Workshop on ancillary data for land surface and Earth system modelling

The evolution of the ECMWF system especially with higher resolution developments, is continuously pushing the land surface model to consider better representation of local processes and application-based outputs. With these objectives, it is necessary to seek increased quality in space and time of the ancillary fields that constrain the model and their related/derived parameters. The presentation would introduce the ECMWF land surface system and its currently used ancillary data set ranging from land-sea mask, physiographic fields, vegetation indices, to soil related hydrological parameters. it would also introduce the perspectives for these ancillaries considering higher resolution, time variability and uncertainties.

Lakes modify the structure of the atmospheric boundary layer. They can have a significant impact on local climate (over 1°K difference in 2-meter temperature) and on local weather (up to 10°K difference in 2-meter temperature).
At the European Centre for Medium-Range Weather Forecasts (ECMWF), lake parametrization was introduced in 2015. Inland water bodies (lakes, reservoirs, rivers and coastal waters) are simulated by the Fresh-water Lake model Flake. The IFS model is used for global weather forecast production from medium to seasonal range, and for reanalysis (e.g. ERA5) generation.
The current lake mask used in IFS is constant over time and represents permanent water over the 34-year period (i.e. 1984-2018). It has been shown that monthly varying lake mask has a significant positive impact on regions with prolonged rain and dry seasons, especially in Malaysia, Indonesia and Papua New Guinea (see Kimpson et al., 2023).
The CERISE project has outlined requirements for the boundary conditions of the IFS model, which will be used for generating high-resolution reanalysis from the present back to the year 1925, i.e. ERA6. One of the main objectives is the generation of high-resolution monthly varying lake mask.
Global seasonally varying water distribution maps were generated based on high horizontal (30m) and temporal (month) resolution satellite data for the past 50 years and some high-fidelity auxiliary data (e.g. coastline shapefiles, elevation datasets). The main data input is from the Joint Research Centre (JRC) Global Surface Water Explorer (GSWE) dataset (Pekel et al., 2016) at 30 m resolution (EPSG:4326) covering the period 1984-2021: ‘monthlyHistory’ – monthly water distribution with water/ notWater/ noData data, to further customise water classes based on a relevant period (e.g. 2012-2021). In addition, are used Copernicus GLO30 dataset at 30 m resolution (EPSG:4326) ‘waterBodyMask’ to separate the ocean from inland water; and numerous regional glacier datasets at 15-100 m resolution (i.e. British Antarctic Survey, QUANTARCTICA, GIMP project, QGREENLAND, Norwegian Institute, Icelandic Met service) to improve water distribution over relevant region.
Global seasonally varying water distribution maps generated for 1992-2021 are fully independent and are purely based on satellite data. Earlier (1962-1991) maps have in general the 1992-2001 period as a baseline and are updated only regionally - based on available reliable satellite information or historic records (i.e. maps, verbal description) with supplementary elevation data criteria.
Generated maps are grouped per 10-year period, each period has one permanent water map and twelve monthly maps (i.e. permanent water + monthly delta). The first available period is 1962-1972 due to booming water-related anthropogenic activities, i.e. building of large reservoirs and irrigation channels, reverting rivers, etc. It is assumed that for the earlier periods (1925-1961) maps for 1962-1971 can be used (due to decreased availability of situ data and its quality to make any assumptions/ calculations or verification). All available periods: 1962-1971, 1972-1981, 1982-1991, 1992-2001, 2002-2011, 2012-2021.
The presentation will outline global seasonally varying water distribution maps and the methodology used to generate them, map comparison with other datasets, as well as first modelling results using these maps.

The Community Land Model is the land surface scheme of several Earth System Models, including CESM, E3SM, NorESM and CMCC. The latest release v5.3, contains numerous updates to the ancillary fields, including transient land-use timeseries going back to 1700, newly updated crop distribution scenarios, soil color maps and other drivers that are modified to be consistent with TRENDY protocols by default. Newer capabilities using the FATES demographic mode also utilize gross land use transitions directly. Further, hydrological ‘hillslope’ tiling requires topographical fields. Other legacy requirements/capabilities include use of spatially varying soil depth, fire suppression from population density, lightening strikes, elevation and slope, timing of agricultural fires, irrigation and fertilization.

