Contribution of EO Large-sample hydrology data and multi-model approach in enhancing model stability and accuracy in arid and semi-arid regions

Water security is crucial for achieving most Sustainable Development Goals, especially health, food production, energy, and climate resilience. Given the many links between water and other SDGs, focusing efforts on achieving the water goal would inevitably facilitate the achievement of the rest. From this perspective, it is feasible for Southern Mediterranean countries, including Morocco, to partially achieve the SDGs by adopting integrated water management systems, of which hydrological modeling is an important part. However, most modeling tools and their structure often show inconsistencies in application from one basin to another, which can be explained by two factors: first, the insufficient and inaccurate input data, and second, the inherent artifacts in the model structure as well as its incompatibility with the characteristics of the basin. In this work, we propose a modeling scheme that seeks to solve the two problems related to data scarcity and model insufficiency in arid and semi-arid regions. We used a multi-source data approach combined with a multi-model approach to forecast water flow in a set of twenty sub-catchments of the Oum Er-Rbia River Basin in central Morocco, we mainly calibrated, validated and tested the model sets parameters as well as their performance behavior. This modeling exercise will lead to a comprehensive understanding of the model transferability, stability, and adaptability according to its application catchment. Our analysis of model’s performances and outputs reveals spatial and temporal variation in the prediction results of each model, where the set of models was divided according to accuracy, stability, and adaptability into high-performance models along the study field (MOPEX3/2, topmodel, hymod, GR4J, and HBV), medium-performance models (sacramento, newzeland1/2, xinanjiang, and mcrm), and failed models (MOPEX1, tank/2, and collie1). The proposed modeling sceme not only enhanced the predictive skills in the study area, but it’s also formed the basis for investigating the characteristics of the targeted catchment and thus facilitated the process of selecting the most appropriate model for each basin. Additionally, the remotely sensed data products helped to solve the problem of data scarcity in poorly or ungauged basins.