Leveraging AI for predicting fallow land and resource allocation in the San Joaquin Valley of California
Environmental Research: Food Systems · 01.09.2026 · 3 (3) · 035024
DOI: 10.1088/2976-601x/ae9294 ↗Kurz gefasst
Fachartikel in der Zeitschrift Environmental Research: Food Systems, online erschienen am 21.08.2026. Verfasst von Abid Sarwar und 3 weiteren Autor:innen. Die Arbeit nennt 54 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: USDA NIFA AFRI SAS Securing a Climate Resilient Water Future, NSF/USDA AI Institute: Agricultural AI for Transforming Workforce and Decision Support.
Abstract (Originalsprache)
California’s San Joaquin Valley (SJV) faces persistent water scarcity, climate stress, and environmental degradation. Land fallowing, or idling cropland, is one management response used to reduce irrigation demand during dry or restricted-water years. Because fallowing changes unevenly across space and time, it complicates water management, strategic land-use planning, and multi-benefit land repurposing. This study forecasts the spatial and temporal distribution of fallowed lands in the SJV using a hybrid machine-learning and deep-learning framework coupled with a cellular automata. Using observed LandIQ-based cropland/fallow maps (training period (2014–2020) for the 2020–2022 validation period, the deep learning based Fallow Prediction Framework (DL-FPF; F 1 ≈ 0.81; accuracy ≈ 0.94) outperformed classical machine-learning models ( F 1 ≈ 0.65–0.67; accuracy ≈ 0.88–0.89). Across the full SJV, DL-FPF identified 70 586.89 ha of gross crop-to-fallow transition from WY2022 to WY2024, equivalent to ≈25% of the WY2022 mapped fallow baseline. The final WY2024 map predicted 265 657.60 ha of Idle/Fallow land, 20 343.04 ha lower than WY2022 (−7.11%). An independent WY2024 LandIQ overlay showed moderate agreement with the DL-FPF prediction, with 10-class accuracy of 0.73, Cohen’s Kappa of 0.66, binary fallow/non-fallow F1 of 0.56, and binary IoU of 0.39; DL-FPF overestimated WY2024 fallow area by 13.34% relative to LandIQ. Gross transition pressure was highest in GW-only GSAs (31.67%), followed by SW-only GSAs (28.87%) and SW + GW GSAs (23.77%), while SW + GW GSAs had the largest absolute gross transition area. By separating transition magnitude, mapped WY2024 fallow area, and reference-map agreement, DL-FPF outputs can better support water allocation, cropping decisions, and multi-benefit land repurposing under the sustainable groundwater management act.
Förderung
- USDA NIFA AFRI SAS Securing a Climate Resilient Water Future
- NSF/USDA AI Institute: Agricultural AI for Transforming Workforce and Decision Support
Zitieren
APA
Sarwar, A., Medellín-Azuara, J., Abatzoglou, J. T., & Viers, J. H. (2026). Leveraging AI for predicting fallow land and resource allocation in the San Joaquin Valley of California. Environmental Research: Food Systems, 3(3), 035024. https://doi.org/10.1088/2976-601x/ae9294
Vancouver
Sarwar A, Medellín-Azuara J, Abatzoglou JT, Viers JH. Leveraging AI for predicting fallow land and resource allocation in the San Joaquin Valley of California. Environmental Research: Food Systems. 2026;3(3):035024. doi:10.1088/2976-601x/ae9294
BibTeX
@article{Sarwar2026,
author = {Sarwar, Abid and Medellín-Azuara, Josué and Abatzoglou, John T and Viers, Joshua H},
title = {Leveraging AI for predicting fallow land and resource allocation in the San Joaquin Valley of California},
journal = {Environmental Research: Food Systems},
year = {2026},
volume = {3},
number = {3},
pages = {035024},
doi = {10.1088/2976-601x/ae9294}
}
EUVIMED