Glacier albedo can be highly variable due to impurities and even algae growth. In addition, the snow grain size and shape affects the albedo of snow-covered glaciers. In reanalysis this can be addressed by using satellite-derived glacier albedo. For the CARRA and CARRA2 projects we have used MODIS data from the Terra satellite and OLCI data from the Sentinel-3 satellite with 500 m resolution. For CARRA2, we have additionally used AVHHR data for the 1980s and 1990s in 5 km resolution. We allow fresh snowfall from the reanalysis model to change the albedo, since neither of the satellites can see through precipitating clouds. We demonstrate the major impact of using satellite-derived albedo.

Great progresses have been made in land surface model developments in recent decades. More processes are incorporated to enhance model functions, especially to illustrate different mechanisms of land-atmosphere interaction. Observation-based data are essential for land model developments in order to control uncertainties introduced by new processes. Multiple sources of data, such as meteorological forcing and land surface properties provides model with necessary boundary conditions. Global and site scale data, including energy, hydrology, biogeochemical and anthropogenic activity variables are used to verify the performance of Common Land Model version 2024 (CoLM2024). Impact of uncertainties in meteorological forcing on CoLM2024 has been evaluated. Precipitation has the highest degree of uncertainty at 4.4%. The forcing uncertainties propagate to the model simulations and cause significant differences in simulated variables. Runoff uncertainty is about 15.7% globally, with a greater impact in low latitudes. Meanwhile, the Open Source Land Surface Model Benchmarking System (Openbench) have been used to evaluate the performance of different models or versions against plot-scale observations and globally gridded reference data for over 100 variables such as latent heat, sensible heat, net radiation, and total runoff. Over 50 metrics are included to quantify CoLM2024 performance in different dimensions. The benchmark tools grounded the land surface model development on a basis of data. The data-driven development of land surface model help reduce the uncertainties propagation in earth system models.

Geospatial datasets are pivotal for numerical weather prediction (NWP) and climate modeling, providing critical information on orography, land use, soil properties, vegetation, and radiative factors such as aerosols. Rapid integration of new, high-resolution data sources into models is essential to keep pace with advancements in satellite technology and increasing model grid resolutions.
The geospatial data preprocessor EXTPAR has been unified and integrated as community software within the COSMO consortium. This robust and flexible tool supports the COSMO and ICON model grids in both NWP and climate applications, enabling streamlined processing and quality assurance of geospatial datasets.
This presentation highlights the role of geospatial datasets in improving forecast accuracy in NWP and explores the outcomes of EXTPAR's integration as a shared resource. Specific examples will demonstrate how consistent and efficient geospatial data handling enhances model performance and facilitates the rapid adoption of new data sources. Future priorities include expanding dataset coverage and ensuring the tool remains adaptable to evolving scientific and operational needs.

The Ensemble of Data Assimilations (EDA) at ECMWF suffers from under-dispersed surface fields, this can limit the effectiveness of land-surface data assimilation. In the current system identical climatological ancillary fields for many land-surface parameters are provided across all members of the EDA. To increase the ensemble spread at the surface we have experimented with the perturbation of two of the land parameters provided in the climatological fields, namely vegetation fraction and leaf area index. The perturbations are generated following the Stochastically Perturbed Parameterisations (SPP) methodology, adding spatiotemporally correlated noise.

Initially these perturbation strategies were tested in an offline ensemble of land models (forced atmosphere with no feedback). The land model ensembles were run for a 2 year spin-up and a full year’s integration with and without the inclusion of perturbations in vegetation fraction and leaf area index across the ensemble. These experiments showed promising increases in spread (~20%) in a variety of surface fields such as 2-metre temperature, dew-point, soil moisture and even variables such as snow depth where the change in bare soil fraction of the grid cell allowed for more/less melting of snow.

After these initial "offline" results, coupled EDA experiments were run including the most promising perturbations in the surface parameters. These full EDA experiments are slow to spin-up but show broadly similar results to the offline experiments for the land-surface variables. In addition, running in the coupled system enables a more thorough analysis of the effect on the ensemble spread and forecast impact for atmospheric variables and several diagnostics will be shown.

