still unclear to what extent the problems detected in multi-decadal simulations also affect initialised seasonal forecasts from state-of-the art models.

In this study, we diagnose ENSO teleconnections from the re-forecast ensembles of nine models contributing (during winter 2021/22) to the multi-model seasonal forecasting system of the Copernicus Climate Change Service (C3S). The re-forecasts cover winters from 1993/94 to 2016/17, and are archived in the C3S Climate Data Store. Regression and composite patterns of 500-hPa height are computed separately for El Niño and La Niña winters, based on 2-month averages in November-December (ND) and January-February (JF). Model results are compared with the corresponding patterns derived from the ERA5 re-analysis. Signal-to-noise ratios are computed from time series of projections of individual winter anomalies onto the ENSO regression patterns.

The results of this study indicate that initialised seasonal forecasts exhibit similar deficiencies to those already diagnosed in multi-decadal simulations, with a significant underestimation of the amplitude of early-winter teleconnections between ENSO and the NAE circulation, and of the signal-to-noise ratio in the early-winter response to El Niño. Further diagnostics have highlighted the impact of mis-representing the constructive interference of teleconnections from the Indian and Pacific Oceans in the early-winter ENSO response over the North Atlantic.

Models with coarse horizontal resolution, such as those used for seasonal prediction, cannot explicitly simulate coastal processes such as sea-breezes. This leads to systematic model errors in coastal convection and rainfall, particularly in island regions with complex topography.

We address this by developing a new parameterisation of sea-breeze driven coastal convection. Our parameterisation consists of two parts: i) identifying coastal points affected by sea-breezes, using large scale prognostic fields to calculate the thermal heating contrast between land and ocean, and ii) modifying the convection scheme to enhance or suppress deep convection in the presence or absence of the sea-breeze.

The parameterisation has initially been tested in single column model (SCM) runs, using prescribed initial conditions from Tropical Warm Pool International Cloud Experiment (TWP-ICE) during a period of sea-breeze convection. We link the calculation of the sea-breeze parameter to the convection scheme by increasing initial parcel buoyancy (increasing parcel temperature while maintaining its relative humidity) where the sea-breeze parameter is positive and decreasing initial parcel buoyancy where the parameter is negative. Our results show a corresponding increase and decrease of convective rainfall for positive and negative sea-breeze parameter values, respectively.

Current work focuses on implementing the parameterisation in global model simulations. Initial experiments indicate i) that the parameterisation works as intended and ii) that it affects the diurnal cycle of precipitation near coasts, with feedback then affecting the large-scale circulation. Rainfall is enhanced during the day in wet coastal regions, while the parameterisation clearly distinguishes dry sea-breezes and does not produce spurious rainfall. Large-scale synoptic rainfall events suppress the sea-breeze evolution in the model as they do in the real world.

Future plans involve tuning the parameterisation, performing long atmosphere-only and coupled model simulations to test climatological effects, and then testing the parameterisation in forecast mode, focussing on sub-seasonal timescales.

We develop a statistical method to assess CMIP6 simulations of large-scale surface temperature change during the historical period (1850-2014), considering all timescales, allowing for the different unforced variability of each model and the observations, observational uncertainty, and applicable to historical ensembles of any size. The generality of this method, and the fact that it incorporates information about the unforced variability, makes it a useful model assessment tool, and more objective than visual inspection of historical plumes. We apply this method to the historical simulations of the CMIP6 multi-model ensemble. We use three indices which measure different aspects of large-scale surface-air temperature change: global-mean, hemispheric gradient, and a recently-developed index that captures the sea-surface temperature (SST) pattern in the tropics (SST#; Fueglistaler and Silvers, 2021). We use the following observations: HadCRUT5 for the first two indices, and AMIPII and ERSSTv5 for SST#. In each case, we test the hypothesis that the model's forced response is compatible with the observations, accounting for unforced variability in both models and observations as well as measurement uncertainty. This hypothesis is accepted more often for the hemispheric gradient than for the global mean, for which half of the models fail the test. The tropical SST pattern is poorly simulated in all models. Given that the tropical SST pattern can strongly modulate the short-term feedback parameter (i.e. the relationship between energy imbalance and global-mean surface temperature anomalies on annual to decadal time scales), with potential implications for the global-mean temperature evolution in decadal time scales, this should be a focus area for future improvements.

The presence of a double Intertropical Convergence Zone (ITCZ) in the tropical Pacific is a persistent feature of global coupled ocean-atmosphere models that gives rise to excessive precipitation south of the equator. The ITCZ position is extremely sensitive to changes in the magnitude and distribution of the Sea Surface Temperature (SST) in the tropical band, due to the strong coupling between SST and convective precipitation. The complexity of the air-sea interactions makes it hard to disentangle the different mechanisms at play to identify the main driving processes behind this ubiquitous bias. Here, we use a coupled ocean-atmosphere regional model, the Coupled-Ocean-Atmosphere-Wave-Sediment Transport (COAWST) Modeling System, to investigate the impact that different parametrizations of the oceanic vertical mixing have on the water column dynamic, SST and subsequently the convective precipitation distribution in the eastern tropical Pacific. The model includes an atmospheric component, the Weather Research and Forecast Model (WRF), and an oceanic component, the Regional Ocean Modeling System (ROMS). The same atmospheric setup, with a resolution of 20km, has been forced with observed SSTs and with two ocean parameterizations for an El-Niño neutral year: the 1995. Different temperature gradients and oceanic stratification give rise to a double ITCZ or to a southward shift of the maximum precipitation band. Particularly in winter, a surface warming of a few degrees south of the equator around 5°S affects the distribution of the sea level pressure. The consequent changes in the surface wind pattern impact the usually asymmetric behavior of the trade winds, the south easterlies are no longer able to cross the equator and converge in the ITCZ in the northern hemisphere.

A targeted group was formed between the Met Office, The Bureau of Meteorology, ECMWF and other forecasting centres within the UM Partnership to understand and improve common biases in Indo-Pacific region and associated teleconnections. In particular, the significant cold SST biases in the Eastern Indian Ocean have a significant impact on the prediction skill of the seasonal forecasting systems (in GloSea5, ACCESS-S and ECMWF model) for tropical rainfall, Australasian monsoon system and the representation of associated key teleconnections in climate models. The multi-disciplinary group aims to consolidate current diagnoses of model errors and develop a new set of diagnostics and metrics to understand model errors using a seamless modelling framework and measure the impact of future improvements, as well as to recommend potential model/DA developments and observation campaigns to mitigate these errors. The current strands of activities explore different hypotheses for the sources of errors in the region: the role of different initial conditions and coupled data assimilation, the bathymetry and potential role of the Indonesian throughflow, the sensitivity to various physics changes in the atmosphere and ocean models and the evolution of key biases and the air-sea interaction processes involved at different timescales using coupled NWP, sub-seasonal to climate modelling frameworks. Key results from the initial investigations coordinated across modelling centres will be presented.

Previous work has shown large cold and dry biases in seasonal forecasts over the equatorial Atlantic. In addition, the correlation between tropical Atlantic rainfall and the North Atlantic Oscillation in seasonal hindcasts is reversed compared to the observations. This work investigates these biases in the Met Office Seasonal Forecast System (GloSea), and uses a novel ensemble based method to estimate the impact they have on forecasts of the North Atlantic.
Winter hindcasts covering the period 1993 – 2016 are analysed for biases in tropical Atlantic rainfall rate. The internal variability in ensemble members is then used to explore the impact of the dry bias in equatorial Atlantic rainfall rate on the North Atlantic. This is done by selecting model members which most closely resemble the bias and model members which most closely resemble the observations and using the differences as a proxy for the model error.
We explore the impact of the equatorial Atlantic bias on Rossby wave sources in the tropical Atlantic, and extratropical circulation.
We find that the tropical Atlantic bias impacts Atlantic Rossby wave sources over the northern tropical Atlantic, a common source of Rossby waves propagating over the North Atlantic. A clear Rossby wave pattern originating from this area appears in the analysis and projects onto the NAO. We argue that this can explain a significant amount of the mean bias over the North Atlantic.

We ask what overlap may exist in physical and statistical approaches to systematic error. A motivation is given for completing a measurement model, or regression model, by accommodating partially overlapping notions of equation error and representation error, where both are taken as signal terms. Implications are that error cross-correlation has the interpretation of nonlinear association, error that is uncorrelated is a lack of association, and all three (linear, nonlinear, and lack of association) are needed to describe the signal of interest. An extended sampling model is shown to offer multiple solutions, with two being comparable to ordinary and reverse linear regression, but offering better bounds on systematic error or bias. A few practical applications are highlighted and a framework for exploring model solutions under controlled settings is given at https://github.com/JuliaAtmosOceanHydro/MeasurementModelDemos

Understanding systematic errors in climate simulations is crucial for both model development and for providing trustworthy information for climate projections. One way to aid this understanding is to use perturbed parameter ensembles (PPEs), which comprise variants of a single model, each using a unique set of values for uncertain model parameters. PPEs can readily expose systematic errors in a single model structure as those which cannot be removed by any combination of model parameter values. Tracking these errors across updates to the model configuration can provide useful insights into improvements and degradations in model errors. This can then feed-back into the model development process and allow for more robust assessments of the plausibility of climate projection information.

Here, we present a study of systematic errors across several PPEs based on the most recent atmosphere-only configurations of the Met Office Hadley Centre climate model, HadGEM3. We highlight key changes in these errors across the configurations, and describe the methods used to analyse these in the shared parameter spaces of the PPEs, both for large-scale means and spatially. Further, we consider the potential impact of these changes on climate projections, including an evaluation of radiative feedbacks. We also look at how systematic errors can lead to biases in the selection of plausible variants for climate projections and suggest strategies to mitigate these.

L.Cavaleri, L.Bertotti, N.Wedi, J.Edwards, J.Bidlot

The problem – The progressive increase of models spatial resolution has led to a correspondingly increase of the quality of surface wind fields, now at a very satisfactory level. This is not the case when, also on the ocean space, wind blows from a coast to offshore. Long term comparison with scatterometer data shows that in this case the model wind speeds are substantially underestimated (up to -20%) in the first 200 km run on the sea (fetch). Present also at the ECMWF Tco1279 level, the results strongly depend on (worsen with decreasing) resolution. This has drastic consequences in coastal winds and to a greater extent in the inner seas. Even now, using the Tco1279 winds in the Mediterranean, in particular the Adriatic, Sea a 16% correction is required to get the measured values and, on a parallel course, the correct wave model results.

Approach – Joining ECMWF and UKMO data, we focused our attention on the Mediterranean and Southern North seas for the October – December 2018 period (also in connection with the disastrous Vaia/Adrian storm). Model surface wind data for all the available resolutions have been used and compared with scatterometer data, taking into account the “fetch”, i.e. the distance run by wind on the sea after leaving the coast. Also the previously crossed orography has been considered.

