Reconciling Field Size Distributions Of NASS (National Agricultural Statistics Service) Cropland Data.

Cropping system models and decision support tools are increasingly structured to provide area-wide simulation and analysis at multiple spatial scales. A major challenge is having access to large volumes of geo-referenced data. The Cropland Data Layer (CDL) products from USDA National Agricultural Statistics Service (NASS) contain digitized data layers of cropland distribution. However, the CDL data has too many single pixels or small pixel clusters that are classified as crop fields due to spectral artifacts and unrealistically large polygons also classified as fields due to CDL's limited ability to distinguish field boundaries. This paper presents a methodology to reconcile NASS CDL data with observed field size distribution data. The reconciliation system consists of several automated processing steps, including field delineation by road networks, streams, open road extension; pixel merging; fishnet parsing; and distribution optimization. The distribution optimization uses both pixel merging and fishnet parsing to match the reconciled data with observed field size distribution data. Observed field size data for rice and cotton (2009-2012), and corn and sorghum (2012) from selected counties in Texas were used to test the algorithm.  The 2009-2012 CDL data underestimated total rice acreage by 2-12% for Colorado County and overestimated total cotton acreage by 1-10% for Dawson County, while the CDL data overestimated the number of fields, which are represented by individual polygons, by up to 714% for rice and 280% for cotton. The number of rice fields in the optimized CDL data is reduced 39-99% for Colorado, Jefferson, Matagorda and Wharton County, with the total acreage change ranging from a 1% increase to a 33% reduction. The large reduction in total acreage mainly occurs during optimization and only for years in which the CDL data substantially overestimated total rice acreage. The number of cotton fields in the optimized CDL data increased by 15% for 2009 and reduced by 35-71% for 2010-2012 for Dawson County, with total acreage reduced by 5-12%. Most of the reduction in acreage is due to areas that were classified as cotton fields but now identified as road networks since the CDL data has only limited capability to separate fields from road networks. Summary data were also provided for other land types. Field size distributions of the reconciled CDL data closely match observed data and are appropriate for use in cropping systems simulations and analyses involving crop rotation, land change, area-wide pest management, and biorefinery site selection.