Land surface temperature (LST) temporal and spatial variability carries a blueprint of the surface energy budget. The temporal variability of LST is mostly dominated by the diurnal cycle of available energy, modulated by cloud presence, and synoptic variability. Averaged over a few weeks, LST filters the cloud and synoptic variability, remaining the diurnal cycle. On these time scales, the daily maximum of LST spatial variability will be mostly dominated by the land surface conditions (e.g. elevation, land cover, vegetation state, soil moisture state, etc.). Therefore, the LST spatial variability of the daily maximum should reflect the different sources of land surface variability, making it an optimal candidate for evaluating the realism of storm-resolving models in representing such surface heterogeneities. LST is also one of the few land surface variables that can be directly derived from satellite data, and there are several available datasets which provide high quality and long data records. In this study, a newly reprocessed GEO-ring Land Surface Temperature (LST) dataset, spanning the period from 2018 to near real-time, with an hourly temporal frequency and 5 km spatial resolution, is presented. This product is employed to evaluate the impact of updated land information in the ECMWF IFS NextGEMS simulations, comparing results from cycle 2 to cycle 3 and to investigate the realism of the NextGEMS storm-resolving models from a land surface perspective.

A keystone of European Space Agency’s (ESA) climate activities is the Climate Change Initiative (CCI), which has been running for more than a decade and is led by the ESA Division on Climate and Long-Term Action. This unique scientific effort, improving our understanding of climate science and the quantification of its key processes, generates global multi-mission and multi-decadal datasets satisfying the requirements for more than 25 Essential Climate Variables (ECVs) defined by the Global Climate Observing System (GCOS). These datasets have fully characterised uncertainties and are validated using independent, traceable, in-situ measurements. They provide an impartial yardstick to understand climate processes and to evaluate, improve and validate climate models, thereby enhancing the quality, credibility, and exploitation of their predictions. CCI data also provide the observational record to study drivers, interactions and feedback due to climate change, with a particular focus on tipping points and the global energy, water and carbon cycles, etc. A particular focus of CCI is to develop data and methodology in support to the reporting requirements under the UNFCCC Paris Agreement, supporting regional and/or national mitigation and adaptation activities,
Through ESA’s close interaction with stakeholders and users and active participation in the relevant entities in the international climate landscape, ESA supports operational capabilities in the international decision and policy making process, policy implementation monitoring. This in turn supports activities of ESA’s members states, for example under the UNFCCC Paris Agreement, addressing national and regional requirements as formulated in their Nationally Determined Contributions and National Adaptation Plans.
The ESA Division on Climate and Long-Term Action works closely with operational climate services: to date more than 20 ECVs have been transferred from CCI to operational production by the Copernicus Climate Change Service (C3S). ESA continues to support the development of these ECVs through ongoing CCI research projects.

The Copernicus Land Monitoring Service (CLMS), jointly implemented by the European Environment Agency (EEA) and the Joint Research Center (JRC), provides various near real times (NRT) products over the terrestrial surface that includes various global biophysical parameters since 2014. There are also new products such as land surface phenology and future ones such as evapotranspiration, land cover at higher resolution that can serve for land surface modeling.
This presentation summarizes the actual CLMS portfolio and its roadmap focusing on the relevant products to see if they can serve Land Surface Modeling purposes. The needs to increase spatial and temporal resolution, and in some cases timeliness, at global scale that can better serve the future of land surface modelling.

The EUMETSAT Satellite Application Facility (SAF) Network provides a robust and reliable framework for delivering high-quality satellite-derived data to support weather and climate applications. In the context of land surface modelling, several SAFs offer products that enhance our understanding of surface properties, hydrology, and energy balance processes, supporting both operational forecasting and climate applications.

This presentation will introduce the SAF Network and its role in delivering consistent, timely, and long-term data products. It will highlight key SAF contributions to land surface modelling, such as surface radiation and energy fluxes (from the LSA SAF), snow cover and soil moisture datasets (from the H SAF), and climate data records of land surface parameters. Furthermore, the session will outline strategic directions under the new SAF Strategy, aiming to enhance the impact and usability of SAF products, including future adaptations to evolving user needs and computational environments.

At Planet, we compute and deliver several Planetary variables: high-resolution indicators of crucial land processes measured from space. These Planetary Variables are processed at high resolution and are available globally, usually within 24 hours after the observation. Planetary Variables include:

Soil Water Content and Land Surface Temperature
We use passive microwave observations and our patented DifSat algorithm to estimate Soil Water Content and Land Surface Temperatures. Clouds are virtually transparent to microwaves, and we provide estimates every two days, regardless of cloudiness. DifSat downscales the coarse-resolution passive microwave observations by combining the information from overlapping footprints, which increases the spatial resolution and removes artifacts from water bodies. We provide our estimates at a 1 km resolution, which can be refined further using infrared imaging to 100 meters.

