Author(s): Yao Hu; Pavel Ivanov; Zherui Xu

Linked Author(s):

Keywords: Bounded rationality; Data scarcity; Groundwater; Irrigation behavior; Statistical modeling; Temporal clustering

Abstract: Accurately modeling and predicting farmers' irrigation behavior is crucial for sustainable agricultural practices and effective water resources management. However, the intricate nature of irrigation behavior, influenced by factors like bounded rationality and various environmental and socioeconomic conditions, poses significant challenges for accurate predictions. While statistical models hold promise, their effectiveness heavily relies on the available behavioral data. While accessing high quality environmental and socioeconomic data has become more feasible, obtaining irrigation behavior data at a relevant spatiotemporal scale remains challenging. Furthermore, these datasets are often noisy and prone to measurement errors. To address these data challenges in irrigation prediction, we developed a methodology framework using Deep learning-based Temporal Clustering (DTC) and Hidden Markov Models (HMMs). This framework allows us to: 1) uncover daily groundwater usage states near pumping wells where daily pumping data were collected; 2) extend predictions to regions lacking pumping observations. We applied this framework to predict daily groundwater use in the High Plains Aquifer Hydrologic Observatory Area (HPAHOA), using pumping data from dozens of wells near the Nebraska study area. Overall, our framework provides a potential solution to the common issue of data scarcity in modeling human decision-making.

DOI: https://doi.org/10.3850/iahr-hic2483430201-313

Year: 2024