Prediction and comparison of urban growth by land suitability index mapping using GIS and RS in South Korea

Abstract

This study compares land suitability index (LSI) maps created using a geographic information system (GIS) with frequency ratio (FR), analytical hierarchy process (AHP), logistic regression (LR), and artificial neural network (ANN) approaches to forecasting urban land-use changes. Various social, political, topographic, and geographic factors were used as predictors of land-use change, including elevation, slope, aspect, distance from roads and urban areas, road ratio, land use, environmental score, and legal restrictions. Then, LSI maps were created using FR, AHP, LR, and ANN approaches, and significance and correlation were examined among the models using relative operating characteristic (ROC), overall accuracy, and kappa analyses. The ROC analyses gave results of 0.940, 0.937, 0.922, and 0.891 for the LR, FR, AHP, and ANN LSI maps, respectively. The highest correlation was found between the LR and AHP LSI maps (0.816911), and the lowest correlation was between the ANN and FR LSI maps (0.759701). The ANN approach produced the highest overall accuracy at 92.3%, followed by 91.74% for FR, 89.12% for AHP, and 88.93% for LR. In the kappa analysis, the highest K ˆ statistic was 45.38% for FR, followed by 40.84% for ANN, 30 representing the city area, the ANN method had a relatively high value of 71.71%, and the FR, LR, and AHP methods had similar accuracies of 57.68, 55.05, and 54.31%, respectively. These results indicate that the FR, AHP, LR, and ANN approaches produced similar LSI maps for Korea.