Authors: Michaela Buenemann*, New Mexico State University
Topics: Remote Sensing, Land Use and Land Cover Change, Arid Regions
Keywords: remote sensing, land cover, desert, WorldView-3, NAIP
Session Type: Paper
Start / End Time: 4:30 PM / 6:10 PM
Room: Marshall North, Marriott, Mezzanine Level
Presentation File: No File Uploaded
Accurate and detailed land cover maps are crucial for sustainable land management and planning, habitat suitability mapping, and a range of other applications. The existing literature on high spatial resolution land cover mapping confirms this. However, little work has been done on mapping land cover in desert landscapes composed of a complex mix of urban and rural materials. The goal of this study was to address this weakness using Las Cruces, New Mexico in the Chihuahuan Desert of the southwestern United States as a case study area. To meet this goal, I used Random Forest to classify high spatial resolution WorldView-3 satellite imagery and National Agriculture Imagery Program (NAIP) aerial imagery. The image classifications considered the original multispectral bands; hundreds of multi-scale image derivatives such as principal component and tasseled cap bands, indices, texture measures, and spatial autocorrelation statistics; as well as numerous topographic variables derived from a digital elevation model of the area. The results suggest that land cover map accuracy can be enhanced significantly through the use of multi-scale image derivatives and ancillary data and that WorldView-3 imagery facilitates the production of much more representative land cover maps than NAIP imagery. The presentation will discuss these results in more depth and address differences in overall and class-level accuracies in the different classifications (e.g., spectral data only vs. spectral and contextual data) of the two datasets (WorldView-3 and NAIP) as well as the relative importance of the spectral and contextual variables in the different land cover mapping efforts.