青藏高原湖泊面积变化在反映气候变化方面具有非常重要的意义。通常的阈值法提取湖泊面积由于受到冻湖、山体阴影和冰雪的影响,进行大范围提取时会存在阈值选择的不确定性和较难实现自动化提取等问题。基于归一化差异水体指数NDWI(Normalized Difference Water Index)、数字高程数据和雪盖指数NDSI(Normalized Difference Snow Index),提出了针对冻湖的湖泊面积插补迭代自动提取方法。结果表明,此方法能够实现非冻湖和冻湖阈值的自动确定,不仅可以在大范围内准确地完成湖泊面积提取,而且能减少山体阴影、冰雪的干扰。该方法相较于传统的基于水体指数的阈值法,具有更高的分类精度。
英文摘要:
Lakes on Tibetan Plateau not only are sensitive indicators of climate change but supply fresh water for people,livestock and agriculture.Therefore it is significant to extract lakes area on Tibetan Plateau and strengthen the understanding of lakes area variation trend under the influence of global warming.However, because of the complex terrain and bad weather conditions on Tibetan Plateau,it is difficult to determine threshold values automatically when using traditional threshold based methods to extract lakes areas for large scale application.This paper proposed a iterative interpolation method to automatically extract the boundary of lake based on NDWI(Normalized Difference Water Index),Digital Elevation Model(DEM) and NDSI(Normalized Difference Snow Index).This method not only can determine threshold of frozen lake and non-frozen lake automatically,but can reduce the interference caused by hillshade and snow.The proposed method and another two methods were applied on the experimental region on Tibetan Plateau and four typical areas which are glacier,frozen lake,non-frozen lake and hillshade in experimental region were selected to evaluate the accuracy of each method.Six Index with mapping accuracy,user accuracy,misclassification errors,omission errors,total errors and Kappa coefficient were used to show the performance of the three methods.The results show Kappa coefficient of the proposed method in this paper reached 0.90which is much higher than the other two methods of 0.83and 0.77;besides,the total errors of classification on water body and land are lower than that of the other two methods.Good performances of the proposed method make it become promising method to be applied in complex and wide regions.