Research Projects

Principal Investigator: Hannah Cloke 

Postdoctoral Research Scientist: Hamidreza Mosaffa 

This project aims to develop a novel hydrology-informed foundation model for land surface hydrology that leverages the power of machine learning (ML) and artificial intelligence (AI), and to integrate it into ECMWF’s Earth System modelling approach. The project will address the challenges and limitations of existing Earth System models in representing land surface hydrology and its feedbacks to the atmosphere particularly at the extended range, and in providing accurate and reliable predictions of hydrological conditions and extremes such as floods, which can support decision making and risk management in various sectors. The foundation model will be constrained using core principles of hydrological scientific understanding such as the closure of the water balance, and will be tested using explainable AI methods to ensure hydrological consistency. Fine tuned models will then be developed for the downstream tasks of (a) improving land surface-atmosphere feedbacks at extended range/S2S scales and (b) forecasting floods. This project brings together different existing research collaborations at ECMWF and will use ECMWF’s portfolio of earth system datasets and archived experiments, as well as targeted new experiments. It will also exploit synergies with major European initiatives in this field at ECMWF such as DestinE and Copernicus EMS. The project is designed to anticipate the long-term horizon and next steps for the ECMWF strategy: taking the Earth System to machine learning. 

Project Aim: to advance the understanding of Earth System hydrological predictability and improve Earth System forecasting through the development of a hydrology informed foundation model for land surface hydrology.  

Research Objectives: Research objectives are divided into Core objectives and flexible stretch objectives to allow the project to have direction but also allowing it to evolve without constraining it too much in this fast-paced field.  

1. Develop and assemble the tools, information and datasets that are required to develop a hydrology informed foundation model of land surface hydrology 
2. Develop and evaluate a global foundation model and test for conceptual & scientific soundness using explainable AI techniques 
3. Develop and evaluate fine-tuned models for Land surface – atmosphere feedbacks at the extended range/S2S 
4. Develop and evaluate fine-tuned models for forecasting floods at different spatial scales and leadtimes
5. Explore the potential of using ECFoundLand to improve process representation, parameterisation and optimisation in ECLand 6. Explore the potential to add a land surface hydrology foundation model to larger foundation model in weather and climate)

Principal Investigator: Sarah Dance 

Postdoctoral Research Scientist: Rishabh Bhatt 

In the future, global numerical weather prediction (NWP) systems will run at kilometre-scale resolutions, allowing better representation of orography and treatment of convection. To fully realise these improvements, forecasts need to be initialised with observation information on appropriate scales. High-resolution satellite data are available. However, only 5-10% of these data are currently assimilated, largely due to a lack of knowledge about spatial and temporal observation uncertainties and the difficulties of taking them into account in NWP systems. Typically, observations are thinned to separation lengths where observation errors may be assumed uncorrelated, with the advantage of reducing the computational cost. Idealised studies have shown that representing correlated observation error covariances in the assimilation algorithm leads to significant improvements in analysis accuracy and forecast skill. However, there are open questions about the computational feasibility of these approaches for operational systems. Thus, the grand challenge addressed by this project is to improve NWP skill by enabling the assimilation of dense observation datasets using new, numerically efficient approaches to take into account the spatial and temporal structures of observation uncertainty. We will begin with research to provide a proof-of-concept using AMSU-A observations, and, once proven, extend this to other observation types. The first step is to estimate observation uncertainty structures, using assimilation residuals and metrological approaches. The next step is to develop novel numerical methods to implement these. 

Principal Investigator: John Methven 

Postdoctoral Research Scientist: Hannah Croad 

The S2S Challenge 

The sub-seasonal to seasonal (S2S) range extends beyond the limit of predictability for synoptic scale weather systems with the loss of predictability stemming from chaos (sensitivity to initial conditions). This is manifest in phase errors in the location and timing of weather systems. However, there are components of the atmosphere with longer range predictability despite the chaotic behaviour; across the S2S range coupling with the land surface, ocean and cryosphere become increasingly important. There is a pressing need to improve predictive skill on the S2S range to the level where it can be useful for decision-making in the food, water and energy sectors. The goal of this project is to identify the atmospheric components that are inherently more predictable, unpick their dynamics, understand why they are more predictable, and use this knowledge to improve the representation of those phenomena in prediction systems.   