Crop Biomass and Vegetation Optical Depth
Using optical and microwave data, we provide insights into the status of vegetation: with Crop Biomass we provide insight into the amount of vegetation, while Vegetation Optical Depth is an indicator of vegetation moisture content and plant health.

Forest Carbon and canopy information
Using high-resolution optical images and various machine-learning techniques, we estimate the amount of carbon captured by forests, together with canopy heights and canopy cover.

During the workshop, we will show how Planetary Variables can be used to obtain soil and surface conditions in near real-time.

The Momentum Partnership brings together ten modelling centres who use the Unified Earth Environment Prediction Framework for a broad range of operational and research applications. Partners are drawn from institutions in both the mid-latitudes and the tropics. At the core of the system is the Unified Model (UM) coupled to the JULES land surface model. Applications range from earth system and climate modelling, through numerical weather prediction at global and convection-permitting scales, to hectometric modelling of urban areas. Seamlessness is an important feature of the framework.

Across the modelling community, the continuing development of ever more highly resolved models requires corresponding increases in the accuracy and detail of ancillary data. This is particularly so in the case of urban areas, which are of especial importance in the context of climate change and increased urbanization. Improved forecasting of environmental hazards necessitates the coupling of different component models of the environmental system and imposes further constraints on the quality of ancillary data, the development of more integrated hydrological forecasting being one example of this.

In this presentation, I will review the current status of ancillary data within the modelling framework and illustrate current and future challenges with examples drawn from across the partnership.

A few years ago the HIRLAM countries took the initiative to apply the second generation of ECOCLIMAP (ECOSG) SURFEX physiography for our operational NWP setups. ECOSG land cover is based on the ESA-CCI global land cover at 300 m resolution. ECOSG came with a few benefits, like e.g. better land/water masks and improved Leaf-Area Index product, but also with one very clear drawback.

The Irish agricultural landscape is characterised by fields surrounded by quite narrow stretches of trees. This landscape gives a substantial effective roughness length but this character of the landscape is not reflected in the ESA-CCI land cover. The drawback of this, in combination with SURFEX roughness formulations, is a substantial underestimation of the roughness length and consequently a positive bias in 10 m wind speed. Our cure to this problem was to introduce what we named fake trees for the low-vegetation covers, which means that we complement the low vegetation with trees. 10 m wind speed scores improved substantially over low-vegetation dominated areas. Our question in this Workshop context is, how should we deal with inconsistencies between how a physically based

These problems with ESA-CCI land cover, in combination with needs for even higher resolution of land cover, inspired new approaches where Machine-Learning methods are now utilized to create a European ML-version of ECOSG at 60 m resolution. A description and a status of these efforts will be presented.

Numerical weather and climate prediction at Météo-France is based on the ARPEGE global model, the AROME model for forecasting in metropolitan France and overseas, and the Méso-NH research model for more specific studies. These models cover a wide range of spatial and temporal scales. The spatial resolution of the global ARPEGE numerical weather and climate prediction models is a few tens of kilometers, while limited area models such as AROME or Méso-NH represent finer-scale phenomena. A better representation of boundary layer processes, turbulence and interactions, including surface-atmosphere coupling, is a major challenge for mesoscale research at Météo-France. There is a growing need to develop operational and research models with hectometric resolution to meet societal needs in response to meteorological, hydrogeological or air quality extremes.
This range of models and the trend towards higher resolution means that we need to define as precisely as possible, i.e. with a resolution appropriate to the models, the ancillary data from which certain parameters are derived and which allow us to better characterize the impact of the surface on dynamics and physics. For example, this will allow us to better characterize orographic drag, surface friction, the effect of surface heterogeneity on the coupling with the atmosphere and the partitioning of surface fluxes as a function of leaf area index, interactions with radiation via albedos, and so on. It is also important to be able to update these auxiliary data over time to improve model realism. In most cases, such changes must be accompanied by adjustments to surface parameterizations or those at the surface-atmosphere interface.
The presentation will focus on the practices used at Météo-France to define ancillary data for operational and climate models, as well as for higher resolution research applications. Since the 2000s, Météo-France has developed the global, kilometer-scale ECOCLIMAP database, which is linked to the SURFEX modeling platform and used by all Météo-France forecast models, and has upgraded it to improve its spatial and temporal resolution. The methods and practices currently used to generate high-resolution ancillary data will be presented, along with their advantages and disadvantages. In addition, work using machine learning to represent land cover in overseas regions for which Météo-France is responsible will be described.

