Abstract. A stand-alone emulator of ECMWF's land surface scheme (ecLand) has been developed. This emulator, aiLand, uses a multi-layer perceptron architecture, chosen for its balance of accuracy and efficiency and for its differentiability, which is crucial for integration into data assimilation and parameter estimation systems. In this study, we introduce a two-stage learning framework that leverages both synthetic land surface model simulations and real-world observations. We first pretrain the surrogate on extensive ecLand outputs to capture the core dynamical behaviour of key land surface states, evaluating its accuracy, long-term stability, and transferability across variables, depths, and climates. We then fine-tune the pretrained model on in situ eddy-covariance flux observations for selected diagnostic variables, validating against independent flux-tower sites. The pretrained emulator reproduces ecLand's prognostic soil state with a 90-day RMSE of 1.19 K for surface soil temperature and 0.014 m3 m-3 for surface soil moisture, and remains stable over continuous 4-year autoregressive integrations. A single globally trained model outperforms biome-specialist baselines in cross-biome transfer, with residual errors concentrating in snow-insulated cold biomes where an insulating snowpack decouples the soil from atmospheric forcing. Fine-tuning on FLUXNET observations reduces latent heat flux RMSE by 30 % and sensible heat flux by 20 % at validation sites, while preserving prognostic state integrity and improving the physical consistency of the surface energy budget: per-site Bowen ratio error drops by 42 % and energy balance closure residuals fall from 13.4 W m2m-2 to 3 W m-2 and below. These results demonstrate that combining physics-based pretraining with observation-based fine-tuning provides a flexible pathway for building accurate, stable, and differentiable land surface emulators suitable for data assimilation and coupled modelling applications.

Abstract. The representation of river routing and floodplain dynamics remains a key limitation in many global Earth system and numerical weather prediction models, despite their importance for hydrological extremes, water resources, and land–atmosphere interactions. This paper presents the implementation of a global hydrodynamic river routing capability within the European Centre for Medium-Range Weather Forecasts (ECMWF) Integrated Forecasting System (IFS) through the integration of the Catchment-based Macro-scale Floodplain (CaMa-Flood) model into the ecLand land surface scheme. The system enables the routine simulation of river discharge, water level, and inundation extent within a physically consistent Earth system modelling framework. We describe the scientific design and technical developments required to transition from research to operational implementation, including coupling strategies, state initialisation, parallelisation using a basin-based hybrid MPI/OpenMP approach, and integration within ECMWF's Research-to-Operations workflow. The implementation maintains consistency across multiple configurations, from offline experiments to fully coupled forecasts, while preserving operational constraints such as computational efficiency and forecast stability. The added value of the integrated system is demonstrated through a range of applications, including global water resources monitoring contributing to the World Meteorological Organisation State of Global Water Resources Report, real-time ensemble flood forecasting, high-resolution regional simulations, and land-surface diagnostics. Results illustrate that river discharge provides a powerful integrative constraint on the land water cycle and supports the identification of model deficiencies in hydrological processes. The system also enables the generation of continuous, global, high-frequency hydrological datasets suitable for both operational applications and emerging data-driven modelling approaches. This work establishes a new capability for global hydrological prediction within an operational Earth system model and provides a framework for future developments toward fully coupled atmosphere–land–river interactions.

Earth observations from satellites are the primary information source for Numerical Weather Prediction models. While some land surface variables, such as surface soil moisture, are assimilated to improve initial land conditions, the use of additional satellite-derived land surface and vegetation products remains limited partly due to systematic model biases. Here we examine the potential of satellite-derived land surface temperature and vegetation indicators to enhance near-surface temperature forecast skill. We build deep learning surrogate models for Numerical Weather Prediction using Long Short-Term Memory networks. Results show that including these satellite datasets improves temperature forecast skill globally across lead times from 1 to 12 days, with the largest improvements at 4-day lead time. Satellite-based predictors are the most relevant variables in about 60% of global grid cells. Among them, sun-induced fluorescence is the most important predictor, reflecting vegetation photosynthetic activity and its influence on surface energy partitioning and near-surface temperature.

