Comparing the impact of mapping error on the representation of landscape pattern on upscaled agricultural maps

Upscaling techniques have been extensively used to produce upscaled maps to fill data gaps serving various Earth observation models by providing area and landscape pattern information. Base maps as input for upscaling techniques inevitably have mapping errors that greatly impact the upscaling performance. However, the influence of mapping error on the representation of landscape pattern of upscaled maps has rarely been explored. To address this issue, the Crop Data Layer (CDL) data for two study sites were first used to generate agricultural maps as the base maps. A probability-based Monte Carlo algorithm was then used to simulate different error levels for the base maps. Two upscaling techniques, Fusing class Membership probability and Confidence level probability (FMC) and a conventional upscaling method (i.e., Majority Rule Based, MRB), were conducted. The results highlight that higher mapping error results in higher change of landscape pattern for upscaled maps. Overall, this work extends our understanding of the influence of mapping error on the upscaling performance. Also, it suggests that next generation upscaling techniques should greatly consider the mapping error and how to accurately present landscape pattern.