ASSESSING GARP MODELING AND EFFECT OF PLANT SAMPLE POSITION ON PREDICTING SUITABLE HABITAT OF Brainea insignis

摘要

Forestry has begun to use 3S technologies (RS, GIS, and GPS) in routine inventory work and scientific research. Modeling ecological pattern of species needs to utilize the combination of 3S technologies and statistics, and it has become an important part in ecology. The study was intended to predict the suitable habitat of cycad-fern in the Huisun Forest Station by using multivariate statistics coupled with a GIS. The ecological pattern of the species was examined by overlaying the layer of cycad-fern samples collected with GPS on the layers of topographic variables and vegetation index derived from SPOT-5 images. We also combined easting and northing coordinates of grid cell with topographic variables to improve the accuracy of predictive models. Three models, decision tree (DT), Genetic Algorithm for Rule-set Prediction (GARP), and discriminant analysis (DA), were developed and validated. Accuracy assessment results indicated that the accuracies of DT with easting and northing added respectively were much greater than those of both GARP and DA with easting and northing added respectively, and GARP with easting and northing added respectively was also better than DA. More importantly, the accuracies of DT with easting and northing added respectively were greatly improved, those of DA with easting and/or northing added respectively were slightly improved, and the opposite was true with GARP. Easting was more effective than northing in improving model accuracy because of east-west distribution with cycad-ferns in the Kuan-Dau watershed of the Huisun area. However, easting and northing predictor variables were found to limit the ability of spatial extrapolation with models and make predictive results look more artificial. We shall attempt to incorporate predictor variables extracted from high spatial resolution, hyperspectral data into models to improve the ability of spatial extrapolation with models in a follow-up study.