Identifying representative crop rotation patterns and grassland loss in the US Western Corn Belt

Crop rotations (the practice of growing crops on the same land in sequential seasons) reside at the core of agronomic management as they can influence key ecosystem services such as crop yields, carbon and nutrient cycling, soil erosion, water quality, pest and disease control. Despite the availability of the Cropland Data Layer (CDL) which provides remotely sensed data on crop type in the US on an annual basis, crop rotation patterns remain poorly mapped due to the lack of tools that allow for consistent and efficient analysis of multi-year CDLs. This study presents the Representative Crop Rotations Using Edit Distance (RECRUIT) algorithm, implemented as a Python software package, to select representative crop rotations by combining and analyzing multi-year CDLs. Using CDLs from 2010 to 2012 for 5 states in the US Midwest, we demonstrate the performance and parameter sensitivity of RECRUIT in selecting representative crop rotations that preserve crop area and capture land-use changes. Selecting only 82 representative crop rotations accounted for over 90% of the spatio-temporal variability of the more than 13,000 rotations obtained from combining the multi-year CDLs. Furthermore, the accuracy of the crop rotation product compared favorably with total state-wide planted crop area available from agricultural census data. The RECRUIT derived crop rotation product was used to detect land-use conversion from grassland to crop cultivation in a wetland dominated part of the US Midwest. Monoculture corn and monoculture soybean cropping were found to comprise the dominant land-use on the newly cultivated lands.