Improving estimates of spatially distributed agricultural land use using pesticide data

Accurate, spatially resolved agricultural land-use data are essential for estimating water demand, evaluating agricultural production, and understanding long-term changes in managed landscapes. However, existing land-use datasets are often temporally discontinuous, particularly for historical periods, and remotely sensed classifications face persistent challenges distinguishing spectrally similar crop types, capturing sub-annual rotations, and separating active from abandoned plantings. We developed the California Pesticide Use Reporting Land-Use Estimator (CalPUR-LUE), which converts grower-reported pesticide-application records into annual estimates of land use by crop class and cropped acreage at approximately 1-square-mile resolution across California. CalPUR-LUE performance was evaluated for 2018 using complementary spatial, categorical, and quantitative comparisons with California Department of Water Resources (CADWR) remotely sensed land-use data and county-level agricultural economic reports. At the county-meridian-township-range-section (COMTRS) scale, CalPUR-LUE and CADWR acreage estimates showed substantial proportional agreement (log-transformed Lin’s concordance correlation coefficient = 0.77) with a mean proportional bias of −15%. Crop-class agreement was high when CalPUR-LUE identified agricultural activity, with user’s accuracy exceeding 83% for all evaluated crop classes. Overall categorical agreement was 67% (Cohen’s kappa = 0.56). Grain crop underestimation contributed substantially to disagreement; however, there is evidence of CADWR systematic overestimation of grain during 2016–2019. County-level agricultural reports further corroborated CalPUR-LUE estimates and, in several cases, aligned more closely with CalPUR-LUE than with CADWR. These results demonstrate that pesticide-use reporting provides an independent and complementary source of agricultural land-use information that can corroborate contemporary estimates and reconstruct annual spatially distributed land use where historical data are limited.