Resource information
A ca. 2010, 30m resolution map depicting annually tilled areas across the conterminous United States was developed. Input sources included fouryears, spanning 2008–2011, of annual national-level coverage Cropland Data Layer (CDL) land cover classifications as produced by the National Agricultural Statistics Service. Derived total land area under tillage from the aggregate CDL product equaled 112.8million hectares (278.7million acres). By comparison, the 2007 Census of Agriculture (CoA) produced an estimate of 122.9million hectares, suggesting the map is under representing tilled area by 10.1million hectares or 8.2%. Regression analysis using state-level summaries showed a strong, albeit biased, correlation (r-squared=0.99) between the CDL derived tilled area and the CoA information. Notable outliers were North Dakota and Montana. Comparisons of the CDL tilled map were also made against the 2006 National Land Cover Dataset (NLCD) land cover product’s Cultivated Crops category. Strong state-level regression agreement (r-squared=0.98) was also found between the NLCD and the CDL acreages, but the NLCD estimated 8.5% more area than the CDL and thus closely matched that of the CoA. However, significant pixel level differences were found between the CDL and the NLCD. Nationally 5.6% of the maps were in disagreement as to whether cultivated or not, a large proportion considering around a seventh of the country’s land area is tilled. States of Arkansas, Montana and Wisconsin had the largest absolute discrepancies between the NLCD and CDL. Accepting the CDL as reference showed a national level NLCD cropland commission error of 23.0% and omission error of 14.5%. Much of what is believed to be problematic in the NLCD could be explained by definitional issues having included alfalfa hay into their cultivated category for many areas. Ultimately, while it is likely that the CDL annually tilled area model is an underestimate of the true total, taken contextually in map form and adjusted for undercount bias it likely is the best available.
