Authors: Santosh Rijal*, Virginia Polytechnic Institute & State University, Qing Wang, Southern Illinois University Carbondale, Guangxing Wang, Southern Illinois University Carbondale, Justin Schoof, Southern Illinois University Carbondale
Topics: Remote Sensing, Geographic Information Science and Systems, Land Use and Land Cover Change
Keywords: Spectral variables, Military training, Land condition prediction, Regression models
Session Type: Paper
Presentation File: No File Uploaded
Monitoring and predicting the changes in military land condition is critical for sustainable training mission and army readiness. In this study, a novel methodology that integrated geographically weighted regression with logistic regression (GWLR) was developed to predict the land condition of Fort Riley Military Installation (FR) and the results were compared to three other regression models i) linear stepwise regression (LSR) ii) logistic regression (LR), and iii) geographically weighted regression (GWR). Three different years i.e. 1990, 1997, and 2001 were chosen for comparing the prediction results. Landsat 5 TM data, training intensity data, and DEM were used to calculate 157 independent spectral variables. The correlation of these variables with field-based disturbance intensity data collected through range and training land assessment (RTLA) program were calculated. The variables that highly correlated with disturbance intensity but had low variance inflation factor (VIF) selected from stepwise regression method were used in the regression model. The coefficient of determination (R2) and root mean square error (RMSE) were calculated for each regression model. Results showed that i) the spatial distribution of disturbance intensity was well demonstrated by all the regression models with higher disturbance in the northwest, and central west of the installation iii) GWR showed a promising method among all the regression model. Local variability based GWLR was able to capture the spatial pattern of disturbance but didn’t necessarily improve the estimation of land condition as compared to other global models.