 

Extended range predictability tied to wave-like phenomena and teleconnections 

Mid-latitude regimes central to S2S prediction are characterized by large-scale patterns with internal variability over timescales longer than individual cyclones. Three key examples include: 

1. Teleconnections from the tropics to the extratropics are large scale: many mechanisms have been proposed but are represented with varying success by S2S systems.   
2. Unprecedented multi-day extreme rainfall events have occurred across Europe in the last two decades (de Leeuw et al, 2016) and they have been linked to troughs in quasi-stationary Rossby waves, while the ridges are related to heatwaves (2003, 2023). However, free running models tend to underestimate such multi-day events for reasons unknown.  
3. Mid-latitude blocking is associated with longer timescales and persistent weather extremes and yet blocking remains one of the greatest challenges for prediction (Woollings, 2018). 

A common feature of these phenomena is persistence and the working hypothesis here is that this stems from long intrinsic dynamical timescales which are the basis of longer-range predictability.  

In this project, we will extract dynamical modes of variability from multi-level global data using the Empirical Normal Mode (ENM) technique (Brunet, 1994) in which each mode has an intrinsic frequency given by its spatial structure (just as a bell has a tone dictated by its shape). The approach enables us in principle to diagnose predictable components of variability and the (stochastic) forcing of those components (Brunet & Methven, WMO S2S Book). Recent developments by the team have opened new opportunities and this project will harness them. 

The proposed work aims to provide the underpinning for long-term development (>5 years ahead) of S2S prediction systems by finding out what modes of the atmosphere are predictable, why this is and how current prediction models represent the structure and time-dependence of those modes. This information will be used to determine why current systems achieve weak S2S signals and how best to improve S2S prediction systems including ensemble design. 

Principal Investigator: Chris Holloway 

Postdoctoral Research Scientist: Mark Muetzelfeldt 

Changes in anvil cloud with warming are one of the largest current sources of uncertainty in climate sensitivity (Sherwood et al 2020). Anvil cloud (including latent heating and radiative effects) also plays a key role in large-scale tropical convective organisation such as the MJO (Benedict et al 2020), the ITCZ (Talib et al 2018), and monsoons (Cetrone and Houze 2009), as well as mesoscale organisation including tropical cyclones (Ruppert et al 2020) and mesoscale convective systems (MCSs, Yuan et al 2011). While Cloud-system Resolving Models (CRMs, ~1-5-km grid spacing with explicit convection) improve the location and timing of deep convection, they often have biases in updraft strength, cloud amount (including anvil life cycle) and organisation. Observations to constrain these biases are not routinely available and therefore are the primary objective of the satellite mission INvestigation of Convective UpdraftS (INCUS), which is led by project partner Sue van den Heever at Colorado State University. INCUS will provide the first tropics-wide investigation of the evolution of convective updrafts and mass flux from its launch date in August 2026 (Kim et al 2023, Prasanth et al 2023).  

Anvil clouds are directly linked to the detrainment of moisture and condensate from deep convective updrafts. These mid-to-upper level stratiform clouds and their source updrafts involve a complex interplay between radiative effects, microphysical properties, latent heating, and convective dynamics. During their life cycle, anvil clouds transition from more surface-cooling thick clouds to more surface-warming thin clouds, and modelling studies show that changing the representation of any of the physical processes above can have a significant effect on cloud evolution and radiative effects (Gasparini et al 2019). Traditionally, km-scale models are compared to rainfall observations in order to improve short-term forecasts of severe weather and precipitation, while coarse-grid climate models are tuned to achieve a realistic top-of-atmosphere radiative balance. However, as we move to global km-scale model simulations we increasingly need to evaluate and improve the key processes that link cloud microphysics, radiative processes, precipitation, latent heating, convective dynamics and the large-scale circulation. We need to capture variability as well as the mean state, cloud as well as precipitation, average events as well as extremes, and we need to do this from cloud scales to global scales (Holloway et al 2014). To do this, we need to evaluate global km-scale models with observations and process models that target the full range of relevant processes across these different scales.  

