Get reliable rangeland science

LANDSAT imagery is a freely available global dataset that is attractive for many remote sensing applications, however, the 30m resolution is not ideal for more detailed analyses such as estimating woody plant cover on a localized scale.� In this study, we researched the feasibility of using 2m high resolution (HiRes) imagery to create a woody cover classification training dataset that would be applied to natural color (4,3,2 band) LANDSAT 8 data over a desert area in Kenya, Africa, and oak-juniper cut plains near Lampasas, Texas, USA.� We began by classifying the HiRes imagery into woody and non-woody classes using Example Based Feature Extraction in ENVI 5.3.� The resulting HiRes raster classification layer was Aggregated into a 30m cell size in ArcGIS 10.3 and reclassified based on percent woody cover per cell (< 20%, 20-40%, 40-60%, 60-80% and >80%). A random sampling of 2,000 cells was taken from the aggregated raster.� Of these 2,000 cells, 1,000 were used as an example training dataset on a LANDSAT image, while the other 1,000 were used to test the classification. Our resulting LANDSAT classifications were 49.7% and 54.7% in agreement with the HiRes classifications in Africa and Texas, respectively. Both sites had better agreement in the lowest (<20%) and highest (>80%) classifications. In Kenya, the <20% LANDSAT cover class matched the HiRes classification in 74.4% of the cells; while the >80% cover class matched in 64.4% of the cells. The Texas site matched 86.59% and 50.75% in the <20% and >80% cover classes, respectively.� Our results showed very similar results in two strikingly different ecosystem types, suggesting that the methodology may be repeatable in other areas as well.� Areas of very low and high woody plant cover were the most discernable in the LANDSAT imagery, while areas of moderate cover were often misclassified.