Results – When blowing to offshore, upon leaving the coast the model surface winds are strongly underestimated. The underestimate depends on the model (stronger for ECMWF, milder or null for UKMO), worsening with the more coarse resolutions. Figure 1a shows the model results in the Mediterranean Sea where the orographic effects are more pronounced. Note the differences between the ECMWF and UKMO results, the former ones approaching unity (model = scatterometer) after 100 or 200 km, depending on resolution, the latter substantially exceeding the measured quantities. Neutral winds have been used, so no ensemble data for ECMWF. It is interesting (panel 1b) that for each model the results for different resolutions tend to converge once scaled with resolution. This strongly suggests that part (at least) of the problem is connected to the number of grid steps required to reach the correct asymptotic value.

a b

Figure 1 – a) Ratio between model and scatterometer wind speeds as a function of “fetch”, i.e. the length run by wind on the sea after leaving the coast. ECMWF and UKMO models, with different resolutions (in km), are shown. b) as panel a), but with fetch evaluated as number of grid steps for each resolution.

The importance of atmospheric composition in Numerical Weather (NWP) and Climate Prediction (CP) has been addressed by many studies in the last few decades. Scientific research in the modelling field has been focused on better understanding atmospheric chemistry's role in the predictive skill of global and regional models, using climatologies or including real-time treatment of aerosols, for example, in operational forecasting systems or research models. Recognizing the role of atmospheric composition as a fundamental component to improve forecast capabilities, WGNE, jointly with the WWRP Subseasonal to Seasonal (S2S) Steering Group and the GAW Scientific Advisory Group (SAG) on Modelling Applications (SAG-APP), has been leading the second phase of the Aerosol project. The project aims to explore the importance of interactive aerosols in short to medium-range and subseasonal predictability.

The WGNE-S2S-GAW Aerosol Project considers the participation of operational meteorological centres as well as research groups from institutions around the world, contributing with their state-of-the-art of integrated chemistry-meteorology numerical modelling. A set of experiments consisting of limited-area runs or sub-domains provided by global models, as well as subseasonal experiments including re-forecasts based on an ensemble approach on the global scale have been provided by modelling groups. The experiments include aerosols' direct (and indirect, optionally) effect and simulations either with no-aerosol loading or with climatological aerosols.

The project is based on the model verification approach to identify the role of aerosols on the predictive skill of short, medium and subseasonal forecasts. The first phase of the project examines meteorological fields and atmospheric circulation patterns, linking them with the physical processes associated with the aerosols. The second phase of the project will address aerosol properties and air-quality verification forecasts. In the present study, we will discuss the changes in near-surface temperature forecast errors of participating models by comparing the experiments considering prognostic and climatological aerosols in both regional and subseasonal timescales.

The Navy Earth System Prediction Capability (Navy ESPC) is a coupled global model consisting of the Navy Global Environmental Model (NAVGEM) atmospheric model coupled to the Global Ocean Forecast System (GOFS), which consists of the Hybrid Coordinate Ocean Model (HYCOM) and the Los Alamos Community Ice Code (CICE). We examine the prediction skill and biases in tropical cyclone (TC) tracks, the large-scale TC environment, and the Madden-Julian Oscillation (MJO) in the operational Navy ESPC 16-member ensemble forecasts initialized weekly during 2017, 2020, and 2021. The prediction skill and biases in the TC environment and MJO are compared to models from the Subseasonal-to-Seasonal (S2S) database including the NCEP CFSv2, ECMWF, and UKMO ensembles.

The Tempest Extremes package is used to identify TCs based upon sea level pressure minima, surface wind maxima, and upper-tropospheric warm cores. Biases in the TC genesis count and number of TC days in Navy ESPC are related to model biases in the genesis potential index (GPI), an empirical index based upon shear, humidity, low-level vorticity, and relative SST anomalies. Navy ESPC prediction skill and biases in the GPI and its component terms are compared to those in the S2S database models. We also examine MJO prediction skill using several different measures, including the Real-time Multivariate MJO (RMM) index and Real-time OLR MJO Index (ROMI) as well as OLR that has been wavenumber-frequency filtered for the MJO. MJO prediction skill and biases are related to the subseasonal prediction skill in the GPI and its component terms. Specifically, we are interested in whether mean state biases or MJO prediction skill and biases are more important to the TC environment prediction skill.

A single-column model (SCM) using the new Met Office generalized mass-flux convection scheme (CoMorph) is modified to implement the damped gravity wave (DGW) approach that enables the interactions between tropical convection and the large-scale tropical circulation. Using the modified SCM, we examine the response of the parameterized convection to idealized forcing perturbations under the influence of an evolving large-scale circulation. The forcing perturbations include either a modification of column’s moisture via imposed drying or moistening, modifications of column’s stability via imposed cooling or warming or a combination of both. We also examine the response of the parameterized convection to changes in the strength of column’s moistening or drying. We explore the sensitivity to the entrainment rate and the closure method within CoMorph. The results from the SCM using CoMorph are compared with those from equivalent cloud-resolving model (CRM) simulations performed using the new and the New Met Office NERC cloud model (MONC).

For the same conditions and model configurations, direct contributions from drying and stabilizing the simulated column is to induce a large-scale descent that shuts off precipitating convection completely, whereas direct contributions from moistening and destabilizing the simulated column is to induce a large-scale ascent and the resulting increase in convective precipitation is about equally divided between the moistening and destabilization effects. For standard closure and default entrainment rate within CoMorph, the response to forcing perturbations is faster, convective response to moistening and destabilization is weaker, and the relationship between mean precipitation rate and column water vapour (CWV) is very different. For the standard closure and lowest entrainment rate within CoMorph, the response to forcing perturbations remains faster, convective response to moistening and destabilization becomes stronger, and the relationship between mean precipitation rate and CWV becomes closer to that obtained in MONC. However, for the lowest entrainment rate within CoMorph, the response to forcing perturbations is slowed down when a CAPE closure is used, and a CAPE closure time scale of 3 hours results in convective response to moistening and destabilization that is very similar to that obtained in MONC.

NOAA is collaborating with the US weather and climate science community to develop the next generation fully coupled earth system modeling capability for both research and operational forecast applications across different temporal and spatial scales. This presentation will first introduce major changes and updates of atmospheric model physics which are targeted for both the global and regional models for short and medium-range weather forecasts and subseasonal to seasonal predictions. Strategies are developed to first test individual physics parameterizations in atmospheric-only forecast experiments in the aforesaid applications and then to further evaluate and improve the parameterizations in the integrated earth system modeling applications to reduce model systematic biases and improve model prediction skills. Significant efforts are made to unify physics parameterizations for all applications to speed up the transition of research to operation (R2O) and to reduce the cost of operational systems maintenance. A few samples will be presented to highlight the success and challenges of introducing new physics parameterizations into the UFS forecast systems.

We introduce and demonstrate the performance of a double moment cloud microphysical scheme (CASIM: Cloud AeroSol Interacting Microphysics) in both midlatitude and tropical settings using the same model configuration. Comparisons are made against a control configuration using the current operational single moment cloud microphysics, and a CASIM configuration that computes cloud droplet number concentration from the aerosol environment. We demonstrate that configuring CASIM as a single moment scheme results in precipitation rate histograms that match the operational single moment microphysics. Results indicate that CASIM performs as well as the single moment microphysics configuration in the midlatitude setting but with the introduction of useful features such as greater extent of light rain around frontal precipitation features and outperforms the single moment cloud microphysics in the tropical setting .

Subseasonal prediction of tropical cyclones (TCs) has many potential applications in energy, emergency response, and defense sectors; however, subseasonal TC prediction remains a challenge due to biases in both large-scale conditions and TCs in coupled global models. Subseasonal forecasts of large-scale environmental conditions, such as vertical wind shear, are generally more skillful than forecasts of smaller scale phenomenon, such as TCs themselves. Thus, model forecasts of environmental parameters can be linked to TC activity and then be used to extend the horizon of useful skill through statistical-dynamical models. The aim of this work is to shed light on the model biases in large-scale environmental conditions that inhibit subseasonal TC prediction.
In this study we evaluate several models from the Subseasonal-to-Seasonal (S2S) database on their ability to capture environmental signals linked to West Pacific TC activity. To isolate the signals in TC activity associated with subseasonal variability we examine events of anomalous accumulated cyclone energy (ACE) relative to each season. These events are used to create composites of the environmental conditions related to TC activity (e.g., shear, humidity, low-level vorticity, upper-level velocity potential) at various lead times for each of the forecast models, which are then validated against the ERA5 reanalysis. To quantify skill, we look at the evolution of the pattern correlation between the model and reanalysis composites over the Indo-Pacific as a function of lead time. We also look at the evolution of model biases of several environmental fields to establish if there is a connection between mean state biases and the ability of the model to capture critical large-scale patterns related to TC activity.

The increasing use of reanalysis of tropical cyclones (TCs) illustrates a great value to risk assessment and to detection of climatological trends. However, the available reanalysis of TCs is short, incomplete and it updates quite slowly. For example, lBTrACS (International Best Track Archive for Climate Stewardship) only includes the complete TC size information starting from 2004 with a time interval of 3 hours, and its reanalysis update takes more than one year after the storm. Taking IBTrACS as a reference, we produced a novel reanalysis named TC_RE. Specifically, we designed a deep learning model to generate hourly reanalysis for the radius of max winds (RMW), average 34 kt wind radii (R34), and intensity in wind speed (Vmax) of global TCs using ERA5 hourly data from 1979 to now. The model shows a test error with 9 kt of Vmax globally, less than the intensity uncertainty of IBTrACS (10 kt), as well as accurate test results of RMW, R34, and Vmax in different basins. The availability of ERA5 combined with the deep learning model enables accurate storm characterisation with only a 5 day delay. The tropical cyclones size (both RMW and R34) are found to be stable between 1979 and 2004 but have increased since then.

In accompaniment with the implementation of Global Ensemble Forecast System Version 12 (GEFSv12), a 31-yr (1989–2019) ensemble reforecast dataset has been generated at the National Centers for Environmental Prediction (NCEP). The reforecast system is based on NCEP’s Global Forecast System, version 15.1, and GEFSv12. The Climate Forecast System (CFS) reanalysis and GEFSv12 reanalysis serve as initial conditions for the Phase 1 (1989–99) and Phase 2 (2000–19) reforecasts, respectively. The perturbations were produced using breeding vectors and ensemble transforms with a rescaling technique for Phase 1 and ensemble Kalman filter 6-h forecasts for Phase 2. The reforecasts were initialized at 0000 (0300) UTC once per day out to 16 days with 5 ensemble members for Phase 1 (Phase 2), except on Wednesdays when the integrations were extended to 35 days with 11 members.

This study presents the evaluations of forecast capability; systematic error characters of surface temperature, precipitation and other variables in terms of spatial distribution and forecast lead-time. Meanwhile, the tropical storm tracks, and MJO forecasts are also presented. The results are also compared with GEFSv10 reforecast or GEFS Subseasonal Experiment (SubX) reforecasts. The impact of the calibrating precipitation and 2-m temperature using the reforecast data is addressed. Currently, the GEFSv12 forecast also serves as a benchmark for the GEFSv13 development. Some comparison between the reforecast and the developing version of the GEFSv13 prototype will be also presented.

Diurnal water vapour cycles are a critical component of the hydrological cycle yet they remain one of the hardest processes to represent in both climate and forecasting models, particularly in the Arctic. Previous work has shown that models tend to under-represent the diurnal cycle, which has consequences for radiative transfer, precipitation, and convection algorithms. Additionally, most studies have been conducted in the tropics and have only considered integrated water vapour (IWV) cycles.