In UPFLO, we will combine regional and global km-scale modelling (and sub-km process modelling) with data from observational field campaigns (including the current WesCon UK campaign on turbulent processes and convection, Barrett et al. 2021; CAMP2Ex in the tropical West Pacific, Reid et al 2023; CINDY-DYNAMO in the tropical Indian Ocean, Gottschalck et al 2013; DCMEX in 2022 which sampled cloud microphysics and related properties in anvil clouds over New Mexico; and the upcoming 2024 ORCESTRA campaign in the tropical Atlantic) to investigate how convective updrafts interact with anvil cloud processes and larger scales. Pre-launch, INCUS research centres on understanding how the Ka-band radars on three planned SmallSats (all passing over the same location within minutes) will be able to measure cloud properties including updraft velocities (Prasanth et al 2023). Specific WesCon case studies in which aircraft and ground-based radars sampled the same convective updrafts will provide valuable insight into this question and help develop INCUS retrieval algorithms.  

In addition to INCUS, data from the EarthCARE satellite mission (launching 2024 with cloudprofiling Doppler-capable radar to study clouds and aerosols) will also be utilised. In preparation for INCUS and EarthCARE, global km-scale ECMWF simulations and explicit convection global and regional simulations from the Met Office K-Scale project will provide large statistical samples of updrafts and related cloud properties with which to address our project objectives as well, and satellite simulators will be applied to model fields. Following launch, INCUS and EarthCARE retrievals will provide novel data sets to constrain updraft strength and cloud amounts in these simulations.  

Through comparisons of km-scale models to observations and process models, we will study model biases in updrafts, anvil clouds and associated radiation fluxes and their sensitivity to parameterisation choices involving key processes including microphysics and turbulent mixing. We will work closely with model developers to facilitate improvement of the representation of convective updrafts and anvil clouds 

Principal Investigator: Sue Grimmond 

Postdoctoral Research Scientist: Silvia Rognone 

URBANE will advance inclusion of urban environments in numerical weather prediction (NWP) models, notably better representing variables that are relevant to people and infrastructure, ultimately supporting forecasts and services. URBANE will exploit observations, including thermal remote sensing at multiple scales (facet, neighbourhood, city, multi-city, to cities globally), as part of constraining surface-energy exchanges.

URBANE is contributing to AFESP science plan Theme 2 Challenges and opportunities in simulating the Earth System at the kilometre-scale, but will also contribute to the other three (i.e., 1,3, and CC) over the 15-year AFESP period. Given the scale of cities and the move to higher NWP resolution[1], representation of cities is a timely challenge towards the service needs of city residents . ECMWF doesn’t yet represent cities operationally, but a simple urban tile is under testing for implementation in summer 2024, while other centres’ early urban representations need fuller evaluation to inform the next-generation urban schemes. Key to these developments is addressing the variability arising from urban form (e.g., building spacings and height) and function (e.g., anthropogenic heat distribution because of people’s behaviour)[2] ^.

Given the city focus of URBANE, the proposed research addresses challenges at the crossroads between seamless prediction[3] and delivering services crucial to human welfare[4]^,[5]. These challenges have been identified internationally as critical next research needs^2,3,[6]. There is a clear demand for services providing much greater detail than is currently operationally possible. The obstacles include lack of: model evaluation, data to evaluate the models, parameters to characterise cities at sub- neighbourhood scale, and sufficient understanding to improve model ‘physics’. Models must quantify the critical influences on surface exchanges impacting the myriad human activities from city operations, infrastructure, to citizen health and  socio-economic well-being.

The central contribution of URBANE will be to unlock new data and exploit them in urban NWP development, enabling model evaluation and model parameterisation across scales (sub-neighbourhood to global cities), boosting urban modelling to the next generation. URBANE’s potential impact will span operational centres and the international urban climate research community. The proposers have led two international urban land surface model comparisons[7] for which we gathered two and 20 neighbourhood-scale eddy covariance sites[8] , respectively. The 50 observation-years provided by the latter remains extremely sparse given the spatial extent and global population exposed to urban conditions. Additional data sets to be exploited include dense 3-d dataset being gathered in the ERC urbisphere[9] and NERC ASSURE projects, as well as satellite land surface temperature (LST) data. Each data set has potential advantages and disadvantages, for example LST is spatially extensive and very relevant to multiple surface energy exchange but a  challenges arise from the 3-dimensional nature of the urban surface[10] .

