| Some specifically available remotely sensed images are usually selected to delineate lake area changes to reflect the influence of environment and human activities. Because of the dynamic changes of some lakes, the outcomes obtained from one image are not representative and cannot reflect the changes of lakes. In this study, 8-day Moderate Resolution Imaging Spectrometer (MODIS) composite data were used to extract 629 lakes in China during the period of 2000 and 2010. Through combining the automatic extraction of training data and Support Vector Machine (S VM) classifier, we obtained, for the first time, the spatial distribution of 629 large water bodies in China every 8 days during the period of 2000 and 2010. The producer’s and user’s accuracies (PA and UA) are 91.06% and 89.81% respectively for the whole China when compared with interpretation results from 30 m resolution Landsat images taken during similar days of MODIS images. Area changes, variability, inundation intensity and rain seasonality were analyzed based on this multi-temporal lake database. The 629 lakes have experienced an increase over the 11-year period, which attributes to the expansion of lake areas in Tibetan Plateau. The number of lakes whose maximum area exceeds 1000 km2 during 2000 and 2010 was 12, and 6 of them were shrinking. The shrinkages of Poyang Lake and Dongting Lake were -54.76 km2/a and -25.08 km2/a, respectively. Lakes in Tibetan Plateau, Northern Xinjiang, Northeast Inner Mongolia, and the Northeast region were with low variability, while lakes in Yangtze Plain, Southern Inner Mongolia, central Xinjiang fluctuated drastically. The inundation intensities of lakes in Tibetan Plateau, North Xinjiang, Heilongjiang, and Jilin increased. On the contrary, the inundation extents decreased in Central Xinjiang, Southern Tibet, Southern Inner Mongolia, Sichuan and Yangtze Plain.The major innovativeness of this work can be summarized in the following three aspects:1. The spatial distribution of 629 large water bodies in China was mapped for the first time at a time interval of 8 days. Different rules were set based on the diversity of water bodies and background features to collect training data. Then an SVM classifier was used to extract water bodies and this method was labor-efficient. The producer’s and user’s accuracies (PA and UA) were 91.06% and 89.81% for the whole China when compared with corresponding interpretation results from higher resolution remotely sensed data. The PA and UA were 93.03% and 92.86%, respectively, over Tibetan Plateau. The PA and UA were 88.13% and 85.39%, respectively, for the remaining lakes in China. These differences are caused mainly by that the shape of lakes in Tibetan Plateau are nearly round and can be easily distinguished from the background, while lakes in the Yangtze Plain are easily confused with mountain shadows, leading to low accuracies. The results indicate that the method used in this study is suitable for delineating major water bodies in large regions, with high frequency.2. The high frequency lake database can provide more accurate area information of lakes and can be a vital supplement to hydrological studies. Lake-area changes can be better analyzed from the prospective of seasonal and inter-annual scale in support of global change research. The establishment of China lakes database at 8-day temporal intervals in this study can be a baseline data for future research.3. This study represents an attempt of paradigm change from the general land cover mapping to high frequency specific land cover type mapping. This database makes it possible for the generation of a new land surface process model, which can be based on more accurate lake dynamic information, and has potential in improving hydrological simulations. |
PDF Full Text Request