Applying spatial autocorrelation and logistic regression to analyze land cover change trajectory in a forested watershed

  • Landslide patches had higher positive spatial autocorrelation as regional hotspots
  • The most significant transformation of land cover was from forest to landslide
  • Lithology was the most important spatial determinant for the change trajectories
Abstract

Taiwan is more susceptible to earthquakes and typhoons because of the unique geographical position in the Northwestern Pacific Rim. It is necessary to understand the effects of such frequent natural disturbances on land cover change for watershed management and disaster mitigation. This study applied both global and local Moran’s I statistics to analyze the spatial autocorrelation of landslide patches in a natural disturbed watershed in eastern Taiwan. The land cover maps extracted from FORMOSAT-2 satellite images acquired in 2005, 2008, and 2011. A logistic regression model validated to predict occurrence probability of change trajectory. Results showed that spatial pattern of homogeneous landslide patches presented on small scales; while heterogeneous landslide pattern was on larger scales. Landslide patches had higher positive spatial autocorrelation and indicated that as regional hotspots in study area. After the trajectory calculation and classification, two unchanged and five changed trajectories dominated the study area. The most significant transformation of land cover was from forest to landslide and channel. In addition, a fittest logistic regression model predicted the occurrence probability of change trajectory in the study area. Among these environmental variables for logistic regression, lithology was the most important spatial determinant for the change trajectories. Curvature and aspect variables were also significant. This spatial statistical model was helpful for predicting the occurrence probabilities of the change trajectories.

Read 159 times