[1] Lean et al. 2019 The impact of spin-up and resolution on the representation of a clear convective boundary layer over London in order 100 m grid-length versions of the Met Office Unified Model https://doi.org/10.1002/qj.3519McNorton et al. 2021:An urban scheme for the ECMWF integrated forecasting system: Single-column and global offline application https://doi.org/10.1029/2020MS002375Sützl et al. 2021: Distributed urban drag parameterization for sub-kilometre scale numerical weather prediction https://doi.org/10.1002/qj.4162

[2]  Barlow et al. 2017: The integration of urban atmospheric processes across scales https://doi.org/10.1175/BAMS-D-17-0106.1

[3] Grimmond et al. 2015: Urban-scale environmental prediction systems. Seamless Prediction of the Earth-System: from minutes to months WMO-1156 https://library.wmo.int/records/item/54696-seamless-prediction-of-the-earth-system

[4] WMO 2019: Guidance on Integrated Urban Hydrometeorological, Climate and Environmental Services Vol. I: Concept and Methodology, WMO-No. 1234 https://library.wmo.int/doc_num.php?explnum_id=9903

[5] Grimmond et al. 2010: Climate & more sustainable cities: Climate information for improved planning & management of cities (producers/capabilities perspective) World Climate Conference – 3: better climate information for a better future, https://doi.org/10.1016/j.proenv.2010.09.016

[6] Urban Meteorology: Forecasting, Monitoring, and Meeting Users’ Needs, Board on Atmos. Sci. & Climate, National Academy of Sciences, https://nap.nationalacademies.org/login.php?record_id=13328

[7] Grimmond et al. 2010 The International Urban Energy Balance Models Comparison Project: First results from Phase 1 https://doi.org/10.1175/2010JAMC2354.1Grimmond et al. 2011: Initial results from Phase 2 of the International urban energy balance comparison project https://doi.org/10.1002/joc.2227 2011Lipson et al. 2024: Evaluation of 30 urban land surface models in the Urban-PLUMBER project: Phase 1 results https://doi.org/10.1002/qj.4589

[8] Lipson et al. 2022 Harmonized gap-filled datasets from 20 urban flux tower sites https://doi.org/10.5194/essd-14-5157-2022

[9] Fenner et al. 2024: urbisphere-Berlin campaign: Investigating multi-scale urban impacts on the atmospheric boundary layer https://doi.org/10.1175/BAMS-D-23-0030.1

[10] Morrison et al. 2023: Simulating satellite urban land surface temperatures: sensitivity to sensor view angle and assumed landscape complexity https://doi.org/10.1016/j.rse.2023.113579Hall et al. 2024: Utility of thermal remote sensing for evaluation of a high-resolution weather model in a city https://doi.org/10.1016/j.rse.2023.113579 Hall et al. 2024: Utility of thermal remote sensing for evaluation of a high-resolution weather model in a city https://doi.org/10.1002/qj.4669 

 

Principal Investigator: Anne Verhoef 

Postdoctoral Research Scientist: Rajseskhar Kandala 

High-fidelity global-scale monitoring and modelling of the spatio-temporal variability of land surface processes and state variables, and their interactions and feedback with the atmosphere, is of crucial importance for reliable Numerical Weather Prediction (NWP) and climate modelling. In this context, key land surface state variables (LSSV) are the land surface temperature ( (LST, a complex mix of vegetation and soil surface temperatures), the near surface and rootzone soil moisture content, soil temperature, as well as snow albedo and snow density. These land surface state variables play a pivotal role in the energy-, water and carbon balance, as well as in the strength of Land-Atmosphere coupling, and how this is affected by soil moisture and soil heat storage ‘memories’. The LSSV, and the land surface characteristics (vegetation- and soil-related properties) and processes that determine them, play a strong role in the predictability of near surface atmospheric state variables (NSASV), such as atmospheric temperature and relative humidity. Despite the implementation of a steady stream of improvements in the ECMWF Integrated Forecasting System (IFS) in recent years (Sandu et al., 2020; Boussetta et al., 2021) there are still considerable and persistent biases in LSSV and NSASV, especially for specific regions, seasons and forecast ranges (beyond 2 weeks).  

