Establishing The Continuous Spatial Scaling Of NDVI Based On Fractal Theory

Spatial scale transformation is one of the basic and important scientific problems in quantitative remote sensing field. Particularly, spatial up-scaling has drawn much attention as it can provide an effective way to solve the difficult problems, e.g. validation of quantitative remote sensing products. However currently there are still some issues concerning the research on spatial up-scaling:1) Based on discrete images from different sensors, transformation formula on several scales can be established, but its scale range is very limited; 2) The lack of reasonable retrieved physical models hampered the development of up-scaling based on them. As an important retrieval, the up-scaling of NDVI also faces these two issues.Based on fractal theory, this research established the continuous spatial scaling model (CSSM) of NDVI to address the above issues. This model was able to quantitatively describe transformation relationships of NDVI on continuous scales. And based on that, NDVI’s validation of different low-resolution images could be implemented rapidly and effectively. Shatian Byland (Beihai City, Guangxi Zhuang Autonomous Region) was taken as experimental area because of its variety of ground objects and high spatial heterogeneity. And Landsat ETM+, GEOEYE-1 and HJ-1B CCD1 images were comprehensively utilized. The systematic research of establishing NDVI’s CSSM was conducted from three aspects:1) establishment of NDVI’s CSSM based on fractal theory,2) analysis of the impacts of different image’s characteristics on determining the most reasonable scale-hierarchy (total scale number, Level) when establishing NDVI’s CSSM, and 3) validation of the application of NDVI’s CSSM.First, NDVI’s CSSM was established based on fractal theory and its availability in validation was demonstrated. Utilizing ETM+ image of Shatian Byland, NDVI’s CSSM was established on different Levels (Level=33,66,100,150, 200,250). Integrating the statistical estimation indexes (r, p, rlo, rup) and taking the conditions (biggest r, p...