Modeling and Prediction of Groundwater Level Fluctuations Using Geoinformatics and Artificial Neural Networks in Al Ain City, UAE

Abstract

Groundwater resources are of utmost importance in arid regions due to the scarcity and non-availability of surface water sources. Applications of geo-modeling integrated with machine learning and artificial intelligence algorithms are rapidly advancing for the sustainable management and conservation of those resources. In this pilot study, we explored the state-of-the-art ANN technologies as innovative techniques in a geo-modelling framework for the prediction of changes in groundwater level of part of Al Ain city, United Arab Emirates (UAE). We used Fill Forward Padding to fill the data gaps and utilized the state-of-the-art ranger optimization algorithm to find the most accurate model that suites our data. The model takes 8 inputs to produce the predicted groundwater levels. Training the model on the data gave the highest accuracy in the 408th epoch, through the following hidden layer configuration [700, 200, 600, 100, 9]. To fill the spatial gaps between the wells we made use of Ordinary Kriging utilizing the Circular semi-variogram model. The built model produced predicted variability in groundwater (2013–2019) based on the observations (2004–2012), thereby strengthening the future use of the system in machine learning. Results and observation data highlighted moderate variability in groundwater levels through 15 different annually averaged maps. These variations are attributed to both natural and anthropogenic factors. Indeed, extensive abstraction, changes in precipitation rates, population growth and urbanization strongly contributed to and controlled the groundwater fluctuations and level variability during the study period throughout the region.