Using Landscape Spatial Relationships to Improve Estimates of Land-Cover Area from Coarse Resolution Remote Sensing

Abstract

A two-stage modeling strategy significantly improves land-cover area estimates from low spatial resolution remote sensing by correcting measurements of class proportions within large blocks of pixels. Vegetation class-type information is developed through supervised classification of Thematic Mapper spectral data at both fine (30 m) and coarse (1020 m) resolutions. Stage 1 models use measurements of landscape spatial properties to estimate the slopes and intercepts of proportion transition relationships between fine- and coarse-resolution classes within randomly located pixel blocks. Following this step, a Stage II model uses a linear estimator to predict true class proportions based on measured coarse-scale proportions and the slope and intercept estimates from the Stage I models. Model development and testing on a training site is followed by testing and inversion for a validation site. Model inversion involves using spatial variables measured only at the coarse resolution as input to the Stage I models. For both the training and the validation data, the procedure results in a statistically significant reduction in error when estimating land-cover area by class type within the sampling blocks.