[A-HW22] River Channel Morphology, Water Resource Management, and Advanced Techniques

[AHW22-P07] Hybrid Deep Learning-Physical Model for Regional Groundwater Level Prediction

*JUN-JIE LIN¹、 Fi-John Chang¹、Meng-Hsin Lee¹ (1.National Taiwan University)

キーワード：Spatial Features, Residual, AutoEncoder, Convolutional Neural Network (CNN), Groundwater level

This study proposes a novel hybrid deep learning-physical model, the Residual-AutoEncoder-Convolutional Neural Network (Residual-AE-CNN), designed to enhance the accuracy of monthly groundwater level predictions through advanced feature extraction techniques. The model leverages diverse data sources, including (1) monthly water supply potential features extracted by an AutoEncoder (AE), (2) residual data between observational groundwater level data and groundwater level forecasts derived from the difference between observed groundwater levels and simulations generated by the HBV hydrological model, and (3) power consumption data from groundwater pumps. These inputs are then processed using a one-dimensional convolutional neural network (1D-CNN) to effectively capture spatial features and temporal dependencies. The model was applied to 18 groundwater monitoring stations in Taiwan’s Zhuoshui River alluvial fan, using data collected from 2007 to 2022. To improve predictive accuracy across different time horizons (T+1 to T+3), independent models are constructed for the proximal, mid and distal fans within the study area. The results demonstrate that the Residual-AE-CNN model excels in extracting spatial and temporal features, significantly improving prediction accuracy and reducing temporal delay errors inherent in the HBV model. Compared to the HBV model, the Residual-AE-CNN achieves a 20–30% improvement in R² across various sub-regions. The findings underscore the model’s potential to support sustainable water resource management and agricultural practices. Moreover, the proposed approach contributes to achieving the United Nations Sustainable Development Goals (SDGs), particularly SDG 2 (Zero Hunger) and SDG 6 (Clean Water and Sanitation), by facilitating efficient water resource allocation and management strategies.