Application of machine learning techniques in the evaluation of local groundwater level using field gravity data: Insights from random forest and random tree

Evaluation of local groundwater levels (GWL) is crucial for sustainable water resource management. This study introduces a novel approach leveraging field gravity data to optimize machine learning (ML) models for local GWL evaluation. This unique approach holds the potential to bring about significant breakthroughs in groundwater management. The performance of two ML-based models, random forest (RF) and random trees (RT), is evaluated using four trials with distinct training and testing datasets. Model efficiency was assessed using R2, root mean squared error, mean absolute error, and Nash -Sutcliffe efficiency statistical metrics. Results indicated high model performance during training across all trials, with notable variability during testing. Trial 4 emerged as most successful, with RF achieving R2 and NSE of 0.998 during training and 0.965 during testing. Similarly, RT maintained high performance with R2 and NSE of 0.994 during training and R2 of 0.951 and an NSE of 0.917 during testing. The primary aim of the study was to assess the efficiency of decision tree-based ML techniques in capturing the characteristics of local GWL fluctuations. Addressing these areas will enhance the predictive accuracy of ML-based models and reliability for local groundwater storage evaluation, contributing to better water resource management and informed decision-making.