Environment and Climate Change Canada constructed a supersite in Iqaluit, Nunavut (63.75°N, 68.55°W) in 2015 as part of the World Meteorological Organization’s Year of Polar Prediction (YOPP) initiative. In 2018, ECCC installed a novel pre-production Differential Absorption Lidar (DIAL) at the supersite that produces water vapour profiles up to 2 km altitude every 20 minutes. Due to the DIAL’s high measurement frequency, it is well-suited for diurnal cycle analysis. We use the DIAL and a co-located Global Positioning System (GPS) to evaluate the ECMWF Reanalysis 5th Generation (ERA5) model and the ECCC Numerical Weather Prediction (NWP) model GEM – HRDPS (Global Environmental Multiscale- High Resolution Deterministic Prediction System).

We find that both numerical products reproduce the DIAL’s diurnal water vapour cycle in phase in the first 1 km of altitude. ERA5’s amplitudes are significantly smaller than the DIAL amplitudes at all altitudes; GEM – HRDPS amplitudes are consistently larger in the first few hundred meters and smaller above 1 km. Neither ERA5 nor GEM – HRDPS are able to reproduce the 12-hr component of the diurnal cycle in either the height-resolved or IWV cycles, suggesting that the representation of some underlying process in the models is incomplete. In conclusion, the numerical products are able to reproduce the overall behaviour of the diurnal cycle, but certain altitude regions require deeper investigation.

A major goal of the Korea Institute of Atmospheric Prediction Systems (KIAPS) is the development of an integrated model available for prediction on a wide range of temporal and spatial scales. To improve the predictability of the extended range beyond about two weeks, the interaction between components of the earth system should be considered in the operational model. The KIAPS is therefore developing a new coupled system, including ocean, sea-ice, wave, and river-routing models, and this study introduces the current status for coupling atmosphere-ocean in the framework of the Korean Integrated Model (KIM).
For an ocean component in the coupled KIM, the Nucleus for European Modelling of the Ocean (NEMO) was selected in consideration of the performance, scalability, and accessibility, along with the research status of other operational institutions. The NEMO version 4.0 was coupled with the KIM through a KIAPS coupler based on the Model Coupling Toolkit (MCT). In the initial version, an algorithm for the bulk formula to calculate turbulent fluxes of momentum, heat, and moisture was different in the atmosphere and ocean components, which was refined to be consistent between the components in the updated version. In addition, a parameterization for cool skin and warm layers was included to represent the skin temperature that is used for calculating turbulent heat fluxes. Compared to the uncoupled atmospheric version, the preliminary results showed that the atmosphere-ocean coupled KIM has a significant impact on the performance of medium-range forecasts and seasonal simulations. Finally, future plans to alleviate regional biases existing in the current coupled model will be discussed.

The Stratosphere-troposphere Processes and their Role in Climate (SPARC) Quasi-biennial Oscillation Initiative (QBOi) aims to improve the representation of tropical stratosphere variability in general circulation models.

QBOi phase-1 coordinated experiments were carried out by 17 models and revealed common biases in the simulated QBOs. In the lowermost tropical stratosphere (~50 hPa, ~21 km) QBO winds are unrealistically weak in nearly all of the models, with underestimates of around 50% being common, and the QBO is meridionally narrower than in reanalyses. In initialised hindcasts the QBOi models do not maintain realistic westward QBO winds at these altitudes. Resolved wave forcing in the lower stratosphere increases strongly with decreasing vertical grid spacing, but at the highest resolution used (~0.5 km) it is unclear if convergence has been reached. At the horizontal resolutions used by the models (from ~100 to ~350 km) almost all the models require substantial forcing by parametrized non-orographic gravity wave drag to drive the QBO. These parametrizations are not well constrained by observations and their parameters can therefore be tuned to improve a model’s QBO representation, but it is unclear whether the same parameter values would remain valid in a changed climate. Idealised future projections by the QBOi models reveal non-robust QBO period changes (suggesting tuning artefacts or inconsistent feedbacks) but also reveal a robust weakening of QBO wind amplitude across the multi-model ensemble (i.e., apparently independent of the tuning). Similar biases and future changes in QBO amplitude are seen in CMIP5 and CMIP6 models.

New QBO-ENSO experiments show that all QBOi models simulate shorter QBO periods during El Nino than La Nina, consistent with observational evidence. However, their period change is quite different among models, possibly due to different gravity wave parameterization used, as well as changes of resolved waves and large-scale circulations. By combining various observational data with model experiments, it is expected new insight will contribute to reducing uncertainties in climate models.

Using new simulations in which zonal mean QBO in the models is constrained by observations, Phase-2 of QBOi is now examining the causes of the systematic biases identified in phase-1, and their consequences for QBO teleconnections to phenomena such as the MJO, stratospheric polar night jet, and NAO.

The predictability of sea ice during extreme sea ice loss events on subseasonal (daily to weekly) time scales is explored in dynamical forecast models. In most regions in the summertime, sea ice forecast skill is lower during extreme sea ice loss events than during nonextreme days, despite evidence that links these extreme events to large-scale atmospheric patterns; in the wintertime, the difference between extreme and non-extreme days is less pronounced. We further explore a case study of the largest Arctic cyclone on record in January 2022 that led to the largest recorded weekly loss in sea ice extent. While the storm was well predicted up to 8 days in advance, subsequent changes in sea ice cover were not, likely due to biases in the forecasts’ sea ice initial conditions and missing physics in the forecast model such as wave-sea ice interaction.

The present study investigates the effect of initial conditions and lateral boundary conditions derived from various analysis and reanalysis datasets [European Centre for Medium-Range Weather Forecasts (ECMWF), National Centre for Environmental Prediction (NCEP) and final analysis (FNL) and National Centre for Medium-Range Weather Forecasting (NCMRWF)] on the prediction of the track, intensity and structure of the extremely severe cyclonic storm Hudhud that develop over the Bay of Bengal in October 2014. Six numerical simulations are conducted at two different initial conditions (12 UTC on 08 October and 12 UTC on 09 October 2014) by using three different datasets namely ECMWF horizontal resolution of 0.75º×0.75º, FNL horizontal resolution of 1.0º×1.0º and NCMRWF horizontal resolution of 0.11º×0.11º on Advanced Research of Weather Research and Forecasting (ARW-WRF) model of version 4.2.2. A double nested domain of the ARW model with a horizontal resolution of 15 km and 3 km is used for all simulations. The model performance and validation have been carried out using the observational data [best-fit track data of the India metrological department (IMD)]. The model predicted track, intensity, and minimum sea-level pressure parameters of ECMWF data are well-matched with the observational datasets. Whereas the model simulated mean absolute maximum wind speed errors are about 4.5 m/s, 12.5 m/s and 9 m/s for ECMWF, FNL and NCMRWF respectively. The statistical analysis such as bias, mean error and standard deviation are estimated for the model predicted parameters from different data sets and presented its significance and importance in the forecast of Bay of Bengal cyclones. This study will be extended with the number of cases to evaluate the performance of the model in predictions of extremely severe cyclonic storms over the Bay of Bengal and highlights the significant changes in dynamic and thermodynamic parameters during the rapid intensification of cyclonic storms.

The Global Carbon Project estimates that the terrestrial biosphere has absorbed about one-third of anthropogenic CO$_2$ emissions during the 1959-2019 period. This sink-estimate is produced by an ensemble of terrestrial biosphere models and is consistent with the land uptake inferred from the residual of emissions and ocean uptake. The purpose of our study is to understand how well terrestrial biosphere models reproduce the processes that drive the terrestrial carbon sink. One challenge is to decide what level of agreement between model output and observation-based reference data is adequate considering that reference data are prone to uncertainties. To define such a level of agreement, we compute benchmark scores that quantify the similarity between independently derived reference datasets using multiple statistical metrics. Models are considered to perform well if their model scores reach benchmark scores. Our results show that reference data can differ considerably, causing benchmark scores to be low. Model scores are often of similar magnitude as benchmark scores, implying that model performance is reasonable given how different reference data are. While model performance is encouraging, ample potential for improvements remains, including a reduction in a positive leaf area index bias, improved representations of processes that govern soil organic carbon in high latitudes, and an assessment of causes that drive the inter-model spread of gross primary productivity in boreal regions and humid tropics. The success of future model development will increasingly depend on our capacity to reduce and account for observational uncertainties.

Warm conveyor belts (WCBs) are weather systems that substantially modulate the large-scale extratropical circulation. As they amplify forecast errors and project them onto the Rossby wave pattern, they are of high relevance for numerical weather prediction. At the same time, the ascent of WCBs is strongly driven by latent heat release from cloud-condensational processes, whose representation in forecast models is prone to uncertainties.

In the first part of this contribution, we investigate the impact of model uncertainty schemes on WCBs and other diabatically driven air streams – detected by trajectory analysis – by running sensitivity experiments with the ensemble prediction system of the European Centre for Medium-Range Weather Forecasts (ECMWF). We show that the ‘stochastically perturbed parametrization tendencies’ (SPPT) scheme results in systematically increased occurrence frequencies of rapidly ascending trajectories, despite the symmetric perturbations imposed. The magnitude of the increase scales with the integrated heating rate along the ascent, which makes the effect more pronounced in (sub-) tropical areas and in autumn and spring than in the midlatitudes and in winter. As the physical characteristics of the trajectories remain unchanged by SPPT, we hypothesize that the unilateral effect on the frequencies results from nonlinearities in the triggering of rapid ascents.

In the second part, we analyze the role of WCBs for the growth of forecast errors in operational medium-range forecasts of ECMWF. We show that forecasts with high WCB-activity over the North Atlantic have on average lower forecast skill than forecasts with low WCB-activity, and that the forecast time when the error growth is maximized is characterized by anomalously high WCB-activity. Composites of normalized forecast errors centered on WCB objects reveal that WCBs are associated with a characteristic spatio-temporal pattern of increased forecast errors, which is primarily associated with the representation of an upper-level ridge in the outflow region of WCBs.

The impact on global simulations of a new package of physical parametrizations in the Met Office Unified Model will be described. The main component of the package is an entirely new convection scheme, CoMorph, but it also includes significant changes to the cloud, microphysics and boundary layer parametrizations. The package, called CoMorph A, is tested in a variety of global model configurations and evaluated against a range of standard metrics. Overall it is found to perform well against the control. Biases in the climatologies of the radiative fluxes are significantly reduced across the tropics and sub-tropics, tropical and extratropical cyclone statistics are improved and the MJO and other tropical waves are strengthened. Additionally it improves most scores in NWP trials. There is still work to do to improve diurnal cycle of precipitation over land, where the peak remains too close to the middle of the day. Interestingly, the basic pattern of regional biases in average tropical precipitation remains largely unaffected by the change of convection scheme. This suggests these must at least in part be caused either by some aspect common to both convection schemes or by a systematic bias somewhere else in the model (possibly unconnected with convection). Significant further developments to CoMorph are planned in the near future that may help answer this.

