[A-HW17] Near Surface Investigation and Modeling for Groundwater Resources Assessment and Conservation

[AHW17-P01] A Novel Approach to Groundwater Forecasting: CNN-LSTM Integration of Climate Variability with Groundwater Pumping Electricity Data

Sumriti Ranjan Patra¹、Hone-Jay Chu¹、*Mohammad Adil Aman¹ (1.National Cheng Kung University)

キーワード：Groundwater, Pumping, CNN-LSTM, Climate

Groundwater forecasting plays a pivotal role in sustainable water resource management. In this study, we propose a cutting-edge approach using a Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) model to predict monthly groundwater levels over Choushui River Alluvial Fan (CRAF) located in Central Taiwan. The model integrates climatic variables, i.e., precipitation and temperature, with groundwater pumping data that includes the electricity power consumption of pumping wells. Notably, our model incorporates electricity consumption data as part of the pumping information, shedding light on the energy-intensive nature of groundwater extraction and providing valuable insights into pumping well operations. Training and validation leverage historical groundwater data, climatic variables, and pumping-related features of up to 15 years (2007-21). Results demonstrate the CNN-LSTM model's superior performance, surpassing traditional time series models in accurately forecasting monthly groundwater levels. The inclusion of electricity consumption data significantly improves forecast precision, emphasizing the importance of considering operational characteristics in groundwater forecasting models. This research contributes to advancing groundwater forecasting methodologies, offering water resource managers a robust tool to enhance decision-making. The CNN-LSTM model's capacity to integrate diverse data sources and capture complex spatio-temporal relationships proves invaluable for sustainable groundwater management, especially amid changing climatic conditions and increasing water demands. The model's incorporation of pumping electricity consumption data along with climate variables adds a nuanced understanding of groundwater extraction dynamics, further enhancing its applicability in real-world scenarios.