Soil moisture plays a critical role in water-limited regions through its strong coupling and feedbacks with vegetation. However, state-of-the-art Land Surface Models (LSMs) used in reanalysis and near-term prediction systems still lack a realistic coupling of vegetation, limiting their ability to properly account for the fundamental role of vegetation in modulating the feedback with soil–moisture.In this study, we incorporate Leaf Area Index (LAI) variability from observations - derived from the latest-generation satellite products provided by the Copernicus Land Monitoring Service - into three different LSMs. The models perform a coordinated set of offline, land-only simulations forced by hourly atmospheric fields from the ERA5 reanalysis. An experiment using interannually varying LAI (SENS) is compared with a control simulation based on climatological LAI (CTRL) in order to quantify vegetation feedbacks and their impact on simulated near-surface soil moisture.Our results show that interannually varying LAI substantially affects near-surface soil moisture anomalies across all three models and over the same water-limited regions. However, the response differs markedly among models. Compared with ESA-CCI observations, near-surface soil moisture anomalies significantly improve in one model (HTESSEL–LPJ-GUESS), whereas the other two models (ECLand and ISBA–CTRIP) exhibit a significant degradation in anomaly correlation. The improved performance in HTESSEL–LPJ-GUESS is attributed to the activation of a positive soil moisture–vegetation feedback enabled by its effective vegetation cover (EVC) parameterization. In HTESSEL–LPJ-GUESS, EVC varies dynamically with LAI following an exponential relationship constrained by satellite observations. Enhanced (reduced) soil moisture limitation during dry (wet) periods leads to negative (positive) LAI and EVC anomalies, which in turn generate a dominant positive feedback on near-surface soil moisture by increasing (decreasing) bare-soil exposure to direct evaporation from the surface. In contrast, ECLand and ISBA–CTRIP prescribe EVC as a fixed parameter that does not respond to LAI variability, preventing the activation of this positive feedback. In these models, the only active feedback on near-surface soil moisture anomalies is negative and arises from reduced (enhanced) transpiration associated with negative (positive) LAI anomalies.Our findings demonstrate that simply prescribing observed vegetation properties in LSMs does not guarantee a realistic coupling between vegetation and soil moisture. Instead, it is shown that the explicit representation of the underlying vegetation processes is essential to activate the proper feedback and capture the correct soil moisture response.

Global streamflow modelling is crucial, as it underlies our capacity to forecast riverine floods able to devastate infrastructure and ecosystems, and adversely affect human lives. To that end, the hydrodynamic Catchment-based Macro-scale Floodplain model (CaMa-Flood) has been included in ECMWF’s Land Surface Modelling System (ecLand), and consequently in the Integrated Forecasting System (IFS). Precipitation, which is partitioned into infiltration and runoff by ecLand’s land surface processes, is eventually converted into streamflow through CaMa-Flood. This means streamflow carries an imprint of both meteorology and land surface processes. This is particularly relevant, as runoff is not available as an observation, while streamflow is, making the latter a key variable for the aggregated evaluation of modelled land surface processes. As this analysis is done under the auspices of the Destination Earth project, it presents the additional possibility to evaluate land surface processes across all operational spatial scales up to the km-scale and temporal scales from daily to hourly. For example, through running daily CaMa-Flood simulations driven with runoff forcing from the control ensemble member and from the Continuous-Extremes Digital Twin (C-EDT), the land surface’s hydrological processes can be evaluated at the spatial resolutions of ~9km and ~4.4km, respectively. Comparing results from these daily simulations with streamflow observations, we found that the C-EDT generates insufficient surface runoff in orographic regions. This stems from the sub-grid runoff parameterization in ecLand, which generates less surface runoff at higher resolutions for the same amount of precipitation, and is therefore not scale-adaptive.Beyond providing hydrological simulations, CaMa-Flood is used in this study as a diagnostic tool for hydrological processes to guide future development of ecLand. More specifically, we have implemented scale-specific orographic parameters in the model’s runoff-generating algorithm, aiming to provide consistent orographic surface runoff generation across spatial scales. Runoff partitioning is important for flood extremes on timescales of a few days, because it directly modulates the magnitude of the flood peak. In addition, it affects the soil moisture, and consequently sub-surface runoff and streamflow on timescales of months to years. Therefore, the efficacy of these adaptations is tested with both long-term land surface experiments with ecLand/CaMa-Flood and fully-coupled 5-day meteorological forecasts with IFS/CaMa-Flood at spatial resolutions of ~29km, ~9km and ~4.4km. For the forecasts on shorter time scales, we assess the flood peak magnitude, timing and durations errors. For the long-term integrations from 1990 – 2025, streamflow time series allow a robust evaluation of the simulations against the observations, and its components as a measure of goodness-of-fit. Moving beyond the KGE, a cross-spectral analysis is applied to evaluate the time signature of the hydrological processes undelrying streamflow and occurring at different time scales, which is especially useful considering the partitioning between surface (fast) and sub-surface (slow) runoff. Through addressing these scale-dependent issues, ample surface runoff generation is ensured, allowing the river hydrology simulated by CaMa-Flood to benefit fully from running meteorology and the land surface at the km-scale.

