Comparison of high spatial resolution imagery for efficient generation of GIS vegetation layers

Traditional vegetation mapping approaches require extensive field reconnais sance, and normally involve delineation of vegetation boundaries through in terpretation of aerial photographs. In order to generate vegetation data la yers within a geographic information system (GIS), boundaries must be digit ized and georeferenced and vegetation attributes coded. A project - conduct ed through the National Aeronautics and Space Administration Affiliated Res earch Center at San Diego State University for Ogden Environmental and Ener gy Services, Inc. - investigated the utility of very high spatial resolutio n (1-m) digital multispectral image data for generating GIS vegetation laye rs. Mapping and digital encoding of vegetation polygons was performed using USGS color-infrared (CIR) digital orthophotographic quarter quadrangle (DO QQ) and Airborne Data Acquisition and Registration (ADAR) 5500 imagery The two data sources were compared in the context of a controlled experiment wh ich tested the utility of the imagery under multiple mapping scenarios. The study area was a habitat reserve within Marine Corps Air Station Miramar n ear San Diego, California. This area primarily supports shrubland vegetatio n types typical of the Mediterranean climate area of southern California. CIR image data derived directly from multispectral digital cameras (e.g., A DAR System 5500) enabled more accurate classification and mapping of vegeta tion than did digital imagery generated from scanned CIR aerial photographs . This result is largely attributed to the higher spectral and radiometric fidelity of direct digital capture, but may also be attributed to more opti mal seasonality for the data of the ADAR acquisition. While the mapping was based upon interactive, visual image interpretation and on-screen digitizi ng the following image processing techniques proved to be helpful for aidin g interpretation: contrast enhancement prior to generating hardcopy prints for field analyses and during on-screen classification and digitizing, per-pixel image classification based on spectral-radiometric pattern recogn ition, and derivation of normalized difference vegetation index maps. The overall accuracy of interpreter-derived vegetation maps was approximate ly 75 percent for the entire study area.