In this study we aim to investigate how the upper tropospheric Rossby wave activity is represented in state-of-the-art climate models. Our analysis focuses on the wintertime large scale circulation over the Northern Hemisphere and the Euro-Atlantic region. Pairs of standard and high resolution historical coupled simulations (performed in the framework of the PRIMAVERA project), have been analysed to investigate model biases in representing the spatial distribution and temporal evolution of wintertime Rossby wave activity and evaluate the benefits of increased resolution. ERA 5 reanalysis have been used as reference in the comparison.
A diagnostic based on Local Wave Activity (LWA) in isentropic coordinates is used to identify Rossby waves and to quantify their amplitude. This diagnostic is then combined with a tool to compute weather regimes, in order to obtain the spatial distribution of Rossby wave activity associated with each weather regime over the Euro-Atlantic (EAT) sector.

When examining the spatial distribution of transient wave activity in the Northern Hemisphere, only a minimal improvement is found in the high-resolution ensemble. On the other hand, when examining the temporal variability of wave activity, a higher resolution is beneficial in all models apart from one. In addition, the Rossby wave activity time series show no evident trends in the historical simulations (at both standard and high resolutions) and in the observations. Over the EAT sector, a marked inter-model variability is found: an increased horizontal resolution improves the models' performance only for some of the models and for some of the regimes. A positive impact of an increased horizontal resolution is found only for the models in which both the atmospheric and oceanic resolution is changed, whereas in the models in which only the atmospheric resolution is increased, a worsening model performance is detected.

Finally, a tracking algorithm is applied to the LWA field to compute the trajectories of RWPs. This allows us to analyse the properties of single RWPs, such as their lifetime, spatial density, genesis and decayment regions etc… and to identify model biases against reanalysis.

The Weather Research and Forecasting Model coupled with Chemistry (WRF-Chem) is used alongwith Aerosols Optical Depth (AOD) from Moderate Resolution Imaging Spectroradiometer (MODIS) satellite retrievals and ground observations to estimate Particulate Matter (PM). The typical dust storm events 17th-22nd April 2010 and 05th-10th May 2010 which has severely affected air quality in North and Northwestern India has been simulated. The model captures the spatial pattern of AOD very well however, it underestimates high aerosol loading in comparison to MODISAOD. The modeled AOD (MODELAOD) shows an underestimation by 37% with MODISAOD over the study region. Therefore, the WRF-Chem model PM10, and MODELAOD are scaled using satellite MODISAOD to provide a better estimation of the particulate pollution. The results shows better estimation, trend and correlation (R=0.83) of the PM10 with hourly observations at Delhi monitoring station and a Mean Bias (MB) of 61μg/m3 during the satellite overpass time. The comparison of estimated PM with daily averaged observations of PM10 from Central Pollution Control Board (CPCB) at stations of Jaipur, Jodhpur, Kota, and Delhi showed a strong agreement with an correlation of (R) of 0.81, 0.70, 0.77 and 0.78, respectively.
Keywords: MODIS, WRF-Chem, AOD, Particulate Matter, PM10, AERONET

Water scarcity is one of the critical challenges of the 21st century, mainly in arid and semi-arid regions, due to the consequences of the growing population and climate change. The present study evaluates the blue water-saving potential through deficit irrigation (DI) and organic mulching (OM) over the Nile Basin countries. It focuses on blue water savings in irrigated agriculture. The AquaCrop model and the global WF accounting standard were used to calculate the blue water footprint of crops for current conditions and a water-saving scenario with DI & OM. The model was performed at a spatial resolution of 5x5 arc-minute grid cells in eleven Nile Basin countries for five selected crops. The blue WF of the selected crops was the largest in Egypt, Sudan, South Sudan, and Tanzania. For the current condition, the total blue water footprint was 48.5 k m3/y, largely located in Sudan (55%) and Egypt (34%). Production of sorghum accounts for the largest share of the blue WF (50%) followed by maize (21%), rice (16%), groundnut (9%), and millet (4%). The largest blue water footprint in Sudan is due to the large harvested area in the region. Deficit irrigation combined with mulching could reduce the current blue WF by 42%. The findings underline that DI combined with OM could reduce blue WF and help in sustainable water use in water-scarce regions.
Keywords: Water-saving; water footprint; deficit irrigation; mulching; Nile Basin countries

Cloud fraction parametrizations (CFPs) are required to represent subgrid cloud cover in simulations ranging from coarse-scale climate to high-resolution regional NWP models. Here, we present a novel, pragmatic approach to CFPs, including a prognostic treatment of frozen cloud fraction with tendencies from several microphysical processes, and a diagnostic treatment of liquid cloud fraction. The latter is diagnosed from turbulent properties, allowing for a bimodal subgrid saturation-departure distributions in entrainment zones.
Simulations were performed with the Met Office Unified Model Regional Atmosphere configuration for a tropical, mid-latitude and arctic domain, each centred around one of the U.S. Department of Energy Atmospheric Radiation Measurement supersites, covering two contrasting seasons at each site. Furthermore, simulations for the entire COMBLE mixed-phase cloud campaign over Norway were performed. Sensitivity experiments included the treatment of the liquid and frozen cloud phase (diagnostic or prognostic), as well as the cloud phase overlap. All simulations include a new double-moment microphysics scheme (CASIM).
Using a novel, holistic evaluation technique, looking at joint, colocated biases of radiation and cloud regimes, it is shown that the hybrid approach outperforms a fully diagnostic or fully prognostic approach to CFP in terms of cloud-radiative effects for simulations with abundant partial cloudiness. However, it is also shown that over the arctic, simulations without a CFP generally exhibit cloud-radiative effects most similar to the observations.
Furthermore, the evaluation shows that most configurations suffer from compensating biases of underforecasting optically thin clouds and overforecasting deep convective clouds in the warm season.
For the COBMLE campaign, it is shown that the hybrid approach combines the benefits of prognostic ice and diagnostic liquid cloud fractions for cloud phase, using two state-of-the-art ground-based remotely sensed phase identification algorithms. The treatment of phase overlap is shown to be important for cloud evolution and surface precipitation rates in this environment.

Convective storms, and their associated hazards (flash floods, hail, lightning and severe winds), are a crucially important forecasting problem. Such events can have wide-ranging impacts on livelihoods and infrastructure, so the timing and location of convective storms, as well as their evolution, are important to forecast accurately. Many operational forecast centres, including the Met Office, currently run numerical weather prediction models at order 1 km gridlength for short-range weather forecasting. While there is evidence that precipitation forecasts at this grid length are more accurate than lower resolution forecasts, there are still significant shortcomings in the nature of the simulated convective cells. Cells in the model tend to be too large and too intense and tend not to organise into Mesoscale complexes as observed. We expect that this is a consequence of the fact that convection is not fully resolved at this grid length as well as illustrating our lack of understanding of the nature of small-scale mixing and microphysical processes. To overcome this, many operational centres (including the Met Office) have started exploring the use of order 100 m gridlength models for simulating convection.
In this study we perform simulations with the Met Office Unified Model (UM) at horizontal grid lengths ranging from 1.5 km to 100 m, which allows us to apply a statistical approach to evaluate the properties of the simulated storms over a range of conditions. While some aspects of convection are improved by going to higher resolution (such as initiation time, storm depth), there are still problems – in particular O(100 m) models tend to produce too many small, precipitating showers. This may be a sign that the subgrid turbulence scheme is not handing over to the explicit turbulence appropriately.

Global weather and climate models have significant regime-dependent systematic errors of cloud and its impacts on radiation, many of which are common across a range of different modelling systems. Here we describe how the systematic cloud and radiation errors in the ECMWF global numerical weather prediction (NWP) model have reduced over the last decade, and how the use of observations and data assimilation has helped to understand the source of errors and implement improvements to the physical parametrizations.

In particular, top-of-atmosphere reflected shortwave radiation is a key quantity for model assessment as it is well-observed globally, fundamental for driving the atmospheric dynamics and is strongly affected by cloud. It is shown how short-range forecasts can be used to understand the model radiation errors and link them to specific deficiencies in moist physics parametrizations. Examples include the radiation biases linked to boundary layer cloud structure over the subtropical oceans, and storm-track radiation biases linked to the representation of supercooled liquid water.

In addition, direct and indirect observations of cloud are becoming increasingly incorporated in data assimilation systems. Cloud-affected (all-sky) radiances at microwave frequencies (and to a limited extent, infrared frequencies) are routinely assimilated at ECMWF and there is increasing interest in the possibility of assimilating radar, lidar and shortwave (visible) wavelengths. This puts increasing emphasis on the fidelity of the model cloud fields in terms of the amount of condensate, cloud phase, vertical structure, particle properties, and their impact on radiation.

There are therefore strong motivations to further improve the cloud and radiation fields in models across all timescales, from hours (data assimilation), to days (weather forecasts) to decades (climate).

The demand for more accurate forecasts of tropical cyclone (TC) track and intensity with longer lead times is greater than ever, in part due to the enormous economic and societal impacts of TCs. Over the past decade, NRL has developed a high-resolution, limited-area prediction system, COAMPS-TC, specifically designed for TC track, intensity, and structure prediction. Through this experience, we have gained insight into the diagnosis of systematic TC forecast errors and have addressed a number of these errors. In this presentation, we will provide a summary of model deficiencies impacting TC track, intensity, and structure forecasts that we have encountered, as well as the components and processes they have been attributed to, and how we have addressed them.
The TC intensity and structure forecast errors are sensitive to the microphysics parameterization including the interaction with radiation, as well as the representation of shallow convection. On the inner-most storm-following grid mesh (4-km horizontal resolution) COAMPS-TC does not utilize a deep convection parameterization. The TC forecast intensity errors are also sensitive to the vertical mixing within the explicit convective clouds. COAMPS-TC is coupled in a two-way sense with an ocean model (Navy Coastal Ocean Model) and we find that the forecast intensity and structure errors, as well as the pressure-wind relationship, are closely dependent on the surface drag coefficient and nature of the air-sea coupling. Rapid intensification (RI) of TCs (equal to or greater than a 30 kt intensity increase in 24h) is notoriously difficult to predict. An evaluation of forecasts since 2018 reveals marked improvements in COAMPS-TC RI predictions over the last several years, although many challenges remain. Track forecast errors are also found to be especially sensitive to the deep convective parameterization, shallow convection, and microphysics, while being relatively insensitive to the air-sea coupling.

During the last decades, CMIP5 models simulate a warming trend in the tropical eastern Pacific that has not been present in observations (Sieger et al., 2019). Associated with this, the Walker circulation has experienced a westward migration while CMIP5 models simulate an eastward migration. This mismatch is still present in CMIP6 models and might affect climate projections worldwide. In the Caribbean region, CMIP6 models project a strong drying at the end of the 21st century. El Niño-like changes of the Walker circulation are the dominant processes driving the drying. The models that project a strong Caribbean drying also simulate generally a strong equatorial eastern Pacific warming trend over the recent decades. Thus, the mismatch between observed and simulated warming trends over the equatorial eastern Pacific questions the reliability of the Caribbean precipitation projections. The warming bias might also have implications for tropical cyclones' projections in the Atlantic and Pacific through the effect of vertical wind shear, which is related to shifts in the Walker circulation. In addition, the double Intertropical Convergence Zone (ITCZ) bias might be influenced by the mismatching trends. The strong influence of El Niño-Southern Oscillation (ENSO) dynamics on the world’s climate demands more in-depth studies addressing the drivers of the Walker circulation and the equatorial Pacific warming bias. A detailed discussion on plausible drivers of the bias will be presented.

