Mapping and analysis of montane rain forest habitats using Landsat TM and elevation data with a geographic information system

Tropical species diversity challenges the use of remote sensing. Community-level assemblages in the Luquillo Experimental Forest (LEF) were mapped using Landsat TM and elevation data together. This method is less subjective than applying an elevation mask to classified spectral data. The vegetation map agreed with a map of the LEF produced in the 1940's using air photos and extensive ground truthing. The new map recognized ecotones not found on the older map. This technique is for mapping remote tropical or Caribbean forests.;GIS literature lacks rigorous quantitative statistical analyses. Using a GIS and statistical software an empirical model was developed to calculate the probability of deforested areas and seven forest types. The categorical nature of the dependent variable violates assumptions of least squares regression. Factor analysis, PCA, and discriminant analysis also require continuous dependent variables.;Logistic regression is probability based and was found to be a very good predictor of vegetation types when elevation, slope, and aspect were provided. Caution is required for statistical analysis of GIS data sets which tend to be extremely large. These data can have statistical significance due to their large size but no biological significance.;Image processing methods usually applied to spectral data were applied to topographic data. The result is a map showing similar conditions of elevation, slope, and aspect grouped together and distinguishing areas which are different. This was a new way of predicting locations of habitats which could be used by vegetation. The potential habitat map was compared to the vegetation map for patterns of agreement between the two maps.;The habitat map was processed with spatial filters of various sizes. The filters searched every pixel in the habitat map and recorded habitat diversity surrounding each pixel. A map of landscape diversity for each size of spatial filter resulted. Palms were associated with most diverse regions.