Change Detection Of Forest Coverage In Sichuan Giant Panda Sanctuaries Using SAR Data

Currently,the sustainable development of World Heritage is facing challenges either from artificial destruction or natural disasters.The amount of World Heritage in China is listed as the Top 2 in the World,which is only less than Italian.Recently,the protection and monitoring of the World Heritage have been increasingly concerned in our country.Once World Natural Heritage gets damaged,forest is often the first element to be affected.Therefore,it is necessary and effective to monitor the change of forest coverage in the damaged World Natural Heritage.Majority of the Sichuan Giant Panda habitats,one of the World Natural Heritage,was selected as the study site in this thesis.This habitat plays a significant role in biodiversity.However,in May 2008,there was a giant earthquake happened in Wenchuan whose epicenter is close to the site.The earthquake complementary with secondary disasters induced a serious damage.Then,all kinds of rare animals and plants were affected due to the degradation of forests.Time-series PALSAR data were used in the classification model to extract forest covers and then for the change detection of the habitat.The corresponding research items were highlighted as follows:(1)Based on the reference high-resolution remote sensing images,samples information and the best combination of time-series information were extracted using PALSAR data,including TAI(?~0HV)? TSD(?~0HH)? TAI(?~0HH-HV).This combination was determined as the input parameters for the following classification.(2)Decision Tree(DT)algorithm was firstly applied for the forest classification in this study.The results showed time-series SAR products used in this algorithm is capable of extracting forest coverages with an overall accuracy of 94.31% and user accuracy of 97.68%.(3)As another classification method,Support Vector Machine(SVM)was used to extract the coverage of forest.Results showed that time-series SAR data is better in forest classification than single polarized data.The corresponding overall accuracy and user accuracy of this method were 86.90% and 82.34% respectively.(4)In order to monitor the change of forest effectively,these two classification models(DT&SVM)were compared in five aspects,including data resources,time-series information,performance,the difficult degree in the implementation of classification and interpretability.Results demonstrated that the DT model indicates a better performance.(5)Four forest/non-forest thematic maps were obtained by the DT model.Statistics results showed that there was a significant damage on forest caused by Wenchuan earthquake in the study site,which was 79.57km~2 in all with a damage rate of 6.78%.Fortunately,there was 10.53km~2 of forest recovered naturally and artificially in three years after the earthquake.This study confirms that time-series PALSAR data is potential in extracting the forest coverage of Giant Panda Sanctuaries in Sichuan,China.Wenchuan earthquake caused a serious damage in the deforestation and fortunately the forest was recovering after this giant earthquake.