Rainfall-runoff modeling using machine learning in the urban watershed of Pishin Lora Basin, Balochistan (Pakistan)

Abstract

Abstract Pishin Lora Basin (PLB) is one of the major and most water-scarce basins in the Balochistan province of Pakistan. The Pishin Lora River flowing through the basin supports agriculture and domestic water use. However, the available scientific research on hydrological modeling in the study area is either limited or outdated. The growing impact of climate change on the basin's hydrology has increased interest in forecasting, which requires flow prediction studies using innovative approaches. In this study, Artificial Neural Network (ANN), a machine learning (ML) data-driven approach, is utilized as an alternative to hydrological models. Multilayer perceptron (MLP) model was structured for the runoff prediction. The model's performance evaluation was done to check the accuracy of the prediction. Remotely sensed thirty-two years (1990–2022) observed daily, weekly, and monthly rainfall, humidity and temperature dataset of Modern-Era Retrospective analysis for Research and Applications-2 (MERRA-2) satellite were used as model inputs to predict the watershed's runoff as the output. Runoff data were generated using Soil Conservation Service Curve Number (SCS-CN) Method. Statistical tests were used for the performance evaluation of the model, including coefficient of determination (R2), mean absolute error (MAE) and root mean squared error (RMSE). Four input combinations were used for the prediction of runoff. With rainfall alone as a input the R2 for daily, weekly and monthly datasets were 0.734, 0.968 and 0.906 respectively. Compared to the daily dataset, the model performed better for the weekly and monthly datasets. This demonstrates that the model is effective at predicting the runoff.