基于MODIS和Landsat的青藏高原两代GIMMS NDVI性能评价

由于AVHRR NDVI数据集本质上具有动态变化的特点,使得数据重叠时段(1981-2006年)的第1代GIMMS NDVIg(简称NDVIg)数据集与第3代GIMMS NDVI3g(简称NDVI3g)数据集也不完全相同。如何理解、对待这些差异,是综合利用两代数据集的研究成果、科学客观地评估地表植被历史变化状况、预测未来变化趋势、指导生态保护与建设工作的前提和基础。该文利用MODIS NDVI和Landsat数据,评估了青藏高原两代GIMMS数据集的性能,并对比分析了区域尺度和像元尺度两代数据集在监测植被长期动态变化方面的异同。结果表明,NDVI3g捕获植被物候变化的能力与NDVIg相当,但NDVI数值明显大于NDVIg,甚至大于MODIS NDVI;与NDVI3g相比,在NDVI分布格局和动态变化方面,NDVIg与MODIS NDVI和Landsat更为相似;尽管两代GIMMS数据集1982-2006年生长季NDVI变化趋势类似,但GIMMS NDVIg倾向于检测到更多的显著变化区域;夏季和秋季的结果与生长季类似,但春季GIMMS NDVI3g则检测到了更大范围的NDVI显著增加,与NDVIg结果的差异主要集中在青藏高原腹地。NDVI数据集是众多生态模型的基础数据,NDVI数据集之间的差异可能会导致模拟结果出现偏差。在使用NDVI数据之前,对NDVI数据的适用性进行评估,是获取更符合当地实际情况、更为客观真实结果的前提。

Performance evaluation of GIMMS NDVI3g and GIMMS NDVIg based on MODIS and Landsat in Tibetan Plateau

Abstract: GIMMS NDVI dataset must be re-calculated every time when New year''s data are added, due to the GIMMS NDVI data set is dynamic in nature, and this leads to differences between GIMMS NDVI3g and GIMMS NDVIg throughout their overlapping period (1981-2006). How to understand and treat these discrepancies is premise and basis for comprehensive utilizing the datasets of GIMMS NDVIg and GIMMS NDVI3g to scientifically detect vegetation''s historic variations, forecast its future tendency and guide the ecological protection and construction. With MODIS NDVI datasets from 2000 to 2012 and 495 Landsat samples of 20 km × 20 km from 2000 to 2006, performances of GIMMS NDVIg and GIMMS NDVI3g were evaluated during the period from 2000 to 2006, and long-term variation of vegetation monitoring used both GIMMS datasets during 1982-2006 were compared and analyzed in this paper. Firstly, absolute values of GIMMS NDVIg, GIMMS NDVI3g and MODIS NDVI with Landsat NDVI were compared. Then, the differences between a Landsat sample-pair (i.e., two 20 × 20 km2 Landsat samples acquired for the same location at different years) and GIMMS NDVIg, GIMMS NDVI3g and MODIS NDVI datasets at the same time points were evaluated. Besides, GIMMS NDVIg and GIMMS NDVI3g with MODIS NDVI during 2000-2006 in term of temporal trends by applying a simple linear regression model based monthly anomalies and the seasonal Mann-Kendall trend test were compared at region, and correlations were conducted at pixel scales. Finally, trends of GIMMS NDVIg with that of GIMMS NDVI3g in three seasons (spring, summer and autumn) and growing season during 1982-2006 were compared at region and pixel scales. The results showed that almost equal capability of capturing variations of seasonal and monthly phenology for both GIMMS datasets was found. The NDVI value of GIMMS NDVI3g was generally larger than that of GIMMS NDVIg, or even larger than NDVI of MODIS NDVI. Compared with GIMMS NDVI3g, patterns and trends of GIMMS NDVIg were more similar to that of MODIS NDVI and Landsat. Although spatial patterns of GIMMS NDVI3g change in growing season during 1982-2006 resembled that of GIMMS NDVIg, wider range characterized significant increase in spring NDVI were detected with GIMMS NDVIg, and the discrepancies between both GIMMS NDVI datasets mainly concentrated in the hinterland of the Tibetan Plateau. The increase trend of vegetation growth in spring using GIMMS NDVIg was more severe than that using GIMMS NDVI3g, but the opposite situation was found in summer. The remarkable difference of NDVI variation in spring may lead to differences in the analysis of the phenology using both GIMMS NDVI datasets of the Qinghai Tibet Plateau. Long-term NDVI datasets are the basic data for many ecological models, the differences among these datasets may influence the accuracy of model results. Before conducting the relevant research using NDVI datasets, the applicability of NDVI datasets is needed to evaluate, and it is the premise to obtain more consistent with the actual situation objective results. Combined with other ecological datasets, such as vegetation coverage fraction, leaf area and vegetation production of historical field data is important to identify the similarities and differences between the two GIMMS NDVI datasets and establish a connection between them for reasonably monitoring vegetation dynamics.