 

Motivation and broad scientific approach: (i) The achievement of a significant reduction of NSASV biases, at a range of spatial and temporal scales, is an important motivation for the proposed research, for the benefit of Society through improved predictions and services. This requires maximum exploitation of the growing range of observations at ever-increasing spatio-temporal resolutions, including data obtained from sensors installed on airborne platforms (such as satellites) as well as in-situ data (from quality-controlled observatories). Continued and pro-active Earth Observation-driven model system development (Balsamo et al. 2018) of the coupled IFS Land Data Assimilation (DA) system (LDAS) and the ECLand model (Boussetta et al., 2021), will allow ECMWF and its collaborators to be prepared when new data from planned satellite missions become available. These kinds of developments form a central aspect of the ECMWF coupled land-atmosphere DA strategy (de Rosnay et al. 2022).  

1. ii) A related motive is the notion that there is a strong lack of realism in the representation of soil and vegetation in Land Models (LMs), including in ECLand, which will impede the predictability of LSSV and NSASV. Vegetation parameters, such as leaf area index (LAI) and vegetation albedo are prescribed, and even if the CO2 assimilation flux informs vegetation growth, the interactions between soil and vegetation are absent or poorly implemented. Moreover, important (interactions between) below-ground processes are missing (e.g. coupled heat- and water flow; Zeng et al., 2011). Additionally, soil parameters are static, defined once at model configuration stage (derived from globally distributed soil maps and empirical pedotransfer functions (PTFs), both of variable quality). PTFs link readily available soil properties (e.g. soil texture) to soil physical parameters (i.e. hydraulic parameters that determine water flow and storage) and thermal parameters (those that determine heat flow and storage), as well as to biogeochemical parameters for carbon/nutrient cycles (Van Looy et al., 2017). The choice and quality of the soil map and PTF leads to large uncertainties and affects model skill. Instead, here we propose the timely and ambitious implementation of a fully dynamic soil-vegetation system in the IFS (via novel developments of LDAS and ECLand), whereby the soil will be considered as a temporally variable medium that can be ‘monitored’ from above. Within the IFS we will enable interaction between soil model parameters, their environmental and anthropogenic drivers, and soil and plant processes, by leveraging the new Multi-Parameter Regionalisation framework (MPR) available from the IFS cycle 49r1. Via a combination of EO and in-situ data, novel DA and mechanistic model equations we will obtain the below-ground calibrated model properties. We will largely focus on the dynamics of the soil structure; here the pore-size distribution (PSD) is crucial. It is largely the PSD, and its spatio-temporal variability, that determines the hydraulic and thermal properties of soil. Soil scientists and agronomists already increasingly view the structure of the soil system and its properties as temporally variable. Efforts are underway to reflect this notion in LMs (Fatichi et al., 2020), which we will build on via expert elicitation and literature surveys. Soil physical properties vary on sub-monthly to seasonal timescales due to land use and management activities, such as tillage and livestock trampling; freeze-thaw effects; vegetation growth (e.g. the beneficial effect of biopores on soil hydraulic properties, and hence on infiltration, percolation, and groundwater recharge) and fire: Finally, impacts of environmental change (e.g. melting of permafrost, and reduction of soil organic matter due to increased soil respiration) will also affect the soil hydro-thermal properties; these effects are important for climate-scale modelling (see Robinson et al., 2019).  Key to these proposed developments (a dynamic soil-vegetation system with improved process descriptions, while making optimal use of EO and in-situ data via DA) is the need for a paradigm shift in the way we currently treat soil hydraulic and thermal theory in LMs. This requires a unifying soil hydro-thermal theory, whereby changes to soil structure (and related PSD) affect both hydraulic and thermal properties, with soil matric potential (ΨsΨs ) as the independent variable (Luo et al, 2022), rather than soil moisture content (θθ). The rationale is that mounting evidence suggests that accurate modelling (and monitoring) of the soil-plant hydraulic continuum, includingΨsΨs-based plant water stress functions, will lead to improved prediction of LSSV and fluxes (Verhoef & Egea, 2014; Sabot et al., 2020; Wang et al., 2021). Regulations of root zone hydraulic properties, plant water status and transpiration can be reliably predicted by theory of the soil-plant hydraulic continuum, considering both above- and below-ground hydraulic traits as well as phenological and physiological parameters. (iii) In this context, we will explore the use of a  ‘vegetation as a root-zone soil sensor’ (VaaSS) approach  for spatio-temporal derivation of subsurface properties, using satellite observables that capture key aspects of the soil-plant continuum, and related water-energy and carbon exchanges, across near (NIR), thermal (TIR) and shortwave (SWIR) infrared, and microwave domains (MW) for determination of vegetation optical depth (VOD), vegetation water content (VWC), and Solar-Induced Fluorescence (SIF) (e.g. Konings et al., 2019, 2021). Fig. 1 provides a summary of the proposed approach. This diagram concerns the STEMMUS-SCOPE model, that combines a state-of-the-art soil physics model (STEMMUS: Zeng et al., 2013; Yu et al., 2018) with the SCOPE ‘Soil Canopy Observation’ model (Van der Tol et al., 2009). The ground-breaking approach described above will be implemented, tested and honed first with the STEMMUS-SCOPE model before we incorporate it into the ECLand-DA system, and subject it to detailed evaluation.  