The Copernicus Climate Change Service (C3S) plays a crucial role in providing high-quality, satellite-derived data for comprehensive climate monitoring across all Earth System domains. The terrestrial domain is of particular significance, as it is home to the majority of the human population and includes key Essential Climate Variables (ECVs) such as soil moisture, land surface temperature, snow cover, and terrestrial water storage. These variables are vital for understanding climate variability, land-atmosphere interactions, and ecosystem dynamics.

The C3S Climate Data Store (CDS) offers access to valuable climate information spanning the past, present, and future. Of particular interest to this workshop are the quality assured Climate Data Records (CDRs) derived from satellite ECVs, alongside reanalysis products. Both provide long-term, spatially consistent datasets. While satellite data offer more direct observations of land variables, reanalysis products help fill observational gaps and provide continuous datasets for global and regional climate assessments. Both types of datasets are invaluable for a wide range of applications, including serving as ancillary information to constrain land surface models and validating advancements in land modelling.

This presentation will provide an overview of the C3S datasets relevant for Earth system modeling, and will showcase some applications of these data. Additionally, it seeks to engage the Earth system modeling community to help shape the future evolution of C3S services.

..

The model’s mean orography acts as the boundary condition for the model dynamics and the drag from resolved orographic gravity waves can have a significant impact on the large-scale atmospheric circulation in weather and climate models. As we approach km-scale horizontal resolutions in global models, more of the orographic spectrum and the impact of orography becomes resolved. However, even at kilometre scale, some of the orographic variance will not be represented on the model grid and must be parameterised.

High resolution simulations for Destination Earth’s global Digital Twin at 4.4 km have shown benefits of increased resolution, for example better prediction of orographic rain, but also increased error over Eastern Asia, indicating that the higher resolution of resolved orography causes additional small horizontal-scale orographic gravity waves breaking above the mid-latitude jet, which affects the global circulation.

Therefore, we focused on improving the mean orography processing and sub-grid scale orography parameterisations with the aim to find a scale-independent formulation that maintains good forecast skill at operational model resolutions and profits from increased resolution at kilometre scale. Changes to the orographic ancillary fields include a new source dataset for surface elevation with 30 m resolution; simplified and harmonised processing across resolutions using conservative interpolation from the source dataset; an updated spectral filtering of mean orography; updates to the computation of the sub-grid orography fields; and optimising parameters orographic parameterisation schemes for the updated fields.

This poster provides an overview of ECMWF’s current ancillary data sets.

Understanding the role of land surface physics and biogeochemistry is essential for advancing climate models and weather prediction, particularly in relation to long-term variability, local feedbacks, and extreme events. Accurate boundary conditions—such as land cover (LC) and land use (LU)—play a key role in enhancing the realism of climate simulations by improving the representation of land-atmosphere interactions that regulate surface energy balance and ecosystem processes. Additionally, they provide the foundation for a more realistic depiction of the terrestrial carbon cycle, including vegetation dynamics and soil biogeochemistry.

The CERISE project aims to generate high-resolution (1 km) LC and Leaf Area Index (LAI) datasets spanning 1925–2020, contributing to next-generation reanalysis datasets (e.g., ERA6-Land) and seasonal forecasts (e.g., SEAS6). In its initial phase, we reconstructed historical LU and LAI using machine learning (ML) models to downscale coarse-resolution LU datasets (LUH2f, HILDA+). Our workflow integrates multiple ML techniques, including Random Forest and XGBoost, to train models on high-resolution LC and LAI satellite observations while actively exploring ways to enhance both performance and interpretability. To infer monthly LAI variations from annual LU inputs, we developed an auxiliary network that captures intra-annual variability. Initial results demonstrate promising performance in reconstructing LC and LAI across various test years and regions, underscoring the feasibility and robustness of this ML-based approach for historical reconstructions.

Future phases, including the CONCERTO and TerraDT projects, will build upon this work by generating consistent high-resolution LU datasets for both historical (1850–present) and future scenarios (present–2100), supporting CMIP7 climate simulations and scenario-based studies. These efforts will incorporate additional auxiliary data (e.g., elevation, soil types, climate indices) to refine feature representation and develop autoregressive models that account for long-term temporal dependencies and dynamic changes. Ultimately, our goal is to develop a robust ML-based emulator capable of producing scalable, high-resolution land surface boundary conditions to support digital twin applications, thereby enhancing climate simulation and prediction capabilities.