The spatial heterogeneity of vegetation types on a landscape has been linked to multiple ecosystem functions, including habitat for wildlife and pollinators, water cycling, human aesthetic values, and nutrient cycling. Although agricultural land uses are sometimes combined into a single unit when quantifying landscape heterogeneity, diverse cropping systems are a valuable alternative to near-monocultural croplands and contribute more strongly to ecosystem service provision, including services such as pest regulation and carbon sequestration that are of direct interest for agriculture. The USDA Cropland Data Layer was used to characterize crop diversity across the contiguous US for 2008-2018. Percentage of each crop type, along with non-crop uses such as forest and development, were calculated for each 4km PRISM climate data grid cell. To better understand the drivers of crop diversity, Random Forest modeling was used to assess the importance of climate, soils, and irrigation for patterns of crop effective richness for the contiguous United States, stratified by USDA Land Resource Region. The models explained 57-89% of the variation in maximum crop diversity, with irrigation being by far the most important explanatory variable in regions where it was employed. The drivers of change from 2008 to 2018 were less clear. Random Forest models explained only 20-60% of the change in agricultural diversity over the eleven-year period; both soil and climate properties were important, with no clear dominant drivers. Potential crop effective richness was greater than actual across the entire region studied, but substantial increases would require irrigation. Major changes in agricultural systems and infrastructure may be necessary to increase agricultural diversity at large spatial extents, and declining availability of water for irrigation could threaten the agricultural systems that are now most diverse.