Glaciers and ice sheets are critical components of the cryosphere and the climate system. In a warming climate, surface temperatures exceed the melting point more frequently and for longer periods, making ice and snowpacks increasingly susceptible to melting. Meltwater from glaciers and ice sheets contributes to freshwater inputs to oceans and rivers, while the ice and snow surfaces provide a cooling effect on the atmosphere. This study presents a new parameterisation enabling a more realistic representation of glaciers and ice sheets in a global land surface model used for numerical weather prediction and reanalyses, accounting for the seasonal evolution of the snowpack and the fractional glacier coverage within grid points. The new scheme has been tested in stand-alone (offline) mode across various scales, from point-level to regional and global simulations, and validated against in situ observations and a range of reference datasets. Results show improvements in the representation of surface temperature, albedo, and snow processes compared to the current scheme, leading to a more accurate simulation of melting events and surface mass balance for the Greenland Ice Sheet. The impact of the new scheme on the hydrological cycle was also assessed for glacier-fed river basins, for which the increased melting generally leads to higher river discharge. It is shown that it improves simulations for basins with an underestimation of their streamflow generation during summer or early spring. However, this effect can be detrimental in basins where the current model already overestimates the discharge, further amplifying the positive bias. Overall, the new scheme enables a more accurate and physically realistic representation of glacier processes, enhancing the representation of cryosphere surfaces of future climate reanalyses for a wide range of scientific applications.

While forecasting of climate and earth system processes has long been a task for numerical models, the rapid development of deep learning applications has recently brought forth competitive AI systems for weather prediction. Earth system models (ESMs), even though being an integral part of numerical weather prediction have not yet caught that same attention. ESMs forecast water, carbon and energy fluxes and in the coupling with an atmospheric model, provide boundary and initial conditions. We set up a comparison of different deep learning approaches for improving short-term forecasts of land surface and ecosystem states on a regional scale. Using simulations from the numerical model and combining them with observations, we will partially emulate an existing land surface scheme, conduct a probabilistic forecasts of core ecosystem processes and determine forecast horizons for all variables.

Lakes modify the structure of the atmospheric boundary layer. They can intensify winter snowstorms, increase/decrease surface temperature and amount of precipitation. It has been shown that monthly varying lake surface cover has a significant positive impact over regions with prolong rain and dry seasons, especially over Malaysia, Indonesia and Papua New Guinea (see Kimpson et al., 2023).At European Centre for Medium-Range Weather Forecasts (ECMWF) current lake mask is constant over time and represent permanent water over the period 1984-2018. To meet reanalysis requirements of monthly varying high-resolution lake mask outlined in CERISE project the Joint Research Centre (JRC) Global Surface Water Explorer (GSWE) dataset (Pekel et al., 2016) was used. Applied methodology, its advantages and drawbacks, as well as first results of monthly lake surface cover maps will be presented.