El Niño–Southern Oscillation (ENSO) is the leading source of global seasonal climate variability, with teleconnections to many regions of the world, therefore its prediction is crucial to overall extratropical seasonal forecast skill. In this study, we evaluate systematic ENSO-related seasonal forecast errors within several different operational forecast models, based on multi-decade seasonal hindcast datasets. We find that the predictions have a systematic westward SST anomaly bias, whereby the eastern-central tropical Pacific SST anomalies associated with ENSO events extend too far to the west for anomalies of either sign. Associated with this SST forecast error is a westward shift of ENSO rainfall anomalies, which in turn affects extratropical seasonal forecast skill through errors in wave propagation from the tropical Pacific. The ENSO-related forecast errors, which are also typical of long free-running climate model simulations, are apparent almost immediately; in fact, they develop so rapidly that they are primarily a function of the seasonal cycle, rather than lead time. That is, the pattern and even amplitude of ENSO-related tropical anomaly errors for a given month are very similar over a range of forecast lead times. Predicted ENSO events also tend to decay too slowly compared to observations, resulting in large systematic forecast errors in the eastern tropical Pacific in late winter/early spring, which are also well-developed at short forecast lead times. The effect of these errors on extratropical seasonal forecast skill will be discussed, as well as possible causes of the ENSO-related biases as seen in the development of the daily forecast error over the first few forecast months.

Climate models have significant biases in simulating the climatological mean of the South Asian Summer Monsoon (SASM) precipitation. From CMIP3 to CMIP6, climate models, continue to simulate dry biases over the northern Indian sub-continent and wet biases over the Indian ocean to the south. Here, we explore whether SASM biases in climate models can be understood in terms of biases in the seasonal migration of the Inter-Tropical Convergence Zone (ITCZ) over the Indian subcontinent. From this perspective, SASM precipitation biases may be interpreted as biases in the placement of the ascending branch of the Hadley cell in this region during boreal summer. We show that SASM precipitation biases in CMIP6 models are coincident with systematic radiation biases over tropical and extratropical regions of both the hemispheres, and these biases are consistent with a southward bias in the ITCZ placement over the region. We hypothesize that southern hemispheric radiation biases, which arise from poor representation of low stratocumulus cloud decks in the southern hemisphere tropics restrict the northward extent of continental ITCZ migration in SASM region. We assess this hypothesis by presenting results from CESM2 experiments with cloud locking over strategic region of interest, in an effort to correct SASM biases. In short, we explore the links between biases in SASM precipitation and southern hemisphere radiation, which will provide guidance for future development of models with realistic SASM precipitation simulation.

The diurnal cycle is the most prominent mode of rainfall variability in the tropics, governed mainly by the strong solar heating and land–sea interactions that trigger convection. Over the western Maritime Continent, complex orographic and coastal effects can also play an important role. Weather and climate models often struggle to represent these physical processes, resulting in substantial model biases in simulations over the region. For numerical weather prediction, these biases manifest themselves in the initial conditions, leading to phase and amplitude errors in the diurnal cycle of precipitation. Using a tropical convective-scale data assimilation system, we assimilate 3-hourly radiosonde data from the pilot field campaign of the Years of Maritime Continent, in addition to existing available observations, to diagnose the model biases and assess the relative impacts of the additional wind, temperature, and moisture information on the simulated diurnal cycle of precipitation over the western coast of Sumatra. We show how assimilating such high-frequency in situ observations can improve the simulated diurnal cycle, verified against satellite-derived precipitation, radar-derived precipitation, and rain gauge data. The improvements are due to a better representation of the sea breeze and increased available moisture in the lowest 4 km prior to peak convection. Assimilating wind information alone was sufficient to improve the simulations. We also highlight how during the assimilation, certain multivariate background error constraints and moisture addition in an ad hoc manner can negatively impact the simulations. Other approaches should be explored to better exploit information from such high-frequency observations over this region.

Increasing the skill of extended-range tropical cyclone (TC) forecasts has the potential to increase disaster response and evacuation lead time and improve ship routing. An important phenomena for these predictions is the Madden-Julian Oscillation (MJO) which dominates variability in the intraseasonal (30-90 day) timescale. We examine the skill in the Navy Earth System Prediction Capability (ESPC), which is a coupled global model run out to 45 days. The MJO in the Navy ESPC has been shown to be too strong and too fast, which has implications for the MJO-TC relationship in the model.

First, we examine the biases in TC track and genesis using the version of the Navy ESPC run for the Subseasonal Experiment (SubX). We examine biases in the genesis potential index (GPI) which combines the environmental impacts of absolute vorticity, vertical shear, relative humidity, and potential intensity. We also examine the mean biases in the GPI as well as the biases in relationship between GPI and the MJO. We find that there are negative biases in the GPI throughout the northern tropics during the boreal summer (May-November) in Navy ESPC. We evaluate how each term in the GPI contributes to this mean state bias and is modulated by the MJO. We also examine the biases in TC tracks and genesis frequency in the model and use process-oriented diagnostics to link the biases in TC tracks and genesis to physical processes in the Navy ESPC.

We next examine the effect that Analysis Correction-based Additive Inflation (ACAI) has on GPI and TCs in the Navy ESPC over the boreal summer (May-November). ACAI is a method which acts to reduce systematic errors in the model through a bias correction applied to the tendency terms of surface pressure, temperature, specific humidity, and horizontal wind components. ACAI reduces the root mean squared error (RMSE) and improves the spread-skill relationship for the total and MJO-filtered OLR and low-level zonal winds. The impact of ACAI on TCs is examined through the mean state of several environmental variables important for TC genesis and tracks. ACAI improves the spread-skill ratio in all these environmental variables, and reduces the bias and RMSE of shear, vorticity, and potential intensity, however moisture biases increase with ACAI.

The Madden Julian Oscillation (MJO) is the primary mode of intraseasonal, convectively coupled wave variability in the tropics, and has teleconnection impacts around the globe. With its spontaneous, intermittent and multi-scale nature it represents a unique challenge for course resolution GCM’s and its parameterized physics. Given its emerging importance in extending forecast predictability in models, monitoring the MJO performance during model development, as with other important modes of variability, is paramount.

In the NCAR Community Earth System Model (CESM2), the MJO is now a well simulated feature, in terms of propagation speed, strength and regional coherence, and we choose to make this model our tool in a series of sensitivity experiments. When observed SSTs are prescribed in the atmosphere model (CAM6) the MJO is largely absent, similar to previous model versions. If the logical hypothesis is that high frequency ocean coupling is necessary to sustain events, then it follows that MJO signal coherence degrades when the hourly coupling frequency is decreased to daily. This confirms previous work on the role of SST anomaly coupling in MJO propagation and strength.

However, a range of modifications to surface forcings that drive CAM6 (including SSTs, the maritime continent specification and regional slab ocean ‘patches’) reveals that the requirement for interactive surface coupling may not be critical for the model’s MJO. We show how the atmospheric basic state changes in response to surface forcing, and how critical aspects including the poleward humidity gradient, surface fluxes and changes in process tendencies also enable the support of MJO events. This is in contrast to experiments that test the role of physical parameterization changes and reveal most MJO sensitivities are secondary. The take home message is that there may be multiple pathways in this model’s ecosystem through which an MJO can be supported.

Accurate representation of the stratosphere in NWP models is important for accurate analysis/reanalysis products. It is also important for extended-range and seasonal forecasting because stratospheric variability can influence tropospheric weather patterns on these timescales. The first part of this talk reviews recent efforts in alleviating stratospheric biases in the ECMWF model, ranging from the global-mean and zonal-mean temperature biases to the inter-annual variability bias affecting the spring-time Southern Hemisphere polar vortex. The second part of this talk focuses on the lessons learned from km-scale global modelling with regards to the representation of gravity waves, which contribute to the driving of the stratospheric circulation and variability.

Two-way coupling between the stratosphere and troposphere is recognized as an important source of subseasonal-to-seasonal (S2S) predictability. Stratospheric processes involving phenomena such as the stratospheric polar vortex and the quasi-biennial oscillation (QBO) can modulate the tropospheric circulation in predictable ways and/or provide forecast windows of opportunity. S2S forecast models may struggle to represent such coupling processes; at longer lead times, drifts in a model’s circulation related to model configurations, biases, and parameterizations have the potential to feedback and affect stratosphere-troposphere coupling. Fortunately, biases within the stratosphere generally provide useful insights into their likely sources because of well known radiative and dynamical controls on stratospheric temperatures and winds.

This presentation will highlight results from an international SPARC-SNAP (Stratospheric Network for the Assessment of Predictability) community effort to diagnose and characterize stratosphere-related biases in S2S models, and better understand their impacts to predictability. We compare biases among more than 10 S2S prediction systems, including systems participating in the S2S database, and other recently released systems from NOAA (National Oceanic and Atmospheric Administration) and NCAR (National Center for Atmospheric Research). Our effort highlights several common biases related to global mean stratospheric temperatures, stratospheric polar vortex variability, and maintenance of the QBO. Our ongoing work seeks to characterize how such biases affect stratosphere-troposphere coupling, and to what extent these may influence forecast skill in the troposphere.

Large biases in ocean and consequently climate models arise from the inaccurate representation of mesoscale eddy dynamics. Eddies are important drivers of oceanic heat transport and atmosphere-ocean-coupling, and eddy kinetic energy makes up a large part of the total kinetic energy in the oceans. However, in many state-of-the-art ocean model setups, effects of mesoscale eddies need to be at least partially parameterized and eddy variability and energy are often substantially underestimated by a factor of two or more.
We present a set of deterministic kinetic energy backscatter schemes with different complexity as an alternative to classical momentum closures that can alleviate eddy related biases at eddy-permitting resolution, including oceanic biases in the mean currents, in sea surface height variability and in temperature and salinity. The complexity of the schemes is reflected in their computational costs, the related simulation improvements, and their adaptability to different resolutions. However, all schemes outperform classical viscous closures and are computationally less costly than a related necessary resolution increase to achieve similar results. We also outline further potential for refinement of these novel schemes.

Ocean-atmosphere coupled climate and forecast models represent the transfer of energy across the air-sea interface by surface latent and sensible heat fluxes using a variety of bulk flux algorithms. These algorithms all apply parameterized transfer coefficients to bulk inputs of surface temperature and near-surface wind speed and air temperature to estimate the flux. The transfer coefficients—which may be affected by low-level stability, surface currents, wave state, salinity, and wind speed—are estimated using a variety of empirical relationships that exhibit a variety of functional forms, and thus produce a range of fluxes for a given set of inputs. Previous work by Brunke et al. (2003) has shown that many flux algorithms used in climate and forecast models overestimate fluxes by 10-20% when compared to in situ fluxes calculated using direct covariance methods. That study also identified the COARE flux algorithm (Fairall et al. 1996; 2003) as one of the least problematic algorithms.

