Predicting groundwater level using traditional and deep machine learning algorithms

Abstract

This research aims to evaluate various traditional or deep machine learning algorithms for the prediction of groundwater level (GWL) using three key input variables specific to Izeh City in the Khuzestan province of Iran: groundwater extraction rate (E), rainfall rate (R), and river flow rate (P) (with 3 km distance). Various traditional and deep machine learning (DML) algorithms, including convolutional neural network (CNN), recurrent neural network (RNN), support vector machine (SVM), decision tree (DT), random forest (RF), and generative adversarial network (GAN), were evaluated. The convolutional neural network (CNN) algorithm demonstrated superior performance among all the algorithms evaluated in this study. The CNN model exhibited robustness against noise and variability, scalability for handling large datasets with multiple input variables, and parallelization capabilities for fast processing. Moreover, it autonomously learned and identified data patterns, resulting in fewer outlier predictions. The CNN model achieved the highest accuracy in GWL prediction, with an RMSE of 0.0558 and an R2 of 0.9948. It also showed no outlier data predictions, indicating its reliability. Spearman and Pearson correlation analyses revealed that P and E were the dataset’s most influential variables on GWL. This research has significant implications for water resource management in Izeh City and the Khuzestan province of Iran, aiding in conservation efforts and increasing local crop productivity. The approach can also be applied to predicting GWL in various global regions facing water scarcity due to population growth. Future researchers are encouraged to consider these factors for more accurate GWL predictions. Additionally, the CNN algorithm’s performance can be further enhanced by incorporating additional input variables.