ASTER-Based Vegetation Map to Improve Soil Modeling in Remote Areas

Abstract

Soil scientists are using spectral data from vegetation to help predict soils for the North Cascades National Park (NCNP) Complex in Washington State, USA. Vegetation is a proxy indicator for the soil forming factors organisms and climate. Two objectives of this research are: (1) can spectral remote sensing data be used to accurately map vegetation in the NCNP; and (2) are soil characteristics and properties significantly correlated with vegetation cover? We characterized soil profiles and vegetation cover at 70 sites within the 30,000 ha Thunder Creek Watershed (TCW) in the summer of 2007. These ground truth locations were used to manually interpret 1 m resolution National Agricultural Imagery Program (NAIP) color photography and classify 700 randomly selected validation locations as coniferous forest, shrub, meadow, heather, rock and snow. We applied Spectral Angle Mapper (ENVI version 4.3, ITT Visual Information Solutions; Boulder, CO) to July, 2007 ASTER data to classify vegetation cover for the TCW, with 3600 ground truth pixels and the 700 NAIP photograph pixels employed for calibration and validation respectively. This classification yielded an overall accuracy of 67% with coniferous forest having a producer’s and user’s accuracies of 84 and 91% respectively.