Machine learning for groundwater levels: uncovering the best predictors

When it comes to water for crops, groundwater reigns supreme. Its affordability and year-round availability make it a game-changer for a nation's economic engine. Geographically, the Barind tract suffers from drought in a historic manner due to the random fluctuation of groundwater levels. Thus, developing an accurate forecasting model plays a pivotal role in any kind of measure to be taken in the future. This study aimed to find the most accurate regression model to forecast the groundwater level from 2021 to 2030 using historical data from 1990 to 2020 by conducting a comparative analysis among Random Forest (RF), Support Vector Machine (SVM), and Gradient Boosting Machine (GBM) simulations. SVM regression came with more reliable accuracy metrics (MAE, MSE, RMSE, SMAPE, NSE, explaining variation, correlation, and Wilmott's Index of Agreement or d value). Compared to RF and GBM, SVM regression achieved better accuracy on the training data (1990-2010) with lower error terms (MAE, MSE, RMSE, SMAPE) and a higher correlation (77%) between predicted and actual values. SVM also explained a larger portion of the data variance (55%), suggesting it captured the key factors influencing groundwater levels, both in training and testing phase. The biasness of SVM regression was found to be closer to the zero-bias line, which indicated that the predicted values were less overfitted than the other models. Even a slow increase (0.003 m/year, as found in this study) in depth can lead to significant depletion over several decades, potentially impacting water availability in the long run. The analysis of future groundwater depth (2021-2030) predicts a relatively stable scenario with an average fluctuation of around 4.9 m from the ground surface. This stability comes after a period of increased fluctuation that began around 2024. Interestingly, the overall trend suggests a rising groundwater depth. This rise is expected to start with a potentially rapid initial increase, followed by a more gradual rise after 2028. Furthermore, collaboration between researchers, policymakers, and stakeholders in the water and agriculture sectors can help ensure that the forecasting models are effectively utilized to support sustainable water management practices in Bangladesh to achieve target 4.07f, which states strengthening groundwater monitoring under the National Water Management Plan (NWMP).