By 2030, over 300 million hectares worldwide will be irrigated, constituting the second most significant anthropogenic influence on land use following urbanisation. Our study focuses on an irrigated Terrestrial Environmental Observatories (TERENO)/Integrated Carbon Observation System (ICOS) site in Germany, unveiling irrigation's immediate effects on soil moisture, latent heat flux, skin and soil temperature. As we strive to seamlessly integrate irrigation processes into the ECMWF Integrated Forecasting System (IFS), our investigation extends to an offline model, ECLand, including dynamical vegetation. Introducing a perturbed precipitation field offers a refined perspective of mimicking irrigation. The feedback provides us with insights into the coupling of simple irrigation representation on thermodynamic variables, ensuring optimal benefits for the IFS. After verification with remote sensing data, the next step involves coupling water fluxes to stomatal conductance via photosynthesis, shedding light on the preliminary influence of irrigation on enhanced vegetation growth. This aims to untangle irrigation effects of increased soil moisture and greening.

Vegetation is a relevant and highly dynamic component of the Earth system and its variability – at seasonal, interannual, decadal and longer timescales – modulates the coupling with the atmosphere by affecting surface variables such as roughness, albedo and evapotranspiration. In this study, we investigate the effects of improved representation of vegetation dynamics on climate predictability and prediction at the seasonal timescale. To this aim, the observational constraints from the latest generation satellite dataset of vegetation Leaf Area Index (LAI) have been integrated in the modeling, including a parameterization of the effective vegetation cover as a function of LAI. The improved vegetation representation is implemented in HTESSEL, which is the land surface model included in the seasonal forecasting (ECMWF SEAS5) systems used in this work. Our results show that the realistic representation of vegetation variability has significant effects on both potential predictability and actual prediction skill at the seasonal time scale. It is shown a significant improvement of the skill in predicting boreal winter (December-January-February; DJF) 2m Temperature (T2M) at 1-month lead time especially over Euro-Asian boreal forests; the improvement is at least in part due to the more realistic representation of the interannual albedo variability that is related to the changes in vegetation shading over snow. Remarkably, from the region with the most considerable T2M improvement originates a large-scale ameliorating effect on circulation encompassing Northern Hemisphere middle-to-high latitudes from Siberia to the North Atlantic. The results indicate that the coupling with the improved vegetation might operate by amplifying locally the signal originating from the North Atlantic sector, therefore improving both potential predictability and actual skill over the region. Concurrently, the improved predictability and skill over the Euro-Asian forests appears to feedback to the large-scale circulation enhancing the representation of the circulation pattern and associated interannual anomalies.

Greenhouse gas emissions, greenhouse gas concentrations and global mean temperature all continue to rise, and in order to stay within the temperature limits stipulated in the text of the Paris Agreement, mitigation action is becoming increasingly urgent However, the fact that we cannot quantitatively and reliably predict future GHG concentrations – and therefore climate scenarios – from assumed future emission pathways is a complicating factor when designing mitigation action. Even more problematic is the assessment the impact or effectiveness of many current or proposed mitigation activities, since it often has to be based on indirect measures such as avoided emissions with respect to a hypothetical baseline, or carbon stored, e.g. in the land or ocean biosphere, neither of which can be directly linked to atmospheric concentrations.In order to provide robust, actionable data that will help Parties to the UNFCCC and other stakeholder design and develop mitigation action and monitor its effectiveness, the World Meteorological Congress in May 2023 endorsed the Global Greenhouse Gas Watch (G3W) as an internationally coordinated framework to provide near-real time GHG (CO2, CH4 and N2O) flux estimates based on atmospheric modelling and atmospheric observations. At COP28 in Dubai, the G3W was formally recognized by the Subsidiary Body for Scientific and Technological Advice (SBSTA-59) to the UNFCCC.Currently a G3W implementation plan is in development, with the aim of submitting it for approval by the WMO Executive Council by mid-2024. Some of the key elements of the plan are a significant strengthening of the global GHG observing capabilities, improved near-real time exchange of both observational data and flux estimates, and routine intercomparision of model output among all participating flux estimation centers.The presentation will introduce the overall G3W development timeline which aims for a full operational capability to be ready for the Second Global Stocktake in 2027-28, with the main focus on the near-time activities planned for 2024-25.

