5:15 PM - 7:15 PM
[AHW22-P04] Integrating Deep Learning and Explainable AI for Groundwater Level Prediction in Taiwan
Keywords:Groundwater forecasting, deep learning, explainable AI, Taiwan
Climate change has profoundly impacted global water resources, intensifying extreme events like droughts and floods, underscoring the need for effective groundwater management. Groundwater level prediction involves complex hydrological and climatic dynamics and is also influenced by human activities like groundwater pumping. Traditional methods, such as numerical models and empirical formulas, face significant challenges in accurately simulating groundwater levels due to the highly nonlinear nature of hydro-meteorological systems. In Taiwan, the Zhuoshui River basin, as one of the country’s primary irrigation zones, relies heavily on surface water from the Zhuoshui River and its underlying groundwater resources. Efficient groundwater allocation, supported by reliable predictive tools, is critical to sustainable river basin management, which serves as a major agricultural hub.
This study utilizes 10 years (2012–2022) of hydrological and anthropogenic activity data, including groundwater levels, river water levels, and pumps’ electricity usage, to develop a multi-input, multi-output, multi-horizon (MIMOMH) deep learning framework for 1- to 3-month ahead groundwater level prediction. For sustainable river basin management, Explainable AI (XAI) techniques further provide deeper insights into the feature importance of input variables within the proposed framework, particularly from the perspectives of individual groundwater and flow monitoring stations as well as groundwater pumps. This research is expected to advance groundwater level prediction and address critical resource management challenges by providing a replicable framework applicable to Taiwan and similar regions.
This study utilizes 10 years (2012–2022) of hydrological and anthropogenic activity data, including groundwater levels, river water levels, and pumps’ electricity usage, to develop a multi-input, multi-output, multi-horizon (MIMOMH) deep learning framework for 1- to 3-month ahead groundwater level prediction. For sustainable river basin management, Explainable AI (XAI) techniques further provide deeper insights into the feature importance of input variables within the proposed framework, particularly from the perspectives of individual groundwater and flow monitoring stations as well as groundwater pumps. This research is expected to advance groundwater level prediction and address critical resource management challenges by providing a replicable framework applicable to Taiwan and similar regions.