Lithological Discrimination Of Carbonate Rocks Covered By Vegetation Using Remote Sensing Data In Southwestern Karst Area, China

Southwestern China karst area covers an about 907,000 km2 area and mainly exposed carbonate rocks. Characteristics of karst area are that soil is difficult to develop and the karst entironment, if the vegetation was felled and destroyed, is very difficult to be resumed or renewed. Therefore, karst rocky desertification is often occur in karst area, causing serious environment issues and leading to progressive impoverishment of local residents. In addition to people's irrational activity, the speed or course of karst rocky desertification is specially influenced by the lithology of carbonate rocks. For example, karst rocky desertification is very easy to occur in pure limestone area. To prevent and recover from karst rocky desertification, it is necessary to know lithological distribution of carbonate rocks.Lithological mapping using remote sensing data is fast and cheap. Since remote sensing come out, a number of researches have been done in lithological mapping or lithological discrimination using remote sensing data in two aspects:First, spectroscopy, lithological discrimination by studying the spectral property of different rocks; Second, technology or method of information extracting. Many of these studies are focused in arid region with thin soil, little vegetation, and well-outcropped strata. However, in southwestern China karst area, rocks were covered by flourishing vegetation and deep soils. The spectral information of the remote sensing images mainly reflects vegetation and soils. So, lithological discrimination by using remote sensing becomes very difficult.In this thesis, the use of multi-source remote sensing data, such as TM, SPOT and ASTER, for lithological discrimination was evaluated in the major test area covered a 900 km2 karst area located in southeastern of Quanzhou County, Guangxi. The spectra of carbonate rocks were measured in field using portable spectrum meter. In the assistant study area, which is located in Donglan County, Guangxi and is a typical karst area too, the relationship among lithology, soil and vegetation was studied by measuring thire chemical element. The rock and its soil and vegetation was conside as a whole system in remote sensing image, lithology was classified by using remote sensing data in the major test area.Through research, following knowledge and conclusions can be reached:There is a close correlation betreen lithology, soil and vegetation. The elements migration follows certain patterns from Lithology to soil and soil to vegetation. Lithology, as the basement of soil and vegetation, play a "source" role in elements migration, which restricts the component and composition of soils elements and plant elements. Different spectral characteristics will be showed on remote sensing image because of different Soil and plant elemental composition.Based on the above understanding, in the area of the same lithology, the lithology, soil and vegetation were consided as a whole system in the present study, and thus the spectral information and texture information of different lithology unit were explored and extracted from remote sensing data under vegetation cover, and was used in lithology automatic identification and classification.The process and results of lithological classification indicate that the accuracy of lithological discrimination using single RS data was 69.36% for ASTER data,64.37% for TM data and 54.41% for SPOT data. The reason should due to different spectral range and band settings of each RS data.Because satisfying classification accuracy using single RS data can't be acquired, rational strategy is to use multi-sources RS data. Each RS data have its own character and bands setting. By combining of multi-sources RS data, those different characters of different kinds of data can reinforce each other, and supplementary information helped to increase accuracy of lithological discrimination may be obtained. For instance, ASTER data have more wide spectral range and more bands than SPOT data, but SPOT data have higher resolution. These two kinds of data reinforce each other in spectral range and resolution. Usually, the more sorts of RS data were used for classification, the higher accuracy was obtained.In classification, in addition to spectral bands, the inclusion of variogram texture images in spectral classification could considerably improves the classification accuracy. The key for pick-up texture image is the size of window. Two factors decided the size of window are:first, texture complexity on remote sensing image; secondly, the exposed width of lithological unit, especially the latter. For the lithological unit which is thin and narrow exposed on the surface, it is reasonable that the window size do not exceed the exposed width of the thin unit.Finally,4 SPOT spectral bands and its 4 texture images,6 TM spectral bands,14 ASTER spectral bands and its 3 texture images extracted from 3 ASTER VNIR spectral bands was used for classification. The final overall classification accuracy is 82.01%.