Vegetation plays a crucial role in the land surface water and energy balance modulating the interactions and feedback with climate at the regional to global scale. The availability of unprecedented Earth observation products covering recent decades (and extended up to real-time) are therefore of paramount importance to better represent the vegetation and its time evolution in the land surface models (LSMs) used for offline analysis/initialization and for the seasonal-to-decadal predictions. Here, we integrate realistic vegetation Leaf Area Index (LAI) variability from latest generation satellite campaigns, available through Copernicus Land Monitoring Service (CLMS), in three different LSMs that conducted the same coordinated set of offline land-only simulations forced by hourly atmospheric fields derived from the ERA5 atmospheric reanalysis. The experiment implementing realistic interannually-varying LAI (SENS) is compared with simulations utilizing a climatological LAI (CTRL) to quantify the vegetation feedback and the effects on the simulation of near-surface soil moisture.The results show that the inter-annually varying LAI considerably affects the simulation of near-surface soil moisture anomalies in all three models and over the same water-limited regions, but surprisingly the effects diverge among models: compared with ESA-CCI observations, the near-surface soil moisture anomalies significantly improve in one of the three LSMs (HTESSEL-LPJGuess) while the other two (ECLand and ISBA-CTRIP) display opposite effects with significant worsening of the anomaly correlation coefficients. It is found that the enhanced simulation of near-surface soil moisture is enabled by the positive feedback that is activated by the effective vegetation cover (EVC) parameterization, implemented only in HTESSEL-LPJGuess. The EVC parameterization works such that the effective fraction of the bare soil being covered by vegetation does vary with LAI following an exponential function constrained by available satellite observations. The increased (reduced) soil-moisture limitation during dry (wet) periods produces negative (positive) LAI and therefore EVC anomalies, which in turn generate a dominating positive feedback on the near-surface soil moisture of HTESSEL-LPJGuess by exposing more (less) bare soil to direct evaporation from the sub-surface layer. On the other hand, in the EC-Land and ISBA-CTRIP models, EVC is fixed in time as it cannot vary with LAI and so the positive feedback described cannot be activated. The only feedback on near-surface soil moisture anomalies that operates in these two models is negative and comes from the reduced (increased) transpiration related to the negative (positive) LAI anomalies.Simply prescribing observed vegetation data into LSMs does not guarantee the introduction of the correct coupling and feedback on climate. In this respect, this multi-model comparison experiment demonstrates the fundamental role of the inclusion of the underlying vegetation processes in LSMs. Ignoring the proper representation of the vegetation processes could lead to unrealistic (and even the opposite effects compared with observations) behaviour in reanalysis and climate predictions.

The Destination Earth (DestinE) initiative of the European Commission applies digital twin technology to the Earth system, enabling bespoke, high-resolution simulations of extreme weather events and climate scenarios. At its core, DestinE features Digital Twins and a Digital Twin Engine — a software framework that connects computing, Earth system models, data, and applications. DestinE’s Digital Twins combine observations, physics-based high-resolution simulations and emerging Artificial Intelligence (AI) methods, and leverage Europe’s most powerful supercomputers, through a strategic partnership with the European High Performance Computing Joint Undertaking (EuroHPC JU). DestinE enables tailored Earth system simulations to dynamically explore future weather and climate scenarios and to address “what-if” questions. It also establishes an operational framework for multi-decadal, multi-model climate projections, linking them to applications that transform vast climate data into actionable insights for climate sensitive sectors. By offering detailed, customized information on weather and climate extremes, DestinE enhances existing capabilities and supports both immediate and longer-term climate adaptation strategies.3 Key points of the paper:• Defines the Destination Earth digital twin concept• Demonstrates the implementation of the world’s first two digital twin prototypes for two distinct purposes, anticipating weather-induced extremes and supporting climate change adaptation and mitigation• Discusses the challenges of the first 2 years of Earth system digital twin technology capacity building, highlighting the future potential for supporting public institutions in their effort to respond and adapt to climate change and extreme events.

