Prediction of groundwater level in the southwest plain of Tehran-Iran by Multiple Modelling (MM) and treating heterogeneity by self-organizing map (SOM)

Groundwater resources are crucial for meeting water supply needs, highlighting the importance of accurate modeling. The study of groundwater level (GWL) fluctuations holds significant implications for various fields such as management studies, engineering design, agricultural practices, and access to high-quality groundwater. With the increasing use of groundwater resources in recent years, there is a growing need for more serious resource management and closer monitoring of consumption.This research utilized three levels to predict fluctuations in groundwater levels. Firstly, the intelligent self-organizing map (SOM) method was used to cluster observation wells (OWs) in order to reduce heterogeneity in hydrogeological environments. Secondly, models including Sugeno fuzzy logic (SFL), recurrent neural network (RNN), and feedforward neural network (FNN) were utilized to predict groundwater level fluctuations based on regional and observational data, including groundwater level data, groundwater abstraction, temperature, and rainfall. Thirdly, the support vector machine (SVM) Artificial Intelligence (AI) strategy was applied to build further understanding, using the results of the second level as input data to improve results. The findings of this study indicate that the SFL model outperforms the other two models at the second level. Additionally, in the third level, the SVM model improved the results, with testing phase accuracies for categories 1, 2, and 3 improving from 0.92, 0.91, and 0.94 to 0.98, 0.96, and 0.99 respectively.