Mapping forest in the southern Great Plains with ALOS-2 PALSAR-2 and Landsat 7/8 data

Southern Great Plains (SGP) of the United States, comprising the states Kansas, Oklahoma, and Texas, spans diverse climatic regions. In recent decades, woody plant continues to expand and form forest (above 2 m in height) across the SGP. However, our knowledge of the forest amount and distribution in this region is very limited. This study aims to map forest, especially evergreen forest (above 2 m in height) in the SGP for the time period of 2015-2017. Annual mosaic data of HH and HV polarization backscattering (25 m) from Phased Arrayed L-band Synthetic Aperture Radar-2 (PALSAR-2) aboard Advanced Land Observing Satellite-2 (ALOS-2), along with their difference (HH-HV) and ratio (HH/HV) were utilized. With the four bands (HH, HV, difference, ratio) of 2017, decision rules of forest were developed based on 30 randomly selected forest plots (as of 2017) across the study area. With the decision rules, a PALSAR-2 based forest map was created for each year from 2015 to 2017. Then an annual maximum normalized difference vegetation index (NDVI) threshold of 0.5, derived from Landsat 8 data of 2017 and the 30 forest plots, was used to filter potential commission error of rough surface and building in each PALSAR-2 based forest map. After that, a forest map circa 2016 was generated, in which each forest pixel was identified as such at least twice during 2015 and 2017. Lastly, a threshold of seasonal NDVI change (0.3) was derived to extract evergreen forest out of the forest map circa 2016. Accuracy assessment for the result forest map suggests a user's accuracy of 99.2% and a producer's accuracy of 88.7% for forest. Accuracy assessment for the evergreen forest map suggests a user's accuracy of 97.3% and a producer's accuracy of 90.5% for evergreen forest. The result forest map, especially the evergreen forest map, paves the way for follow-up studies on forest resource and woody plant encroachment in the SGP.