In this study, we review a newly developed surface flux diagnostic that evaluates model surface flux biases that arise from biases in the bulk inputs (winds, temperature, humidity) and from parameterized transfer coefficients. This evaluation indicates that the double ITCZ bias that is present in many CMIP6 models is partly rooted in surface flux algorithm biases. Next, we assess changes to the mean state climate and tropical variability when updating the default Large and Yeager (2004) bulk flux algorithm to the COARE3.5 bulk flux algorithm in the DOE E3SM and NCAR CESM2 models. The double ITCZ bias is reduced in both coupled and uncoupled simulations of both models, while improvements to the Madden-Julian oscillation are more pronounced in uncoupled experiments. We conclude with a proposed framework to understand how surface fluxes impact atmospheric buoyancy profiles and convection in the tropics.

UGROW is an ECMWF cross-departmental project focused on Understanding systematic error GROWth from hours to seasons ahead. UGROW-IO is one of the project’s focus themes and assessed the lead-time-dependent evolution of atmospheric and oceanic forecast errors in the Indian Ocean, most notably the cold SST and easterly wind bias in the Equatorial Eastern Indian Ocean (EEIO), which has immediate relevance for prediction of the Indian Ocean Dipole. The errors are more pronounced during boreal summer, when the model rapidly develops an easterly surface wind bias, visible already at week 1. This bias amplifies with time via coupled feedbacks, and eventually manifests in SST, being one of the dominant errors at the seasonal time scales in SEAS5.
We have studied the dependency of this error to ocean and atmosphere initial conditions, ocean and atmospheric resolution and different model cycles. We conclude that there are two fundamental and independent sources of errors that lead to the SST errors in SEAS5. The first one is of atmospheric nature, and is largely related with too stable easterly circulation, characterized by the lack of response of the local winds to local surface heating. This induces an easterly bias which leaves the model predominantly in a state with a shallow thermocline and cold SSTs in the EEIO. The second error is of oceanic origin, associated with a too shallow thermocline, which enhances the SST errors arising from wind errors. This type of ocean state mainly comes from the ocean initial conditions, which depends on both the quality of the assimilation and the ocean model. The version of the ocean model used for the forecast can also play a non-negligible role at seasonal time scales, by amplifying or damping the subsurface errors in the initial conditions due to the strength of the atmosphere-ocean coupling in this region.

In the historical simulations of CMIP5 and CMIP6, nearly all coupled climate models underestimate Arctic wintertime air temperatures. The too cold mean conditions are always connected with positive bias in the thickness of sea ice in the Arctic Ocean, hence introducing substantial uncertainty into projections of future Arctic climate change. We assume that the absence of representation of heat fluxes through sea ice leads in global climate models contributes to this cold bias. Using the EC-Earth3 model, we introduce a new parameterization to better simulate the turbulent heat fluxes through sea ice leads which effects are most prominent during winter. We study the possibilities of minimizing model bias in Arctic surface fields based on the CMIP6 historical simulations. The long-term sea ice satellite observations and ERA5 reanalysis are used for model evaluation and assessment of changes in the mean states and climate trends in the Arctic (e.g. sea ice extent, area, volume, and surface air temperature) resulting from the parameterized effect of leads during a period of rapid Arctic change (1980-2014). In winter, the enhanced heat fluxes diminish the known cold bias in the regions with excessive sea ice cover (the Barents Sea and the Fram Strait) and thick ice (north of Greenland). Nevertheless, the new parameterization adds warm bias, particularly reflected in sea ice concentration, in the Bering Strait to the EC-Earth CMIP6 historical simulations. To assess the role of leads in determining the surface energy budget in the Arctic and to determine how much we can improve the systematic biases in the near-surface air temperature and the representation of near-surface gradients under strongly-stable stratification, we will include the same implementation and analysis with the Norwegian Earth System Model, NorESM2.

This presentation evaluates the simulation of wintertime statistics of the near-surface atmosphere and surface energy budget observed in the Central Arctic during the MOSAiC campaign with short-term forecasts from seven state-of-the-art operational and experimental forecast systems. Five of the forecast systems are fully-coupled ocean-sea ice-atmosphere models. It is only in a coupled system that models can simultaneously simulate the impact of radiative effects, turbulence, and precipitation processes on the surface energy budget and near-surface atmospheric conditions. The focus of the presentation is on coupled processes unique to the Arctic, such as, the representation of liquid-bearing clouds at cold temperatures and the representation of a persistent stable boundary layer. We show that models still struggle to maintain liquid in clouds at cold temperature, with only one of the seven models producing cloud liquid similar to observations. Only two models simulate the observed distinct bi-modal clear-sky and cloudy modes. One of these models has cloud liquid similar to observations and the other produces enough cloud ice without cloud liquid to produce two distinct modes. Three of the models have a distinct clear-sky modes but underestimate the cloudy mode. Two of these models are the only models that produce the observed shutdown of turbulence for strongly stable near-surface conditions. Using a diagnostic that displays all three terms of the surface energy budget, it is seen that these three models have variability in two unobserved regimes; clear-skies with large upward conductive ground fluxes and small sensible heat fluxes, and large downward sensible heat fluxes and small conductive ground fluxes. These biases significantly impact the ability to simulate the evolution of the coupled Arctic system.

Warm conveyor belts (WCBs) are rapidly ascending airstreams in extratropical cyclones that affect the dynamics and predictability of the large-scale midlatitude flow. Based on reanalyses and reforecasts of the subseasonal-to-seasonal (S2S) database, this study systematically evaluates the representation of WCBs in ECMWF’s sub-seasonal forecasts which is now possible through the development of a novel machine learning diagnostic. The study unveils how the representation of WCBs is linked to systematic biases and the predictive skill of the large-scale flow over Europe. In the first part, we show that blocking over the European region is preceded by anomalously high WCB activity over the western North Pacific and the North Atlantic. Though this link is reasonably well represented in correct predictions of European blocking, misforecasts are characterized by an alternative pathway towards European blocking. In the second part, we investigate reasons for the enhanced WCB activity over the western North Pacific and find that it is linked to the Madden-Julian Oscillation (MJO). Accordingly, we consider WCBs to be a linking element between the MJO and known teleconnection patterns towards the midlatitudes. In reforecasts of the S2S database, this link between the MJO and WCB activity weakens substantially towards subseasonal lead times. This likely contributes to a weakening of known teleconnection patterns and prevents the full exploitation of the MJO-related predictability potential.

The ECMWF reanalyses (most recently ERA5) are a critical resource in understanding climate change. Using a state-of-the-art data assimilation system to retrospectively analyse historical observations, we build up a comprehensive picture of how the global atmosphere has changed since observations began up to the present day. Data assimilation techniques such as 4D-Var produce an optimal blend of direct information from the observations with background information from the model, which compensates where the measurements are incomplete in space or time. Broadly speaking, in modern periods when we have access to constellations of multi-sensor carrying satellites, reanalyses are highly constrained by the observed data and rely less on background information from the model. In pre-satellite times reanalyses rely much more on model information to augment the sparse observation network. Inevitably, in doing so, the resulting analyses retain some systematic error characteristics of the model. This is evident from artefacts in the anomaly time-series when major changes in the global observing system occur.

From the operational implementation of weak-constraint 4D-Var on the 30th of June 2020, we have built up a detailed understanding of systematic model errors. A possible innovation for the ERA6 reanalysis is to use a climatology of model error estimates (diagnosed during the present-day well observed era) to correct for model systematic error during poorly observed periods. Pilot experiments, where poorly observed periods have been simulated by deliberately removing data, suggest that this technique works well and will produce significantly improved reanalyses with less sensitivity to changes in the observing system.

Understanding and limiting the spread of ocean carbon sink projections are crucial to effectively guide the development of climate mitigation policies, determine an accurate future carbon budget and subsequently climate change. The North Atlantic and the Southern Ocean are two of the most intense sink regions for anthropogenic CO2 emissions. Nevertheless, CMIP5/6 models simulate growing inter-model spread in future carbon uptakes in these regions. In this study, we apply an emergent constraint approach to reduce the projections uncertainties under the high-CO2 future scenario. The efficiency of surface-to-deep transport of anthropogenic carbon is commonly identified as the key mechanisms driving the systematic inter-model spread. For the North Atlantic region, we further use a genetic algorithm to optimize our identified emergent constraint relationship by isolating the region where contemporary model bias strongly correlates with the projection spread. Our study consolidates the importance of improving representations of anthropogenic carbon ventilation mechanisms in models and sustaining carbon and watermass monitoring network in these regions to improve the fidelity of future model projections.

The impacts of the Madden-Julian Oscillation (MJO) on the South American monsoon season (December, January, and February – DJF) and their possible changes during positive (El Niño – EN) and negative (La Niña – LN) phases of the El Niño-Southern Oscillation (ENSO) are analyzed in the UK Met Office Unified Model Global Ocean Mixed Layer configuration (MetUM-GOML3). Experiments sixty years long, with and without ENSO cycle, considering lower (200 km) and higher (90 km) spatial resolution, are performed to assess if the ENSO influences the MJO and its teleconnections to South America (SA). Simulations without ENSO show: (1) an established MJO extratropical teleconnection triggered by enhanced convection in the central-east subtropical South Pacific (SP) (source region), and its strongest impact on precipitation over SA in phase 8, earlier than in observations (phase 1); (2) an extratropical teleconnection, triggered by suppressed convection over the same region, with strongest impact on precipitation over SA in phase 4, with opposite sign; (3) increased horizontal resolution enhances the MJO convection and the anomalous circulation-precipitation dipole over SA, mainly over subtropical SA. However, the extratropical teleconnections at upper levels are slightly shifted east at higher resolution due to an enhanced SA westerly jet with respect to the lower resolution. The ENSO affects the basic state and the MJO convective anomalies, which modulate the MJO teleconnections and their impacts on SA in simulations with ENSO cycles. The extratropical teleconnections and their impacts are stronger under ENSO with respect to those in simulations without ENSO. Hence, both EN and LN states in the model generate forcing in the central-east subtropical SP that more efficiently triggers teleconnections than simulations without ENSO, indicating nonlinear ENSO effects on MJO anomalies over SA.

Keywords: Coupled global models; ENSO-MJO Interaction; South American monsoon; Teleconnections.

Accurate representations of soil state (soil moisture and soil temperature) in numerical weather prediction (NWP) model is crucial for better prediction of large-scale (synoptic) to small-scale eddies) weather activities. In this sense, land data assimilation has been performed by direct improvisation in the initial fields of soil state in recent years. Nevertheless, no study has reported its impact on boundary layer fog simulation.
The present study is a novelty attempt from Winter Fog EXperiment (WiFEX) to investigate the land data assimilation impact on the fog life cycle. As a case study, we have chosen a very dense fog (Visibility < 200m) event that occurred over the Indo Gangetic Plain (IGP) region during 24-25 January 2018. The Noah-MP Land Surface Model (LSM) based High-resolution land data assimilation system (HRLDAS) is used to develop high-resolution (2 km grid resolution) land surface products over the same region. The outcome of SM/ST from HRLDAS is utilized to initialize the land surface fields in the Weather Research and Forecasting (WRF) model. Prior to the case study, the quality of developed high-resolution gridded SM/ST product for 2017–2018 boreal winter month (December-January) is examined with reanalysis and observational dataset. It is found that the actual soil state over the observational site was drier in condition which was reasonably represented in HRLDAS output compared to reanalysis soil products. Therefore, two sensitive experiments have been carried out in the WRF model: CNTL (with default SM/ST) and EHRLDAS (replaced only SM/ST initial field by HRLDAS product) during fog event. The results reveal that the bias in micrometeorological parameters (T2, RH2, WS10) are significantly improved in EHRLDAS simulation, where it shows small root mean square error (RMSE) and better Index of Agreement (IoA) than CNTL simulation. As a result, error in fog onset timing is notably reduced and the vertical representation of fog is skillfully demonstrated.