References Balsamo, G et al. (2018) Satellite and In Situ Observations for Advancing Global Earth Surface Modelling: A Review, doi: 10.3390/rs10122038; Boussetta S, Balsamo G… de Rosnay, P et al. (2021) ECLand: The ECMWF Land Surface Modelling System, doi: 10.3390/atmos12060723; Boussetta, S, Balsamo, G et al. (2013) Impact of a satellite-derived leaf area index monthly climatology in a global numerical weather prediction model, doi: 10.1080/01431161.2012.716543; de Rosnay, P…Balsamo, G, et al. (2022) Coupled data assimilation at ECMWF: current status, challenges and future developments, doi: 10.1002/qj.4330; de Rosnay, P et al. (2020) SMOS brightness temperature forward modelling and long term monitoring at ECMWF, doi: 10.1016/j.rse.2019. 111424; Fatichi, S et al. (2020) Soil structure is an important omission in Earth System Models, doi:10.1038/s41467-020-14411-z; Konings, A G et al. (2019). Macro to micro: microwave remote sensing of plant water content for physiology and ecology, doi:10.1111/nph.15808; Konings, A G et al., (2021) Detecting forest response to droughts with global observations of vegetation water content, doi:10.1111/GCB.15872;  Luo S. et al. (2022) Soil water potential: A historical perspective and recent breakthroughs, 10.1002/vzj2.20203; Robinson, D A et al. (2019) Global environmental changes impact soil hydraulic functions through biophysical feedbacks, doi:10.1111/GCB.14626; Sabot, M…Verhoef, A et al. (2020) Plant profit maximisation improves predictions of European forest responses to drought, doi: 10.1111/nph.16376; Sandu I, et al. (2020) Addressing near-surface forecast biases: outcomes of the ECMWF project ‘Understanding uncertainties in surface atmosphere exchange’, doi: 10.21957/wxjwsojvf; Van der Tol, C…Verhoef, A, Su, Z (2009). An integrated model of soil-canopy spectral radiances, photosynthesis, fluorescence, temperature and energy balance, doi: 10.5194/bg-6-3109-2009; Van Looy K.., Verhoef A et al. (2017) Pedotransfer functions in Earth system science: Challenges and perspectives, doi.org/10.1002/2017RG000581; Verhoef, A & Egea, G (2014) Modelling plant transpiration under limited soil water: comparison of different plant and soil hydraulic parameterizations and preliminary implications for their use in land surface models, doi:10.1016/ j.agrformet.2014.02.009; Wang, Y, Zeng, Y, et al. (2021). Integrated modelling of canopy photosynthesis, fluorescence, and the transfer of energy, mass, and momentum in the soil–plant–atmosphere continuum (STEMMUS–SCOPEv1.0.0), doi:10.5194/gmd-14-1379-2021; Yu, L, Zeng, Y, et al. (2018) Liquid-Vapor-Air Flow in the Frozen Soil, doi: 10.1029/2018jd028502; Zeng, Y et al. (2011) A simulation analysis of the advective effect on evaporation using a two-phase heat and mass flow model, doi:10.1029/2011WR010701; Zeng, Y & Su, Z (2013) STEMMUS: Simultaneous Transfer of Energy, Mass and Momentum in Unsaturated Soil. (ITC-WRS Report). University of Twente, Faculty of Geo-Information and Earth Observation (ITC), Enschede, The Netherlands, pp. 6161–6164.