Comparison of Different Artificial Intelligence Techniques to Predict Floods in Jhelum River, Pakistan

Abstract

Floods are among the major natural disasters that cause loss of life and economic damage worldwide. Floods damage homes, crops, roads, and basic infrastructure, forcing people to migrate from high flood-risk areas. However, due to a lack of information about the effective variables in forecasting, the development of an accurate flood forecasting system remains difficult. The flooding process is quite complex as it has a nonlinear relationship with various meteorological and topographic parameters. Therefore, there is always a need to develop regional models that could be used effectively for water resource management in a particular locality. This study aims to establish and evaluate various data-driven flood forecasting models in the Jhelum River, Punjab, Pakistan. The performance of Local Linear Regression (LLR), Dynamic Local Linear Regression (DLLR), Two Layer Back Propagation (TLBP), Conjugate Gradient (CG), and Broyden–Fletcher–Goldfarb–Shanno (BFGS)-based ANN models were evaluated using R2, variance, bias, RMSE and MSE. The R2, bias, and RMSE values of the best-performing LLR model were 0.908, 0.009205, and 1.018017 for training and 0.831, −0.05344, and 0.919695 for testing. Overall, the LLR model performed best for both the training and validation periods and can be used for the prediction of floods in the Jhelum River. Moreover, the model provides a baseline to develop an early warning system for floods in the study area.