Keywords

land use, remote sensing, sensitivity, errors, logistic regression, abandonment

Start Date

1-7-2012 12:00 AM

Abstract

Land-use change (LUC) is a common process around the world and LUC models help elucidate LUC. Models are commonly parameterized with LUC maps derived from satellite imagery. However, such LUC maps have errors, and it is unclear how sensitive spatially explicit LUC models are to such errors. We studied the effects of errors maps on spatially explicit LUC logistic regression models of agricultural land abandonment within one Landsat footprint in Eastern Europe that covered the part of Lithuania. The selected footprint had six matching image dates (Spring, Summer and Fall) that were important to separate land-use classes for pre- (circa 1989) and post-abandonment (circa 2000). We simulated errors maps classifying all possible 49 sub-optimal image dates combinations with non-parametric support vector machines (SVM) classifier. We assessed the sensitivity of the spatially explicit LUC logistic models that had socio-economic and environmental variables to the mapping errors for the produced 49 LUC maps. When fewer image-dates combinations were used, the spatially explicit logistic regression LUC models were prone to the mapping errors. Results suggest avoiding using the classifications lower than 80% of individual class accuracy for the spatially explicit logistic regression models of agricultural land abandonment in Eastern Europe.

COinS
 
Jul 1st, 12:00 AM

Sensitivity of Spatially Explicit Land-use Logistic Regression Models to the Errors Land-use Change Maps

Land-use change (LUC) is a common process around the world and LUC models help elucidate LUC. Models are commonly parameterized with LUC maps derived from satellite imagery. However, such LUC maps have errors, and it is unclear how sensitive spatially explicit LUC models are to such errors. We studied the effects of errors maps on spatially explicit LUC logistic regression models of agricultural land abandonment within one Landsat footprint in Eastern Europe that covered the part of Lithuania. The selected footprint had six matching image dates (Spring, Summer and Fall) that were important to separate land-use classes for pre- (circa 1989) and post-abandonment (circa 2000). We simulated errors maps classifying all possible 49 sub-optimal image dates combinations with non-parametric support vector machines (SVM) classifier. We assessed the sensitivity of the spatially explicit LUC logistic models that had socio-economic and environmental variables to the mapping errors for the produced 49 LUC maps. When fewer image-dates combinations were used, the spatially explicit logistic regression LUC models were prone to the mapping errors. Results suggest avoiding using the classifications lower than 80% of individual class accuracy for the spatially explicit logistic regression models of agricultural land abandonment in Eastern Europe.