Stochastic parameterizations are continuously providing promising simulations of unresolved atmospheric processes for global climate models (GCMs). To better represent organized convection in the Climate Forecast System version 2 (CFSv2), the SMCM parameterization is adopted in CFSv2 in lieu of the pre-existing simplified Arakawa–Schubert (CTRL) cumulus scheme and has shown essential improvements in different large scale features of tropical convection. One of the features of earlier SMCM is to mimic the life cycle of the three most common cloud types (congestus, deep, and stratiform) in tropical convective systems. In this present study, a new cloud type, namely shallow cloud, is included along with the existing three cloud types to make the model more realistic. The cloud population statistics of four cloud types (shallow, congestus, deep, and stratiform) are taken from Indian (Mandhardev) radar observations and a Bayesian inference technique is used to generate key time scale parameters required for the SMCM as implemented in CFSv2 (hereafter CFSSMCM-4cloud). The 4-cloud simulation improves many aspects of the mean state climate compared to CTRL, and 3-cloud (CFSSMCM-3cloud, where the three most common cloud types are considered) simulation. Significant improvement is noted in the rainfall PDF over the global tropics. The global distribution of different clouds, mainly low-level and mid-level clouds, is also improved. The 4-cloud simulation shows significant improvement with respect to the double ITCZ (The Intertropical Convergence Zone) problem as well as overall organized convection. The convective and large-scale rainfall simulation is investigated in detail.

Atmospheric models, used for either weather or climate applications, encompass so-called parameterizations, which aims at summarizing and quantifying the impact on the resolved model variables of radiative, thermodynamical, or chemical processes, as well as dynamical processes that occur at scales smaller than the computational grid. Though these parameterizations are developed on a physical basis, some simplifications underlying them introduce parameters that need to be properly calibrated to achieve a skillful model. Given the number of parameters (several tens), the possible number of performance metrics used for validation or evaluation, and the computational cost of models, the modelling experts need help to better address this calibration bottleneck.

In this work, we experiment the history matching with iterative refocusing framework with the CNRM-CM6-1 climate model to assess whether the model current deficiencies are related to poor model calibration, or if they critically rely on the scientific content of the model. Using a rather small physics perturbed ensemble of short simulations as a learning dataset, Gaussian processes are used to cheaply explore the full space of model parameters and identify the part of it, which provides model configurations compatible with references, given a set of usual performance metrics and the various sources of uncertainty in the whole process. The calibration framework also builds on several waves of true model simulations, to parsimoniously increase the size of the learning dataset only where the surrogate model uncertainty needs to be reduced. We show that several new configurations, in which many CNRM-CM6-1 biases are significantly reduced or even removed (e.g., precipitation over West Africa, regional biases in cloud radiative effect), can be found. Though, some CNRM-CM6-1 biases are truly structural (e.g., biases over eastern sides of tropical ocean basins), calling for further understanding and parameterization development.

In NWP models, near-surface forecast variables like 2-m temperature (T2M) or 2-m relative humidity (RH2M) can be subject
to substantial systematic forecast errors for a variety of reasons, e.g. limited quality of external parameter data and
physical properties of soil and vegetation derived therefrom, poor knowledge of soil moisture in deeper layers, and simplifications
or inadequacies in the parameterization of related physical processes. Attributing specific forecast errors to one of these
potential sources is difficult due to the lack of data, and tuning exercises frequently end up with improvements in
some regions or seasons and degradations in others. Significant progress can be made, however, by using information from data
assimilation to adaptively adjust uncertain model parameters in order to minimize systematic errors.

For the ICON-NWP forecasting system operated at DWD, such kind of adaptive parameter adjustment has been developed for several
uncertain parameters affecting the Bowen ratio, the diurnal temperature amplitude and, in the presence of snow cover, the grid-point
averaged albedo. The algorithm builds upon the assimilation of RH2M and T2M by computing time-filtered assimilation increments of these
quantities, which are subsequently used as a proxy for the respective model bias. Time-filtering is accomplished by a Newtonian relaxation approach with a
time scale of 2.5 days, which is needed to separate random from systematic errors and to remove diurnal cycles from the average biases.
In addition, a proxy for the diurnal temperature amplitude bias is computed. Based upon these quantities, a variety of model parameters
whose basic values are estimated from external parameter data are adaptively adjusted, e.g. the minimum stomata resistance of plants, the
minimum bare-soil evaporation resistance, the heat conductivities of the soil and the skin layer, and the snow albedo.

The adaptive parameter adjustment has recently been operationalized in both the global and regional forecasting systems of DWD. The reduction of
T2M and RH2M RMS-errors achieved by this measure has substantial regional and seasonal variability but typically lies on the
order of 5% on a hemispheric average, gradually decreasing with forecast lead time but staying statistically significant throughout the whole
forecast range (180 h in the global system). A weaker but still significant positive impact is seen in the radiosonde verification
of lower-tropospheric temperatures.

This talk will demonstrate the potentials of a hybrid modeling approach called Combined Hybrid-Parallel Prediction (CHyPP), which augments the numerical model with a machine learning (ML) component. This approach, which does not require making any changes to the numerical model, makes frequent periodic interactive ML corrections to the evolving numerical model solutions. CHyPP is not a post processing technique, because unlike a post-processing technique, which corrects only the final numerical solution, it interacts with the numerical solution. CHyPP corrects the full model state for the effects of model errors, substantially reducing both the systematic and transient components of the numerical solution errors caused by model errors. Results will be shown for an implementation of CHyPP on the low resolution atmospheric general circulation model SPEEDY. They will include the results of both forecast experiments in the 0-14-day range and multi-year climate simulations. Special attention will be paid to the investigation of the specific atmospheric processes whose limited accuracy representation in the numerical model leads to the largest ML corrections. The talk will also discuss different approaches for the use of ML to include SST as a prognostic model variable in a numerical atmospheric model.

Initialized coupled model prediction, ranging from subseasonal to decadal timescales, is a key use case of coupled climate models and central to operational climate services. Coupled model biases lead to the emergence of prediction biases and, in most cases, the deterioration of prediction skills. A posteriori bias correction based on climatological model drift is widely used in coupled predictions, but it is inadequate in dealing with nonlinear dynamics, especially coupled dynamics between model components.
In this presentation, we will discuss prognostic coupled model bias correction based on both data assimilation (DA) and machine learning (ML) methods. A new bias correction technique, ocean tendency adjustment (OTA), was implemented in GFDL’s SPEAR real time seasonal prediction system. OTA applies the climatological DA increments to the model prognostically as tendency terms to reduce model bias. While the idea of OTA resembles other bias correction methods used in some reanalysis products, the tendency correction terms for OTA are produced in such a manner that they can be applied to initialized coupled model predictions to prevent model drift of the ocean component, which in turn reduces model drift in other model components such as atmosphere and sea ice. In SPEAR, OTA improves the coupled model’s ability to predict the observed climatology, which leads to improved prediction skill in certain seasonal climate variability.
Despite the effectiveness of OTA in SPEAR, the tendency terms are currently only climatological averages. As part of an international collaborative effort, we used ML models, trained on Argo-era ocean state estimation and DA increments, to improve upon the existing OTA. The goal is to predict flow-dependent tendency terms that can minimize ocean model drift in a coupled model simulation. We will also discuss how learning from DA increments connects to and complements ML-based physical parameterization.

Starting to resolve the oceanic mesoscale in climate models is a step change in model fidelity. This study examines how certain obstinate biases in the midlatitude North Atlantic respond to increasing resolution (from 1° to 0.25° in the ocean) and how such biases in sea surface temperature (SST) affect the atmosphere. Using a multi-model ensemble of historical climate simulations run at different horizontal resolutions, it is shown that a severe cold SST bias in the central North Atlantic, common to many ocean models, is significantly reduced with increasing resolution. The associated bias in the time-mean meridional SST gradient is shown to relate to a positive bias in low-level baroclinicity, while the cold SST bias causes biases also in static stability and diabatic heating in the interior of the atmosphere. The changes in baroclinicity and diabatic heating brought by increasing resolution lead to improvements in European blocking and eddy-driven jet variability. Across the multi-model ensemble a clear relationship is found between the climatological meridional SST gradients in the broader Gulf Stream Extension area and two aspects of the atmospheric circulation: the frequency of high-latitude blocking and the southern-jet regime. This relationship is thought to reflect the two-way interaction (with a positive feedback) between the respective oceanic and atmospheric anomalies. These North Atlantic SST anomalies are shown to be important in forcing significant responses in the midlatitude atmospheric circulation, including jet variability and the stormtrack. Further increases in oceanic and atmospheric resolution are expected to lead to additional improvements in the representation of Euro-Atlantic climate.

Correctly simulating anvil cloud fraction in models is crucial for determining the cloud radiative effects and the general atmospheric state. Here, we use a convection-permitting model with a single-moment and a double-moment microphysics scheme to study the dependence of anvil cloud fraction to horizontal model resolution. In small-domain simulations of radiative convective equilibrium, we find that the anvil cloud fraction keeps increasing from about 10% to about 30% when the horizontal model resolution increases from 4km to less than 100m. By decomposing anvil cloud fraction as the product of detrainment and lifetime, we find that the increase of anvil cloud fraction is primarily due to the increase of detrainment. Several factors may contribute to the increase of detrainment. In the single-moment simulations, the increase of detrainment can be linked to enhanced evaporation and sublimation of cloud. With finer resolution, each unit area of cloud on average has longer perimeter, which can enhance sublimation of anvil clouds in the environment and drive more environmental subsidence. The enhanced environmental subsidence must be balanced by more mass flux in convective updrafts, which lead to more vertical convective mass flux convergence and thus more detrainment at anvil levels. The single-moment scheme likely exaggerates the sublimation effect because of the saturation adjustment assumption, although it serves as a useful and simple illustration of the microphysical processes. In the double-moment simulations, the increase in detrainment with finer resolution is primarily linked with weaker upper-troposphere stability, which is likely due to enhanced entrainment. The weaker upper-troposphere stability requires larger environmental subsidence to balance the similar amount of radiative cooling, leading to greater horizontal cloud mass detrainment and higher anvil cloud fraction.

In the H2020 Next Generation Earth Modelling Systems (NextGEMS) project, we aim at building a new generation of eddy- and storm-resolving global coupled Earth System Models. At storm-scale resolutions (1-9 km), deep convective systems are becoming at least partially resolved, which means that the parameterisation of deep convection can possibly be switched off.

