Explore machine learning-based models to predict regional groundwater levels and assess groundwater vulnerability in the Zhuoshui River basin of Taiwan

Climate change has led to the frequent occurrence of extreme weather, which has highly impacted the environment all over the world. In Taiwan, plenty of challenges relevant to groundwater over-exploitation have arisen due to increasing water demands caused by economic expansion or water shortages caused by droughts. So how to predict groundwater levels and assess groundwater vulnerability are high-value research topics.

The Zhuoshui River basin in Central Taiwan has encountered over-exploitation and severe land subsidence, and therefore has raised the safety concerns of the Taiwan High Speed Rail that runs through this basin. Droughts also occur frequently in this basin due to climate change, leading to increasing groundwater demands from industrial and agricultural sectors. This study intends to explore machine learning-based models to predict regional groundwater levels and assess groundwater vulnerability for the Zhuoshui River basin. We first collected 20 years (2000-2019) of hydro-meteorological data including groundwater level (5 aquifers), rainfall, and runoff for model construction. Next, from a suite of potential factors, we use the random forest to identify input factors of the machine learning-based model. Thirdly, we construct the machine learning-based model to predict groundwater levels. Finally, we provide a novel integration of the prediction results and other geological factors such as land use, soil type and terrain slope to assess groundwater vulnerability of the Zhuoshui River basin. The goal of this study is to create a regional groundwater vulnerability map to reveal the vulnerability degree (high, medium, and low) based on a vulnerability index (VI). The analyzed results indicate our findings are related to several sustainable development goals (SDGs) and can support governments in water resources planning.