Short-term forecasting of monthly water consumption in hyper-arid climate using recurrent neural networks

Abstract

Freshwater supply is a major challenge in regions with limited water resources and extremely arid climatic conditions. The objective of this study is to model the monthly water demand in Kuwait using the nonlinear autoregressive with exogenous input (NARX) neural network approach. The country lacks conventional surface water resources and is characterized by extremely arid climate. In addition, it has one of the fastest growing populations. In this study, linear detrending is performed on the water consumption time series for the period from January 1993 to December 2018 to eliminate the influence of population growth before application to the NARX model. Monthly temperature data are selected as exogenous input to the NARX model, because they are strongly associated with the water consumption data. Correlation analyses are performed to determine the input and feedback delays of the NARX model. The results demonstrate that the recurrent NARX model is efficient and robust for forecasting the short-term water demand, with a Nash-Sutcliffe (NS) coefficient of 0.837 in the validation period. Seasonal model assessment shows that the model performs best during the critical summer season. The NARX-based recurrent model is established as a powerful and promising tool for predicting urban water demand. Thus, it can efficiently aid the development of resilient water supply plans.