Effect of Neighborhood Size on Morphometric Variables and Their Relationship with Vegetation Cover within Three Geomorphologically and Climatically Different Sub-Watersheds in Southwest Iran

The aim of this research was to study the importance of the neighborhood scale in modeling the relationship between vegetation cover and morphometric variables using the classification and regression trees algorithm (CART) in southwestern of Iran. For this purpose, the second Modified Soil-Adjusted Vegetation Index (MSAVI2) was calculated from a Landsat 8 image, and eight morphometric variables were derived using the Wood method in four neighborhood scales (90×90, 150×150, 210×210, and 270×270 m) from a 30 m SRTM digital elevation model. The results of the Kruskal-Wallis test confirmed that in some sub-watersheds, neighborhood-scale change can have a significant effect on slope gradient, profile curvature, specific catchment area, LS factor, and topographic wetness index. The results showed that in each sub-watershed different morphometric variables are most related to the spatial distribution of the MSAVI2 index and the value of the Spearman correlation coefficient between them is slightly affected by the neighborhood scale. CART models based on the MSAVI2 index and 270×270 m morphometric variables with a kappa coefficient of 0.55 and 0.78, respectively, had the best performance in classifying vegetation types. The elevation smoothed, which is the least affected by the neighborhood scale, was recognized as the most important predictor in the CART model. However upscaling led to the increasing importance of other morphometric variables, especially slope gradient, in classifying vegetation types and finally improving the accuracy of the CART model. Overall, the present results indicate that the application of multi-scale geomorphometric analysis with respect to the geomorphology of the study area can improve the performance of prediction models related to vegetation cover to an appropriate extent.