AI-Driven Hydro-Insights: Proactive Water Resource Management for Sustainable Agriculture in the Face of Climate Change

Climate change profoundly affects natural water resources by increasing extreme rainfall and persistent drought events. This impact has led to a rising likelihood of over-extraction of groundwater by Taiwanese farmers due to insufficient water resources. Quantifying groundwater pumping activities is challenging, thereby prompting this study to introduce a hybrid AI model combining a Convolutional-based Autoencoder with LSTM. The objective is to explore the spatiotemporal relationship between hydrometeorology and groundwater for providing a quantitative assessment of groundwater resources. To construct the model, a comprehensive dataset spanning two decades (2000-2019) is utilized, incorporating information from 33 groundwater monitoring wells in the Jhuoshuei River basin of Taiwan. Two types of datasets, observation and simulation, are employed for a robust analysis. The hybrid AI model yields accurate three-month-ahead forecasts for shallow groundwater in the Jhuoshuei River basin, with R2 performance ranging from 0.70 to 0.87 for T+1 (short-term forecasts) and from 0.42 to 0.69 for T+3 (long-term forecasts). The significance of these forecasts lies in their potential to empower farmers to increase crop cultivation efficiency. The long-term forecasts aid in formulating strategic plans for crop cultivation and fallow periods, promoting efficient agricultural management. Simultaneously, the short-term forecasts empower farmers to enhance irrigation efficiency, leading to a reduction in regional water consumption. This proactive approach aligns with Sustainable Development Goals (SDGs) 11 and 12, fostering sustainable water resource management practices. In essence, this hybrid AI model emerges as a valuable tool for proactive and adaptive water resource management, particularly crucial in the context of evolving climate conditions. Keywords: Groundwater management, AI, Deep Learning, regional forecast, machine learning, SDGs, Taiwan