However, this introduces at least two systematic errors. The first error is that at storm-resolving scales, explicit convection is triggered too late and too scarcely because subgrid-scale heterogeneity is not resolved, and convective inhibition is thus too hard to overcome. The second error is that at storm-resolving scales, the mixing of updrafts with their immediate environment is underestimated, as most of the scales on which the mixing happens are not resolved.

Therefore, as soon as the parameterisation of deep convection is switched off, tropical deep convection is associated with too much convective instability, too high updraft buoyancy and too strong updraft velocities, resulting in too intense and too localised precipitation events. The large-scale pattern of mean precipitation is biased as well, particularly the amount of precipitation in the inter-tropical convergence zone is strongly overestimated.

Here we focus on the rain band over the tropical Pacific at 5°N, which is 2-3 times stronger in the IFS simulations for the NextGEMS project than in observations. We perform a sensitivity study, testing how different assumptions in the moist physics schemes relevant to storm-resolving scales affect the results. We show that by choosing a setup that is more suitable for storm-resolving simulations, we can significantly improve the representation of convective precipitation in the IFS, both with respect to the intensity and the spatial pattern in the tropics.

The signal to noise paradox that climate models are better at predicting the real world than their own ensemble forecast members highlights a model error present in almost all current seasonal forecast systems. A consequence of this paradox is that detecting signals in decadal forecasts requires 100 times more ensemble members, representing a significant computational cost. One hypothesis as to why the paradox exists is a deficiency of atmospheric eddy feedback in climate models due to insufficient spatial resolution.

By computing the magnitude of the feedback between transient eddies and large scale flow anomalies in multiple seasonal forecast systems, this study confirms that current systems underestimate this positive eddy feedback, and shows that this deficiency is strongly linked to weak signal to noise ratios in ensemble mean predictions.

Consistently, increased eddy feedback is shown to be linked to greater teleconnection strength between the El Nino Southern Oscillation and the Arctic Oscillation and to stronger predictable signals. A technique to estimate the potential gain in skill that may come from eliminating eddy feedback deficiency is presented, showing that model skill could double in some extratropical regions, significantly improving predictions of the Arctic Oscillation.

Destination Earth – DestinE – is an ambitious initiative of the European Commission, in support of its Digital Strategy and the Green Deal. Bringing together scientific and industrial excellence from across Europe, DestinE will contribute to revolutionising the European capability to monitor and predict our changing planet, complementing existing national (meteorological services) and European efforts (Copernicus services).

Based on the integration of extreme-scale computing, Earth system simulations and the real-time exploitation of all available environmental observations, DestinE will develop high-accuracy digital twins, or replicas, of the Earth. DestinE will thus to allow users of all levels the ability to better explore natural and human activity, and to test a range of scenarios and potential mitigation strategies.

Under the European Commission's leadership, and in coordination with the Member States, scientific communities and other stakeholders, ECMWF, ESA and EUMETSAT are the three entrusted entities tasked with delivering the first phase of the DestinE by 2024.
ECMWF will be responsible for building the ‘digital twin engine’ software and data infrastructure and for using it to deliver the first two high-priority digital twins, while European Space Agency (ESA) provides the platform through which users will access the service, and the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) develops the data repository.

This talk will give a high-level introduction of the programme, and it will particularly focus on the first two priority digital twins. The Digital Twin on Weather-Induced and Geophysical Extremes will provide capabilities and services for the assessment and prediction of environmental extremes. The Digital Twin on Climate Change Adaptation will support the generation of analytical insights and testing of predictive scenarios in support of climate adaptation and mitigation policies at multi-decadal timescales, at regional and national levels.

DestinE’s digital twins will rely on Earth system modelling and data assimilation – the process of combining information from observations and models to distil the most likely current state of the Earth system. Their development will push these capabilities further than ever. Observations will come from many sources, including devices like mobile phones and the internet of things. In addition, new approaches from machine learning and artificial intelligence will be used to improve the realism and efficiency of these digital representations of our world. To increase the added value of the digital twins for societal applications, they will be co-designed and tested with users from impact-sectors such as water management, renewable energy, health and agriculture.

The configuration of the Met Office Unified Model being submitted to CMIP6 has a high climate sensitivity. Previous studies have suggested that the impact of model changes on initial tendencies in numerical weather prediction (NWP) should be used to guide their suitability for inclusion in climate models. In this study we assess, using NWP experiments, the atmospheric model changes which lead to the increased climate sensitivity in the CMIP6 configuration, namely, the replacement of the aerosol scheme with UKCA GLOMAP-mode and the introduction of a scheme for representing the turbulent production of liquid water within mixed-phase cloud. Overall, the changes included in this latest configuration were found to improve the initial tendencies of the model state variables over the first 6 hours of the forecast, this timescale being before significant dynamical feedbacks are likely to occur. The reduced model drift through the forecast appears to be the result of increased cloud liquid water, leading to enhanced radiative cooling from cloud top and contributing to a stronger shortwave cloud radiative effect. These changes improve the 5-day forecast in traditional metrics used for NWP. This study was conducted after the model was frozen and the climate sensitivity of the model determined; hence, it provides an independent test of the model changes contributing to the higher climate sensitivity. The results, along with the large body process-orientated evaluation conducted during the model development process, provide reassurance that these changes are improving the physical processes simulated by the model.

In the dynamical model (NWP or GCM) diagnostic framework, atmospheric dynamical interactions or teleconnection are classically measured as lagged correlations between climate variables at remote locations. However, correlation does not imply causation. In other words, correlation is incapable to legitimately infer a “cause-and-effect” relationship between two events. Naturally, causation seems to require not just a correlation, but a counterfactual dependence. Causal discovery algorithms go beyond correlation-based measures by systematically excluding common driver effects and indirect links. Here, we propose a causal network for model evaluation as a type of process-oriented framework. Based on data-driven causal fingerprints, the causal network can understand differences between models and observations based on the physical process which potentially influences model biases in simulating precipitation and temperature.
Our approach is in three folds. First, we will apply a novel causal network algorithm named PCMCI to the observed data (reanalysis datasets) to find the ocean-atmosphere interactions and their causality on precipitation and temperature. This network will apply mainly to those regions which are known to host sources of precipitation predictability such as the Atlantic, tropical Pacific, and the Indo-Pacific Maritime Continent and its surrounding seas. Secondly, a similar casual network (PCMCI) will be deployed to atmosphere-ocean variables from dynamical models. Finally, the resulting causal fingerprints from observation and models can be compared using pairwise comparisons of all possible links in a network. For example, we will test if a link from MJO to precipitation found in observations is also detected in S2S data sets. For more quantification, following Nowack et al. (2020) a modified asymmetric F1-score which is the harmonic mean of precision (fraction of links in models that also occur in observation) and recall (fraction of links in the observation that are detected in models) will apply to understand differences between models and observations. F1-scores vary between 0 and 1 (perfect network match), so a higher value of F1-score implies Fingerprints closer to observations are associated with smaller precipitation biases in climate models.

References:
Runge, J., Nowack, P., Kretschmer, M., Flaxman, S. & Sejdinovic, D. Detecting and quantifying causal associations in large nonlinear time series datasets. Sci.Adv. 5, eaau4996 (2019).
Nowack, P., Runge, J., Eyring, V. & Haigh, J. D. Causal networks for climate model evaluation and constrained projections. Nat. Commun. 11, 1415 (2020).

In this study we examine the mean-state West Pacific subtropical high (WPSH) in the Met Office Unified Model (MetUM). Observational estimates are contrasted with climate simulations to evaluate the model’s ability to capture this important climatological feature. MetUM exhibits robust systematic biases in its representation of the WPSH, including a weakening of the anticyclone and a location too far east, which leads to an underestimation of the southwesterly monsoon flow over East Asia and contribute to seasonal precipitation errors in the area. To understand the origin and development of the systematic errors, we use a collection of model simulations of varying complexity, including climate integrations, long-range NWP hindcasts and a balance model, the semigeotriptic (SGT) model. With these tools we show that most of the circulation errors in the WPSH are corrected when tropical convection occurs in the right location. We then examine how MetUM’s deep convection biases in the region arise and find a large drying of the boundary layer by the convection scheme that is balanced mainly by local surface fluxes. In places with low exchange coefficient (places with light surface winds), the surface fluxes are not able to support deep convection over a long time period and the convection error is established.

In the DIMOSIC (DIfferent MOdels, Same Initial Conditions) project, forecasts from different global medium-range forecast models have been created based on the same initial conditions. The dataset consists of 10-day deterministic forecasts from seven models and includes 122 forecast dates spanning one calendar year. All forecasts are initialized from the same ECMWF operational analyses to minimize the differences due to initialization. All models are run at or near their respective operational resolutions to explore similarities and differences between operational global forecast models. This study aims to address the questions: ''How much difference does the choice of model formulation make for the forecast quality?” and “Can one identify models that have similar forecast characteristics?''. The presentation will highlight differences in forecast skill and biases between the models, and also explores the forecast differences between the models, e.g in terms of multi-model ensemble spread.

Many global models have substantial biases in their predictions of the Asian monsoon. In the Met Office Unified Model, for example, the monsoon trough is more zonally oriented than is observed which leads to underpredictions of summer rainfall over southern and eastern Asia. These circulation and precipitation errors have persisted over many cycles of research-to-operations, and appear robust to significant developments of all major parametrizations in the model. Here, we address a simple question: why are these biases systematic? That is, why have they not been removed by optimization of parameters in the model's phyiscs? Using a Perturbed Parameter Ensemble of AMIP simulations, we show that a strong constraint exists which prevents the Unified Model from simultaneously producing an unbiased monsoon and unbiased global top-of-atmosphere radiation fluxes. We use this constraint to define a scalar parameter, the "systemicity" of the ensemble, the magnitude of which measures the conflict between the constraints and therefore how "untunable" the model is. We identify the drivers of this parameter, and show that it is related to an inability to independently affect the properties of tropical and extra-tropical clouds. We suggest ways in which it could be reduced in future model versions.

The Arctic has warmed substantially over the last decades and will continue to do so owing to the ongoing global warming in conjunction with polar amplification. The changing mean state poses many challenges to the design, evaluation and calibration of subseasonal-to-seasonal forecasting systems. Furthermore, any inconsistencies between observed and reforecast trends degrade the forecast skill and point to deficiencies in either the physical modelling or the initialization methods. Here, we assess the consistency of boreal winter trends of surface air temperature (SAT) in the Eurasian Arctic between the ERA5 reanalysis and ECMWF sub-seasonal reforecasts initialised from ERA5, for the 20-year period 2000-2019. We present methods to quantify robustness and importance of the reanalysis trends and to test and quantify inconsistency of reforecast trends with the trends in the reanalysis. We find that the reforecasts clearly underestimate the reanalsyis warming trend of about 2 K per decade at lead times beyond two weeks. By week six, the reforecast trend is less than half of the reanalysis trend, with very high statistical significance. We perform a series of numerical experiments to investigate potential reasons for the trend underestimation. These concern the sea-ice thermodynamic coupling to the atmosphere, the snow scheme over land, cloud radiative properties, and possible remote influences from the North Atlantic and the Tropics. The outcome of these experiments provides guidance as to which improvements are needed in the physical forecast model and data assimilation methods in order to faithfully represent climate variability and change in subseasonal-to-seasonal forecasting systems.