Regional-specific influence consideration for groundwater level prediction using machine learning-based approach

Groundwater level (GWL) is one of the important climate change components, directly and indirectly related to geoenvironmental disasters such as ground subsidence, flood, and drought. For two decades, many data-based approaches, including machine learning (ML), have mainly focused on the optimization algorithm and computational process. In this study, a new methodology was proposed for considering the regional-specific infiltration characteristics of given study sites. For this, the time delay between the influencing factors and GWL of each study site was introduced to the ML process, based on the trend-fitting technology of the moving average. Two study sites in South Korea were selected and the performance of the proposed methodology was verified. In addition, it was found that the proper time lag consideration for the GWL prediction is quite important than the amount of learning data length. This indicates that regional-specific characteristics need to be taken into account for more effective and precise GWL prediction.