Impact of Climate Variables on Reservoir Outflow: An AI Approach; A Case Study of Khanpur Reservoir, Pakistan

Abstract

This research examines the effect of climate factors on the outflow of the Khanpur Reservoir in Pakistan by employing ANN for modeling. To forecast the reservoir outflow, the hydrological and climate variables such as seepage, maximum temperature, rainfall, live storage, evaporation, humidity and inflow are utilized in the analysis. ANN models with 5x5 (5 neurons in each two layers), 10x10, 15x15, 20x20, 25x25 neurons combination were created and evaluated, and it was found that the model with 20 neurons in each hidden layer had the best performance having R² value of the training and validation set was 0.95 and 0.85 respectively. The sensitivity analysis through interaction profiles revealed that live storage is the most significant predictor of reservoir outflow due to its high main effect (84410) and comparatively low interaction with other variables (total effect: 9790). This underscores the importance of accurately measuring and incorporating live storage in reservoir management strategies. Moreover, inflow had moderate direct impact (0.097) and the highest overall impact (0.423) revealing it as another most significant and sensitive predictor of outflow. Rainfall exhibited minimal direct influence with main effect: 0.029) but significant interaction effects (total effect: 0.411) showing its moderate effects on outflow. Total seepage and maximum temperature had small main effects: 0.021 and 0.038, respectively, but their moderate total effects were 0.367 and 0.341, respectively, pointed to their indirect significance. In contrast, evaporation and relative humidity exhibited very small individual and combined impacts on outflow (main effect: 0 and 0.014; total effect: 29.03 and 0.129, respectively). These results reveal the strong impact of climate variables on reservoir outflow prediction and capability of ANN in modelling these aspects. This research adds to the existing literature on employing AI and machine learning for optimizing reservoir management and advancing water management techniques under the new climate reality.