Scaling an Artificial Neural Network-Based Water Quality Index Model from Small to Large Catchments

Scaling models is one of the challenges for water resource planning and management, with aim bringing developed into practice by applying them to predict quality quantity catchments that lack sufficient data. For this study, we evaluated artificial neural network (ANN) training algorithms index in a source catchment. Then, multiple linear regression (MLR) were developed, using predicted ANN variables, as dependent independent respectively. The most appropriate MLR model has been selected on basis Akaike information criterion, sensitivity uncertainty analyses. performance was then variable aggregation disaggregation approach, upscaling downscaling proposes, data from four very large- three large-sized eight medium-, small- seven small-sized catchments, where they are located southern basin Caspian Sea. algorithms, including Quick Propagation, Conjugate Gradient Descent, Quasi-Newton, Limited Memory Levenberg–Marquardt, Online Back Batch index. results show highest mean absolute error observed WQI, LM algorithm; lowest values LMQN CGD algorithms. Our findings also indicate upscaling, aggregated could provide reliable index, since r2 coefficient varies 0.73 ± 0.2 large 0.85 0.15 downscaling, disaggregated ranges 0.93 0.05 0.97 0.02 medium catchments. Therefore, scaled be applied perform rapid assessment study area.