The STARS4Water project recently released a report on “Improving existing models based on multi-scale model integration“.
This report presents the results related to Task 3.3 (WP3) of the STARS4Water project. This task focuses on multi-scale model integration through the benchmarking of land surface and hydrological models. The primary objective of this task is to evaluate the ability of different hydrological models to simulate key components of the water cycle, including river discharge, soil moisture, evapotranspiration, snow water equivalent (SWE), and water table depth (WTD). By systematically comparing multiple modelling approaches against observed datasets, this task aims to improve the accuracy and applicability of hydrological models, ensuring they can effectively support decision-making in water resource management. The motivation behind this work arises from the need for reliable, high-resolution hydrological simulations that can help address climate change impacts, increasing water demand, and the complexities of managing freshwater systems across diverse hydroclimatic regions.
The benchmarking assesses the performance of seven models—LISFLOOD (Europe and Drammen), CLM, ParFlow-CLM, TSMP, and wflow_sbm (Europe and Rhine)—applied at different spatial and temporal scales. These models employ varying approaches, from conceptual parameterized structures to fully physically based simulations, and are evaluated against in-situ and remotely sensed hydrological observations. The results indicate that no single model excels across all hydrological variables and regions, highlighting the importance of model intercomparison. The benchmarking also reveals significant spatial variability in model accuracy, with better performance in regions with high-quality observational datasets and lower skill in areas with complex hydrological conditions, such as snow-dominated basins or regions with heterogeneous groundwater systems.
A key outcome of this study is the recognition of the critical role that robust simulated and observed datasets play in hydrological modelling. Observed data are essential for validating models and ensuring their reliability, while high-quality simulations offer valuable insights into hydrological processes, especially in regions where in-situ measurements are limited. The combination of these datasets enhances the ability to evaluate water cycle components more comprehensively, supporting the development of more accurate forecasting systems and sustainable water management strategies.
Additionally, this task explores how simulated datasets can be leveraged to construct machine learning models for hydrological applications. The study demonstrates how data-driven techniques such as Long Short-Term Memory (LSTM) networks and Random Forest (RF) can exploit outputs from physically based models to refine estimations of water table depth anomalies, SWE, and groundwater storage changes, in the Seine, Drammen and Duero River Basin, respectively.
Overall, the results of Task 3.3 highlight the value of benchmarking in improving hydrological models and the necessity of integrating multiple datasets to enhance water resource assessments. Future work will focus on refining model calibration and further validating simulated outputs against observational records.
More information can be found in the following STARS4Water project Deliverable:
Aguilera, H., Gomez, V., Martinez, P., Preiml, M., Glas, M., Weerts, A., Ramos, M-H., de Lavene, A., Engeland, K., Gelati, E., Hisdal, H., Hegdahl, T., Kollet, S., Avila, L., Keller, V., Rickards, N., Kossida, M., Salamon, P., Grimaldi, S., Bisselink, B. (2025): Improving existing models based on multi-scale model integration. Horizon Europe project STARS4Water. Deliverable D3.3.
