Generating a 40-years AVHRR NDVI Composite Time Series forEurope at 1 km resolution

Kurzfassung

Image compositions from remote sensing data are crucial for a variety of scientific applications, such as the analysis of multi-decadal time series. The Advanced Very High Resolution Radiometer (AVHRR) instrument has been providing daily data with a spatial resolution of 1 km since the early 1980s, enabling analyses of environmental processes associated with climate change. The Normalized Difference Vegetation Index (NDVI) is the most commonly used parameter for monitoring vegetation conditions. In particular, the availability of multi-decadal NDVI data with high temporal resolution, such as AVHRR-derived products, provides a unique opportunity to investigate seasonal and interannual variations in vegetation. However, to enable such analyses, a consistent NDVI time series over the AVHRR time span needs to be created. The aim of this study is hence to thoroughly evaluate the effects of different compositional procedures on AVHRR NDVI composites, as no standard procedure exists to date. The project TIMELINE (TIMe Series Processing of Medium Resolution Earth Observation Data assessing Long-Term Dynamics In our Natural Environment) of the German Remote Sensing Data Center (DFD) of the German Aerospace Center (DLR) aims to generate a homogeneous multi-decadal time series from AVHRR/1, AVHRR/2 and AVHRR/3 data. The higher-level NDVI (L3) is calculated from the data of the red AVHRR channel and the near-infrared (NIR) channel of the L2c Surface Directional Reflectance (SDR) products. A comprehensive range of thirteen different compositing methods were implemented, including single and multi-criteria approaches utilizing e.g. maximum band or index values, the geometry of the scene and quality information related to pixel reflectance. Also, the approach used in the standard Moderate Resolution Imaging Spectroradiometer (MODIS) product is recreated as a benchmark. Daily, decadal and monthly composites over Europe and North Africa were calculated for the year 2007, and the resulting datasets were thoroughly evaluated against six criteria: NDVI value distributions, spatial and temporal consistency, local coverage homogeneity, and spatial and temporal agreement with MODIS NDVI composite products. The analyses showed that algorithms that only consider the visibility and illumination conditions tend to introduce strong geometric artifacts into the NDVI maps. Similarly, the uncertainty information available for the SDR product used was not in itself a sufficient selection criterion and led to unstable results. On the other hand, approaches that rely exclusively or at least 50 % on the NDVI (such as maximum value compositing) often lead to an overestimation of NDVI and to saturated pixels. The median approach, in which the median value of all NDVI observations is selected, proved to be an exception to this rule due to its inherent balancing effect and robustness against geometric artifacts. Consequently, the MED approach was selected as the most powerful compositing algorithm considering all aspects investigated. However, the combination of NDVI value and acquisition angles as a criterion for best pixel selection also proved to be a promising approach. The generated TIMELINE NDVI time series, currently covering the period from 1981 to 2018 , showed consistent behavior and close agreement with the standard MODIS NDVI product, which could allow a smooth integration of both datasets. The TIMELINE L3 NDVI products are gridded and projected onto LAEA-ETRS89. The data are provided as netCDF files including six layers: (i) the NDVI for the composite period; (ii–iii) the Julian day and time of acquisition of the selected observation; and (iv–vi) layers that enable the user to further evaluate the reliability of each NDVI value, namely, a quality layer with information on atmospheric correction and angular information, a layer with the variance of all suitable NDVI values, and a layer with the number of available observations. The analyses performed show the enormous influence of a compositing process on the resulting composites, especially when longer compositing intervals are considered. Furthermore, the combination of NDVI value and geometry conditions as selection criteria for the NDVI map composition was also shown to be a promising approach that could be further refined in the future. Future work could therefore focus on improving the quality filtering of the input data, on fine-tuning the scores assigned to angles of coverage or NDVI value and, most importantly, on extending the assessment to additional years, especially the early 1980s, when far fewer observations are usually available for compositing . Furthermore, the applicability of the generated time series for vegetation anomaly monitoring and phenological analyses will be demonstrated.