The Diurnal Land–Atmosphere Coupling Experiment (DICE) aims to explore the complex interactions between the land surface and atmospheric boundary layer, which are generally not well understood and difficult to isolate in models. The project involves over 10 different models, combining expertise from both land‐surface and atmospheric boundary‐layer modelling groups. A simple three‐stage methodology is designed to assess land–atmosphere feedbacks. Stage 1: the individual components are assessed in isolation, driven and evaluated against observational data; stage 2: the impact of coupling is investigated; stage 3: the sensitivity of the stand‐alone models to variations in driving data is explored. For this initial study, a 3‐day clear‐sky period in the mid‐west United States over, an assumed simple, predominantly grass surface was simulated using data from the CASES‐99 field campaign. Key conclusions from the study include: (1) the memory of vegetation state within land‐surface models needs attention; (2) the height of atmospheric forcing for land‐surface models is important, particularly for the nocturnal boundary layer, and this has implications for both observations and vertical resolution for atmospheric models; (3) land–atmosphere feedbacks reduce errors in simulated surface fluxes at the expense of the accuracy of the variables that the models are designed to simulate (e.g., temperature, humidity, and wind speed); (4) problems remain in representing the stable boundary layer in atmospheric models; (5) the mixing of temperature and humidity within the boundary layer may need to be represented separately; (6) differences in daytime profiles of heat, moisture, and momentum between models are mainly due to the way the models erode the inversion at the top of the boundary layer, rather than differences in the surface fluxes. Resultant variations in modelled boundary‐layer heights have a substantial impact on relative humidity and could partially explain variations in coupling strength between models in the Global Land–Atmosphere Coupling Experiment.

The Diurnal Land–Atmosphere Coupling Experiment (DICE) aims to explore the complex interactions between the land surface and atmospheric boundary layer, which are generally not well understood and difficult to isolate in models. The project involves over 10 different models, combining expertise from both land-surface and atmospheric boundary-layer modelling groups. A simple three-stage methodology is designed to assess land–atmosphere feedbacks. Stage 1: the individual components are assessed in isolation, driven and evaluated against observational data; stage 2: the impact of coupling is investigated; stage 3: the sensitivity of the stand-alone models to variations in driving data is explored. For this initial study, a 3-day clear-sky period in the mid-west United States over, an assumed simple, predominantly grass surface was simulated using data from the CASES-99 field campaign. Key conclusions from the study include: (1) the memory of vegetation state within land-surface models needs attention; (2) the height of atmospheric forcing for land-surface models is important, particularly for the nocturnal boundary layer, and this has implications for both observations and vertical resolution for atmospheric models; (3) land–atmosphere feedbacks reduce errors in simulated surface fluxes at the expense of the accuracy of the variables that the models are designed to simulate (e.g., temperature, humidity, and wind speed); (4) problems remain in representing the stable boundary layer in atmospheric models; (5) the mixing of temperature and humidity within the boundary layer may need to be represented separately; (6) differences in daytime profiles of heat, moisture, and momentum between models are mainly due to the way the models erode the inversion at the top of the boundary layer, rather than differences in the surface fluxes. Resultant variations in modelled boundary-layer heights have a substantial impact on relative humidity and could partially explain variations in coupling strength between models in the Global Land–Atmosphere Coupling Experiment.

