Using machine learning for long-term calibration and validation of water quality ecosystem service models in data-scarce regions.

Water quality ecosystem service (ES) modeling tools help inform freshwater management across landscapes. However, the validity of such models depends on the availability of water quality data for validation and calibration, limiting their application in regions where monitoring is limited. This study presents a methodological framework that combines machine learning (ML) and spatial extrapolation to enhance ES modeling in data-scarce contexts (https://github.com/LSU-EPG/Puerto-Rico-ES-Project/tree/main/Data_Scarcity_Framework). Focusing on Puerto Rico, we leverage ML to reconstruct temporal gaps in nutrient trends. We then use ML to apply this reconstructed data to automate calibration and validation of nutrient retention ES models. We transfer validated parameters to unmonitored catchments using hydrogeological similarity. Our results demonstrate that ML preserves critical patterns in nutrient dynamics, while using calibrated parameters across hydrologically similar basins yields accurate predictions in ungauged watersheds. Our framework enhances ES model scalability, offering a tool to inform evidence-based water quality management in data-limited regions.