Mapping Agriculture Expansion in Mato Grosso, Brazil, Using Coefficient of Variation Analysis on Synthetic Aperture Radar Time Series

Synthetic Aperture Radar (SAR) is a useful resource for analyzing dynamic signals on Earth. SAR time series stacks provide rich, regularly sampled, and weather-independent observations that can be used to identify, map, and monitor agricultural areas. While the general applicability of SAR for agriculture area mapping has been demonstrated in previous papers, this study analyzes the applicability of SAR time series data for mapping rapid expansion of agriculture areas in dynamically changing environments. We employ a SAR-based methodology to track cropland expansion in Mato Grosso, Brazil, from 2016 through 2020. This region was selected due to extensive land cover changes within the timeframe of available Sentinel-1 imagery. Furthermore, independent reference information on crop extent is only available infrequently in this region, requiring alternative methods to facilitate annual crop monitoring. Our methodology calculates the ideal coefficient of variation (CoVar) threshold to differentiate crops from non-crop areas. CoVar is a unitless statistical measurement, defined as the standard deviation divided by the mean. It is applied here on the radar brightness backscatter values for single pixels in year-long time series stacks. High CoVar values are associated with areas of greater change. Throughout the growing and harvesting season, agricultural areas experience changes in the moisture content and structure of the crops that lead to higher CoVar values than surrounding landcover types. The ideal threshold value to differentiate crops from non-crops is extracted using a threshold algorithm trained against reference data provided by the USDA, and from Copernicus Global Landcover. Our paper will outline the algorithm used in our approach and introduce the input data needed to train the algorithm. We will summarize the results of our performance assessment activities, specifically those that evaluated how crop mapping performance behaves as the time difference between algorithm calibration and analyzed SAR observations increases. We find that CoVar analysis of SAR time series can provide timely updates on changes in cropland extent in the Mato Grosso region, and has the potential to supplement and support existing crop mapping techniques.