1 August 2007 Generalized linear models for mapping land cover using satellite measurement and digital terrain data
Author Affiliations +
Abstract
This paper explores an approach for predicting land cover types of central Montana, USA based on satellite measurement and digital terrain data. We assume a non-linear inherent relationship existing between land cover types and Landsat TM reflectance and terrain variables. To measure this relationship we use Generalize Linear Models (GLMs), which are mathematical extensions of ordinary least-square regression models. Specifically, stepwise logistic regression technique is applied to optimize the predictive model. For the analysis of the significance of dropping or adding terms, the Akaike information criterion (AIC) is used. Likelihood Ratio Test (LRT) is applied to test the validity of explanative potential of predictor variables. We use cross-validation method to evaluate the predicative accuracy of land cover mapping using GLMs. Finally we table the relative risk ratios of GLMs. Since relative risk ratios explicitly represent the explanative efficiency of predictor variables, their ranking can pick up the variables with significant explanatory potential in discriminating land cover types, which will be significative for simplifying the predictive models. It is anticipated that GLMs will be valuable extension to semi-automatic classification of remotely sensed imagery, and an effective tool for land cover mapping.
© (2007) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Xiong Rao, Jinping Zhang, Brian M. Steele, Roland L. Redmond, "Generalized linear models for mapping land cover using satellite measurement and digital terrain data", Proc. SPIE 6751, Geoinformatics 2007: Cartographic Theory and Models, 675108 (1 August 2007); doi: 10.1117/12.759481; https://doi.org/10.1117/12.759481
PROCEEDINGS
11 PAGES


SHARE
Back to Top