Global trends in vegetation seasonality in the GIMMS NDVI3g and their robustness

Analysing changes in vegetation seasonality of terrestrial ecosystems is important to understand ecological responses to global change. Based on over three decades of observations by the series of Advanced Very High Resolution Radiometer (AVHRR) sensors, the Global Inventory Modelling and Mapping Studies (GIMMS) Normalized Difference Vegetation Index (NDVI) dataset has been widely used for monitoring vegetation trends. However, it is not well known how robust long-term trends in vegetation seasonality derived from GIMMS NDVI are, given inevitable influences from sensor and processing artefacts. Here we analyse long-term seasonality trends in the GIMMS third generation (NDVI3g) record (1982–2013). Changes in vegetation seasonality are decomposed into changes in duration (related to growing season length) and timing (related to peak growing season). We compare seasonality trends from the previous version (NDVI3g v0) with those in the subsequently released version (NDVI3g v1) and, for their common period, with those derived from MODerate Resolution Imaging Spectroradiometer (MODIS) collection 6 NDVI. We find that NDVI3g v0 shows marked seasonality trends for 1982–2013 over more than one-third of the global vegetated area. Long-term trends based on v1 are generally consistent with v0, but v1 shows a strong trend towards earlier timing across the Arctic regions that is absent in v0. NDVI3g v0, v1, and MODIS all point towards an increased duration across the tundra of North Asia and later timing across North Africa. However, several discrepancies are also found between the NDVI datasets. For example, for the North-American tundra, MODIS shows earlier and v0 later timing, while MODIS shows an increased duration and v1 a reduced duration. For North Africa, v0 and v1 exhibit a reduced duration that is absent in MODIS. We conclude that both the primary observations and the subsequent processing can have a marked influence on inferred seasonality trends, and propose that the robustness of trends should be examined and corroborated using alternative data sources wherever possible.