ASSESSING TILLAGE PRACTICES AND WATER USE EFFICIENCY ACROSS WINTER WHEAT FARMS IN CANADIAN COUNTY, OKLAHOMA

Abstract

Improving crop productivity and water use efficiency (WUE) under a changing climate is a challenge for winter wheat (Triticum aestivum L.) systems across the globe, including those in the Southern Great Plains. This study used multi-sensor satellite imagery, machine learning, and field observations to map till and no-till winter wheat fields and assess gross primary productivity (GPP), evapotranspiration (ET), and WUE in Canadian County, a key wheat-growing region in Oklahoma. Sentinel-2, Landsat 8/9, PlanetScope, Sentinel-1 Synthetic Aperture Radar (SAR), field observations, and United States Department of Agriculture (USDA) Cropland Data Layer were used in the analysis. The random forest model showed good performance in training (overall accuracy, OA, of 66 to 86%) but showed limited ability to characterize out-of-sample plots (OA of 57% from Sentinel-1 to 66% from PlanetScope). Till farms were more widespread across the county, while no-till farms were more localized and spatially clustered in the 2025 growing season. Validation of seasonal GPP, ET, and WUE with flux tower observations showed good agreement between modeled and observed values, suggesting the potential of using satellite-derived products to assess field-level variability in crop water use and efficiencies. Notably, seasonal GPP, ET, and WUE varied between till and no-till winter wheat fields across growing seasons, with no-till fields generally showing higher GPP and WUE across several years, while ET differences between management practices were smaller. This study demonstrates the potential of multi-sensor remote sensing and machine learning for monitoring tillage practices and assessing crop-water productivity in winter wheat systems.