摘要:
长时序归一化植被指数(Normalized Difference Vegetation Index,NDVI)对植被生长监测等研究至关重要,但当前NDVI数据集存在时序覆盖和分辨率难以兼顾的问题,极大的限制了其应用。针对该问题,设计多流程处理框架融合AVHRR和MODIS数据,通过时域滤波、辐射归一化、时空融合、残差校正等处理,消除数据噪声、传感器与分辨率差异的影响,生成长江流域/长江经济带1982—2020年250米16天的时空连续NDVI产品,并开展区域植被时空变化分析。结果表明,融合产品精度较高,年内、年际精度变化稳定,平均相关系数达到0.87,且与植被覆盖率呈一定正相关。基于该产品,分析发现近40年研究区植被覆盖整体上呈缓慢增长趋势,各省市NDVI均值均有增长且增速与年度均值成正相关,除灌木外其余植被年度NDVI也都波动增长。
Abstract:
Objectives: Normalized Difference Vegetation Index (NDVI) is a key parameter for characterizing vegetation distribution and dynamics. NDVI time series can directly reflect the variations of large-scale vegetation, which makes it widely used in ecological and environmental studies. Over the years, numerous satellites capable of acquiring NDVI time series have been launched, producing a range of related products. However, individual NDVI datasets often face trade-offs between temporal coverage and spatial resolution, limiting their applicability in regional studies requiring both long-time coverage and high spatial detail. Methods: To address these challenges, a multi-step processing framework was developed to combine the respective advantages of MODIS and AVHRR GIMMS3g products, incorporating temporal filtering, radiometric normalization, spatiotemporal fusion, and residual correction to improve data quality, reconcile sensor differences, enhance spatial resolution, and correct temporal changes. Firstly, temporal filtering is applied to both GIMMS3g and MODIS datasets to remove noise and generate seamless NDVI time series, and then radiometric normalization is performed pixel-by-pixel on AVHRR data using MODIS data as a reference to account for sensor differences. A spatiotemporal fusion method is subsequently employed to integrate the complementary strengths of AVHRR and MODIS, i.e., the longer temporal coverage of AVHRR and the finer spatial resolution of MODIS, to enhance the spatial resolution of early AVHRR products. Finally, residual correction is applied to regions with significant surface changes over the past 40 years, resulting in a longterm NDVI dataset at 250-meter spatial resolution covering the period from 1982 to 2020. Results: The fusion result achieves a spatial resolution consistent with MODIS, with smoother spatial characteristics compared to original MODIS observations. Spatially, in 93% of the study area, the correlation coefficient (r) between the fused dataset and MODIS observations exceeds 0.7, while 68% of regions exhibit a mean absolute deviation (MAD) below 0.05. Temporally, the r values between the fusion results and true MOIDS data exceed 0.85, and MAD values remain below 0.1, underscoring the high reliability of the dataset for spatiotemporal vegetation analysis. The accuracy can also show stable inter-annual and intra-annual variations for different regions, indicating a reliable spatiotemporal variation of the fusion products. Conclusions: Based on this product, the analysis reveals that the average NDVI in the study area has increased from 0.52 to 0.60, indicating a gradual improvement in vegetation coverage over the past 40 years. Vegetation coverage in the Yangtze River Basin and Economic Belt has remained largely stable over the past four decades, with an overall slow growth trend. Mean NDVI values across all provinces have increased, with positive correlations with annual NDVI averages; all vegetation types except shrubs have also shown fluctuating increases in annual NDVI values.