River discharge has direct influence on the water-food-energy-environment nexus and can have devastating impacts during extreme events with rapid onsets such as floods. Floods often occur after extreme precipitation events, which are challenging to forecast accurately, both in time and space. Unresolved small-scale processes and features, including convection and orography, have a detrimental effect on precipitation and consequently hydrological forecast skill. This calls for a spatial resolution increase in Numerical Weather Prediction (NWP) models, including their land component. The Destination Earth programme of the European Commission addresses this with globally coupled forecasts at spatial resolutions down to the km-scale with lead times of 5 days: the Digital Twin on Weather-Induced Extremes (EDT). These meteorological forecasts are used to force ECMWF’s Land Surface Modelling System (ECLand), the land component of the Integrated Forecasting System (IFS), to generate runoff. Subsequently, the river-routing scheme CaMa-Flood, effectively 1-way coupled to the IFS, is used to route runoff in rivers and to produce hydrological simulations. Essentially, CaMa-Flood will be part of the continuous component of the EDT, which in phase 2 of Destination Earth will provide daily high-resolution forecasts to monitor extreme events, such as floods, in real time. As river discharge acts as a natural integrator of the water cycle, CaMa-Flood can be used as a diagnostic tool to assess the hydrological response to increases in spatial resolution of the forcing and the river-routing network. In this study, two data products are derived: i) long-term hydrological simulations forced by atmospheric analysis data (e.g. ERA5 or ECMWF operational analysis) and ii) hydrological forecasts (daily forecasts in June - July 2021 and January - February 2022 as well as selected flood cases). To assess their quality, these data are validated with point-observed river-discharge time series. Analysis shows that the long-term hydrological simulations benefit from spatial resolution increases in the meteorological forcing and to a lesser extent from spatial resolution increases in the river-routing network. This is evidenced by higher Kling-Gupta Efficiency (KGE), higher correlations and lower biases across 876 river stations in Europe. Further, hydrological forecasts also benefit from higher spatial resolution meteorological forcing, evidenced both by higher correlations of the continuous summer/winter forecasts against river discharge observations from 798 river stations across Europe and by more pronounced flood peak magnitude for selected flood cases. These results highlight the added value of high resolution for hydrological forecast accuracy.

The most useful weather prediction for the public is near the surface. The processes that are most relevant for near-surface weather prediction are also those that are most interactive and exhibit positive feedback or have key roles in energy partitioning. Land surface models (LSMs) consider these processes together with surface heterogeneity and, when coupled with an atmospheric model, provide boundary and initial conditions. They forecast water, carbon, and energy fluxes, which are an integral component of coupled atmospheric models. This numerical parametrization of atmospheric boundaries is computationally expensive, and statistical surrogate models are increasingly used to accelerate experimental research. We evaluated the efficiency of three surrogate models in simulating land surface processes for speeding up experimental research. Specifically, we compared the performance of a long short-term memory (LSTM) encoder–decoder network, extreme gradient boosting, and a feed-forward neural network within a physics-informed multi-objective framework. This framework emulates key prognostic states of the Integrated Forecasting System (IFS) land surface scheme of the European Centre for Medium-Range Weather Forecasts (ECMWF), ecLand, across continental and global scales. Our findings indicate that, while all models on average demonstrate high accuracy over the forecast period, the LSTM network excels in continental long-range predictions when carefully tuned, extreme gradient boosting (XGB) scores consistently high across tasks, and the multilayer perceptron (MLP) provides an excellent implementation time–accuracy trade-off. While their reliability is context-dependent, the runtime reductions achieved by the emulators in comparison to the full numerical models are significant, offering a faster alternative for conducting experiments on land surfaces.

Heat extremes have severe implications for human health, ecosystems, and the initiation of wildfires. While they are mostly introduced by atmospheric circulation patterns, the intensity of heat extremes is modulated by terrestrial evaporation associated with soil moisture availability. Thereby, ecosystems provide evaporative cooling through plant transpiration and soil evaporation, which can be reduced under water stress. While it has been shown that regional ecosystem water limitation is projected to increase in the future, the respective repercussions on heat extremes remain unclear. In this study, we use projections from 12 Earth system models to show that projected changes in heat extremes are amplified by increasing ecosystem water limitation in regions across the globe. We represent the ecosystem water limitation with the ecosystem limitation index (ELI) and quantify temperature extremes through the differences between the warm-season mean and maximum temperatures. We identify hotspot regions in tropical South America and across Canada and northern Eurasia where relatively strong trends towards increased ecosystem water limitation jointly occur with amplifying heat extremes. This correlation is governed by the magnitude of the ELI trends and the present-day ELI which denotes the land–atmosphere coupling strength determining the temperature sensitivity to evaporative cooling. Many regions where ecosystem functioning is predominantly energy-limited or transitional in the present climate exhibit strong trends towards increasing the water limitation and simultaneously experience the largest increases in heat extremes. Sensitivity of temperature excess trends to ELI trends is highest in water-limited regions, such that in these regions relatively small ELI trends can amount to drastic temperature excess trends. Therefore, considering the ecosystem's water limitation is key for assessing the intensity of future heat extremes and their corresponding impacts.

