Artificial neural networks for estimating historical daily missing evaporation to support sustainable development in Saudi Arabia

Abstract

Evaporation (EVAP) is a crucial component of the water cycle; however, its estimation is challenging due to its complexity and numerous influencing factors. Estimating EVAP is essential for identifying the environmental effects of heavy metals. It enables the forecasting of contamination risks, concentration changes, and the establishment of suitable mitigation suggestions, particularly in semi-arid and arid regions. This research presents a novel approach to evaluating the efficacy of regional models in estimating missing EVAP at a gauging location in the Al-Medina region. The estimates were generated using time series data on wind speed (WS), relative humidity (RH), temperature (TEMP), and evaporation (EVAP) from January to December (1974–1977; 2007–2009) through models employing the artificial neural network (ANN) feedforward backpropagation (FFBP) technique. The initial phase involved the development and training of the ANN, utilizing the FFBP technique in MATLAB (Version R2015a). The optimal network was then used to predict the EVAP values for 1974–1976, a missing parameter at the gauging site, by employing TEMP, RH, WS, and EVAP data from 2007 to 2009. The second stage includes verifying the predicted EVAP values (1974–1976) by using them to estimate the EVAP values for 1977 at gauged sites. Four ANNs (T1-T4) with distinct configurations were built and trained using the FFBP algorithm. The model's predicted values are compared with the actual EVAP values observed at measurement sites. The value of R² for the optimal topology was determined to be 0.981, with a mean squared error (MSE) of 0.019.