Fine grain adjustments to address pixel-level misclassifications of a land cover data layer change derived landscape metrics relevant to areawide pest management

The availability of spatial data on agricultural land cover, such as the cropland data layer (CDL) developed by the US Department of Agriculture, has facilitated areawide pest management study. The CDL is derived from a raster-based process that primarily uses data layers of satellite images to produce a georeferenced land cover classification. Here, we considered (i) the extent that adjustments at the pixel level to address land cover misclassifications of the CDL changed simple composition (percent cover) and more complex composition (diversity) and configuration metrics, and (ii) if changes were sensitive to spatial scale (10 km2 and ≤ 3 km2 areas) and land cover (three crops and 3 seminatural lands). Next, we considered (iii) whether the landscape metrics derived from the original CDL as modified by 2 adjustment methods affected outcomes of a stepwise regression to consider selected landscape metrics as indicators of cotton fleahopper (Pseudatomoscelis seriatus) and sorghum aphid (Melanaphis sorghi) abundance. Changes in the metrics resulted from applying 2 adjustments (majority filter and manual) to extracted buffers taken from the CDL, with indication of overcorrection using the manual method. Selected metrics were sensitive to landscape structure (8 of 26 two-way interactions of adjustment, scale, and land cover were significant). In regression exercises, configuration metrics of cotton, sorghum, and grassland were the main metrics selected, and fewer metrics were selected when derived from adjusted buffers. Overall, the majority filter adjustment showed moderate sensitivity to scale and land cover and appeared to clarify landscape structure (particularly configuration of land covers) that influences two cotton and sorghum pests. Attention to data quality and downstream analysis for areawide pest management study is advisable going forward for this CDL and likely other land cover data products now available worldwide.