In the 2022 summer, western–central Europe and several other regions in the northern extratropics experienced substantial soil moisture deficits in the wake of precipitation shortages and elevated temperatures. Much of Europe has not witnessed a more severe soil drought since at least the mid-20th century, raising the question whether this is a manifestation of our warming climate. Here, we employ a well-established statistical approach to attribute the low 2022 summer soil moisture to human-induced climate change using observation-driven soil moisture estimates and climate models. We find that in western–central Europe, a June–August root zone soil moisture drought such as in 2022 is expected to occur once in 20 years in the present climate but would have occurred only about once per century during preindustrial times. The entire northern extratropics show an even stronger global warming imprint with a 20-fold soil drought probability increase or higher, but we note that the underlying uncertainty is large. Reasons are manifold but include the lack of direct soil moisture observations at the required spatiotemporal scales, the limitations of remotely sensed estimates, and the resulting need to simulate soil moisture with land surface models driven by meteorological data. Nevertheless, observation-based products indicate long-term declining summer soil moisture for both regions, and this tendency is likely fueled by regional warming, while no clear trends emerge for precipitation. Finally, our climate model analysis suggests that under 2 ∘C global warming, 2022-like soil drought conditions would become twice as likely for western–central Europe compared to today and would take place nearly every year across the northern extratropics.

We analyze the performance of the European Centre for Medium-Range Weather Forecasts (ECMWF) and UK Met Office (UKMO) meteorological models in predicting offshore blowing wind in coastal areas. Our attention is mainly on the Mediterranean coast, up to 200 km distance from the shore. We compare forecast neutral winds with Advanced Scatterometer measurements. The results indicate that the ECMWF forecasts systematically underestimate wind speed with respect to scatterometer data, while the UKMO model tends to overestimate. A cross-analysis suggests that, better than fetch, model biases are a function of the model horizontal discretization, hence of the number of grid steps the wind runs over the sea. The steepness and roughness of the land orography before entering the sea, together with the related drag parameters, appear to have a strong role in determining the coastal offshore wind values. Surface drag over land tends to reduce the related wind speed, and it takes a few grid points to adjust to the smooth sea surface conditions. This is supported by a detailed study with a high resolution grid, but different orography resolutions. In these simulations, practically identical coastal wind speed distributions are found when the subgrid orography schemes are switched off. Vertical cross-sections of potential temperature and wind in bora and mistral conditions illustrate the strong role of gravity waves, wind channeling and turbulent diffusion in coastal numerical weather prediction. Besides local wave modeling, this may be relevant for offshore wind farms, typically situated within 5-50 km from the coast. Practical experience shows that European Centre for Medium-Range Weather Forecasts coastal wind speeds, particularly when blowing from land to sea, are often underestimated. We have quantified this error by comparing model data with satellite measured ones. The results indicate that surface wind speeds progressively increase the further we go from the coast, reaching the correct values at about 200 km offshore. Analysis of the inner land distribution and related orographic characteristics strongly suggests that the underestimate is associated to the roughness of the surface the winds pass across before reaching the coast. This roughness is strongly dependent on the model resolution. The passage of the air flow over steep orography before reaching the coast leads to gravity waves, that is, vertical oscillations of the main flow, that strongly affect the distribution of the surface coastal winds and the time, and space, required to reach the correct values. Meteorological model often underestimates offshore blowing wind speeds. We quantify and explain the reasons for it Rather than distance, the number of grid steps off the coast, about 10, is crucial in reaching offshore the correct wind speeds Coastal orography is crucial in determining the underestimate at the coast, with also gravity waves determining the fetch evolution