Assessing the Impact of Geological Map Detail on Process-Based and Data-Driven Hydrological Models

Abstract

Although large-sample hydrology data sets are increasingly used to advance predictions in ungauged basins, the influence of landscape data quality on model regionalization remains insufficiently explored. This study investigates whether geological catchment attributes derived from maps of increasing detail—global, continental, and regional—improve parameter transfer and model regionalization. To ensure robustness across model approaches, we applied both a semi-distributed process-based hydrological model using hydrological response units (HRUs) and a data-driven Long Short-Term Memory (LSTM) model. The analysis covered a total of 130 catchments in the Moselle (Central Europe) and Garonne (southwestern France) basins. We conducted five model experiments differing only in the representation of geological information: a benchmark without geology, a benchmark with random geology classes, and configurations based on the global-, continental-, and regional-scale geological maps. Model performance was evaluated using a modified Nash-Sutcliffe (NSE) metric for daily streamflow, as well as Pearson correlation and relative bias for three streamflow signatures: baseflow index, slope of the flow duration curve, and half-flow date. Across both basins and modeling frameworks, increasing geological detail consistently improved predictive performance under space–time evaluation. While differences in NSE were modest, improvements were pronounced for streamflow signatures: only models using the more detailed geological information, especially the regional map, consistently reproduced spatial variability in baseflow and flow regime characteristics. These findings highlight the importance of integrating high-quality geological data into hydrological modeling, particularly for improving predictions in ungauged basins through more reliable parameter transfer and regionalization.