Evaluating sampling designs for assessing the accuracy of cropland extent maps in different cropland proportion regions.

Abstract

The GFSAD30m cropland extent map has been recently produced at a spatial resolution of 30m as a part of NASA MEaSUREs’ Program Global Food Security Data Analysis (GFSAD) project. Accuracy assessment of this GFSAD30m cropland extent map was initially performed using an assessment strategy involving a simple random sampling (SRS) design and an optimum sample size of 250 for each of 72 different regions around the world. However, while statistically valid, this sampling design was not effective in regions of low cropland proportion (LCP) of less than 15% cropland area proportion (CAP).The SRS sampling resulted in an insufficient number of samples for the rare cropland class due to low cropland distribution, proportion, and pattern. Therefore, given our objective of effectively assessing the cropland extent map in these LCP regions, the use of an alternate sampling design was necessary. A stratified random sampling design was applied using a predetermined minimum number of samples followed by a proportional distribution (i.e., SMPS) for different cropland proportion regions to achieve sufficient sample size of the rare cropland map class and appropriate accuracy measures.The SRS and SMPS designs were compared at a common optimum sample size of 250 which was determined using a sample simulation analysis in ten different cropland proportion regions. The results demonstrate that the two sampling designs performed differently in the various cropland proportion regions and therefore, must be selected according to the cropland extent maps to be assessed.

Department

Natural Resources and the Environment

Publication Date

4-23-2019

Journal Title

Journal of Geography, Environment and Earth Science International

Publisher

Yadav and Congalton

Digital Object Identifier (DOI)

10.9734/jgeesi/2019/v20i430111

Document Type

Article

Rights

© 2019 Yadav and Congalton

Link to Full Text

Share

COinS