Utilizing deep learning to investigate the impacts of climate change on groundwater dynamics and pumping variability

Abstract

Climate change occasionally leads to unprecedented groundwater level decline. This study investigates the impacts of climate change on groundwater level and pumping electricity consumption concurrently over an agricultural region in Central Taiwan. A hybrid deep learning model i.e., CNN-LSTM was employed to predict future groundwater level and pumping electricity using monthly precipitation and average temperature as exogenous inputs from 2007 to 2021. We adopted future projections for these climate inputs from the latest CMIP6 climate models spanning 15 years (2022–36) for shared socioeconomic pathways (SSPs) scenarios (SSP245 and SSP585). Results revealed that the groundwater level fluctuations in the region are indirectly influenced by climate-induced pumping, as evidenced by exacerbated energy consumption for groundwater extraction during periods of high temperatures and reduced precipitation. Future projections imply that pumping electricity is expected to rise by 2.5–5 % while groundwater level may decline by more than 1 m due to a reduction in average precipitation of about 50–150 mm and a rise in temperature of about 0.7–1.3 °C, respectively over the next 15 years (2022–36). Most notable impacts are witnessed for irrigation-intensive regions such as mid and distal fans. This study necessitates the need to collect and involve pumping along with climate information for a more pragmatic assessment of future groundwater level that may indirectly threaten the food and water security in the region through proficient management of groundwater resources.