Novel Approaches for Regionalising SWAT Parameters Based on Machine Learning Clustering for Estimating Streamflow in Ungauged Basins

Streamflow prediction in ungauged basins (PUB) is necessary for effective water resource management, flood assessment, and hydraulic engineering design. Spain is one of the countries in Europe expected to suffer the most from the consequences of climate change, notably an increase in flooding. The authors selected the Miño River basin in the northwest of Spain, which covers an area of 2,168 km², to develop a novel approach for predicting streamflow in ungauged basins. This study presents a regionalisation of the soil and water assessment tool (SWAT), a semi-distributed, physically based hydrological model. The regionalisation approach transfers SWAT model parameters based on hydrological similarities between gauged and ungauged subbasins. The authors used k-means and expectation−maximisation (EM) machine learning clustering techniques to group 30 subbasins (9 gauged subbasins) into homogeneous, physical, similarity-based clusters. Furthermore, the regionalisation featured physiographic attributes (basin area, elevation, and channel length and slope) and climatic information (precipitation and temperature) for each subbasin. For each homogeneous group, the SWAT model was calibrated and validated for the gauged basins (donor basins), and the calibrated parameters were transferred to the pseudo-ungauged basins (receptor basins) for streamflow prediction. The results of the streamflow prediction in the pseudo-ungauged basins demonstrate satisfactory performance in most of the cases, with average NSE, R², RSR, and RMSE values of 0.78, 0.91, 0.42, and 5.10 m³/s, respectively. The results contribute to water planning and management and flood estimation in the studied region and similar areas.