A catchment-scale analysis of the impact of land data assimilation on surface fluxes and river discharge with a land surface model

Hydrological modelling plays a crucial role in Earth System Models at many scales, affecting results in numerical weather predictions through surface fluxes and storages. Data assimilation is used in weather models to ascertain accurate initial conditions for weather forecasts based on observations. At the land surface, the land data assimilation system (LDAS) exploits observational data to update selected fields of the state vector, such as soil moisture or snow cover. In turn, soil moisture and snow cover affect the terrestrial water cycle through the partitioning of runoff components, and consequently streamflow.

Whereas LDAS configuration choices are usually driven by improvements in meteorology, here we aim to investigate how they impact streamflow. To this end, several LDAS configuration are used to run ECMWF’s Land Surface Modelling System (ecLand), which generates grid-wise runoff routed as streamflow in rivers by the Catchment-based Macro-scale Floodplain (CaMa-Flood) hydrological model. Two experiments, one with snow analysis and the other with soil moisture analysis, are compared against the baseline without any data assimilation over 1990-2023 using meteorological forcing from ERA5. In addition, a baseline experiment without any data assimilation is run. Therefore, differences in the model output of these experiments can be attributed to the data assimilation procedure. Snow cover analysis uses ESA-CCI data from 1990 to 2010, and IMS snow cover data from 2010 to 2023. The soil moisture analysis assimilates ERS-SCAT products from 1992 to 2006 and ASCAT soil moisture products from 2007 to 2023, as well as gridded SYNOP observations of 2-m temperature and relative humidity.

Streamflow is the main diagnostic variable to quantify the impact of LDAS on river hydrology, as observational data is available. A filtering procedure was applied to the Global Runoff Data Centre (GRDC) dataset to ensure sufficient observations to represent local climatology: at least 25 daily values per month, for seasonal representation, and a minimum of 19 years over the 33-year period. Monthly climatologies of simulated discharge and surface fluxes are calculated to identify catchment-scale patterns. Surface fluxes, such as evapotranspiration and (sub-)surface runoff, and streamflow output of the respective experiments are compared for 342 gauged catchments, across 209 distinct river systems in the northern hemisphere.

The Kling–Gupta Efficiency and its components are computed for each catchment, allowing the assessment of bias, correlation, and variability separately. Assessments show a widespread impact of LDAS configuration on the hydrological skill of ecLand-Cama-Flood system: for 48% of the stations the baseline experiment has a higher hydrological skill, while 20% of the stations benefit from snow cover analysis and 32% from soil moisture analysis. Then, to assess hydrological processes across horizontal and vertical spatial scales, performance of data assimilation is analysed as function of catchment characteristics such as upstream drainage area, orography, and spatial variability of orography.

This study pinpoints catchments where river hydrology benefits or is negatively impacted by land data assimilation and directly supports further development of ecLand.