Enhancing the accuracy of metaheuristic neural networks in predicting underground water levels using meteorological data and remote sensing: A case study of Ardabil Plain, Iran

Groundwater is an essential water source for many uses worldwide, including domestic, agricultural, and commercial. It is also the most valuable water source, known as national wealth in some communities. Therefore, modeling and predicting groundwater levels are essential in groundwater management. This study uses Artificial Neural Network-Particle Swarm Optimization (ANN-PSO), Artificial Neural Network-Genetic Algorithm (ANN-GA), and Artificial Neural Network-Ant Colony Optimization (ANN-ACO) algorithms to predict groundwater level changes in the Ardabil Plain in northern Iran. The modeling process includes the use of weather data and groundwater level analysis. The main aim is to increase the model's accuracy in predicting groundwater levels without relying on historical water statistics as input data. The parameters studied include precipitation, temperature, runoff, and withdrawal from wells, springs, or aqueducts, completed in three ways for the processing model. In addition to meteorological data, remote sensing techniques were employed to obtain necessary data, including runoff estimation and land use analysis. Evaluation statistics such as Nash-Sutcliffe efficiency (NSE), correlation coefficient (r), and Root Mean Squared Error (RMSE) were used to evaluate the model's accuracy. The results show that the ANN-ACO offers greater accuracy than the other methods due to the shared database. In particular, among the three algorithms, the ANN-GA offered the best accuracy for observation well 2 (NSE: 0.655, r: 0.820 and RMSE: 0.055) and 4 (NSE: 0.832, r: 0.910 and RMSE: 0.021) while for level 13, the ANN-PSO performed the best with NSE, r and RMSE values of 0.500, 0.720 and 0.019, respectively. The method proposed in this study demonstrated its ability to improve prediction by increasing the model's accuracy in the same algorithm and with the same data, up to 76% just by using data analysis. Results demonstrate significant improvement in prediction accuracy, highlighting the effectiveness of the proposed methodology in groundwater level prediction. © 2024