A multi-scale framework for BOD5 prediction from water quality to hydro-geomorphic interpolation

This study develops and validates a hydro-geomorphic interpolation model (HyRIM) to predict biochemical oxygen demand over 5 days (BOD₅) concentrations at unmonitored locations and help optimize monitoring networks. The research first establishes the significant spatiotemporal heterogeneity of BOD₅ drivers in the Haraz River Basin, Iran, revealing a distinct regime shift where physical parameters control BOD₅ in the headwaters, while nitrate becomes dominant downstream and during colder seasons. The HyRIM integrates key hydro-geomorphic parameters, including stream order, channel sinuosity, and in-stream travel time, with hyperparameters optimized through a data-driven calibration process. Evaluated using a robust leave-one-out cross-validation (LOOCV) scheme, the final calibrated HyRIM demonstrates exceptional predictive power (R² = 0.941, root-mean-squared error [RMSE] = 2.06), significantly outperforming conventional regression models. The findings demonstrate that HyRIM is a powerful tool for generating high-resolution pollution maps and designing more efficient monitoring strategies in complex river systems.