Climate-Driven Forecasting of Agricultural Land Cover Change in North Dakota’s Red River Valley Using Remote Sensing and Machine Learning

Abstract The Red River Valley (RRV) of North Dakota, a major agricultural hub in North America, has experienced notable shifts in crop patterns over recent decades. However, few studies have quantitatively examined how long-term climatic trends influence spatial changes in crop patterns or how these climate variables can be used to predict future agricultural land cover in this region. This study evaluates climate–agriculture dynamics from 1997 to 2023 and forecasts land cover changes for 2033 using an integrated framework of remote sensing, geospatial analysis, and machine learning (ML). Historical rainfall and temperature records from the North Dakota Agricultural Weather Network (NDAWN), crop-specific land cover data from the United States Department of Agriculture (USDA) Cropland Data Layer, and topographic variables from the North Dakota GIS Hub were analyzed using Google Earth Engine (GEE). Temporal trends revealed a consistent warming trend during the April–September growing season from 1997 to 2023(by approximately 1.4–1.8 °C), accompanied by fluctuating rainfall variability across the RRV. Spatial analysis identified expansion of corn and soybeans cultivation between 1997 and 2023, with corn increasing from 7.43% to 24.17% and soybeans from 19.55% to 46.68% of total cropland, while sunflower cultivation declined sharply from 12.26% to 1.45%. Using Auto Regressive Integrated Moving Average (ARIMA) models for climate forecasting and Random Forest (RF)algorithms for agriculture land cover prediction, the 2033 scenario indicates a major shift toward spring wheat cultivation that was projected to cover over 70% of agricultural land while corn and soybeans are expected to decline to 13.9% and 13.6%, respectively. Model performance was strong, achieving an overall classification accuracy of 99.0% and a kappa coefficient of 0.988. Hotspot analysis further indicates spatial concentration of dominant crops, emphasizing emerging monoculture risks and potential threats to soil fertility and food security. These findings underscore the value of ML and remote sensing for anticipating climate-driven agricultural transitions and informing adaptive land-use strategies in vulnerable agroecosystems. Graphical Abstract The graphical abstract summarizes how climate data and machine-learning (ML) techniques were used to study and predict changes in agricultural land cover (ALC) in North Dakota’s Red River Valley. Weather station observations and satellite-based remote sensing provided historical rainfall, temperature and agricultural land cover (ALC) data from 1997 to 2023. These climate variables influence crop growth and were analyzed along with major agricultural land-cover types, including corn, soybeans, spring wheat, and sunflower. The ARIMA (Auto-Regressive Integrated Moving Average) model was applied to forecast rainfall and temperature for the year 2033. These projected climate values, combined with historical crop patterns, were then used in a Random Forest (RF) model implemented in Google Earth Engine (GEE) to predict future agricultural land-cover distribution. The results show how crop patterns may change by 2033, with spring wheat becoming dominant while corn and soybean areas decrease. Overall, the graphical abstract visually illustrates the workflow starting from data collection and climate forecasting to ML-based prediction of future agricultural conditions, visualizing its hotspot.