Measurement and Classification of Long-Term Seasonal Trends: Procedures and Results using a 30-Year Monthly Series of Global NDVI

Abstract:

With the advent of long image time series as a part of the emerging earth observing system, it is now possible to examine the geographic distribution of trends in seasonal change. However, image time series of earth phenomena exhibit many forms of variability that detract from the signal of interest including cloud contamination, path radiance and short term inter- and intra-annual variability. To accommodate this, a special procedure called Seasonal Trend Analysis (STA) was developed that uses a combination of harmonic analysis and robust statistical trend measurement to reduce the effects of noise and short-term variability. A Contextual Mann-Kendall test is then used to test the trends associated with each of five harmonic shape parameters developed by STA. A three-class categorization of significance then leads to 243 possible classes of seasonal trend. In a study of a 30-year series of global monthly Normalized Difference Vegetation Index imagery (GIMMS NDVI3g, 1982-2011), it was found that over half (56.30%) of vegetated land surfaces exhibited significant trends in seasonality. Further, almost half (46.10%) of these significant trends belonged to just three classes of seasonal trends. Class 1 consisted of areas that experienced a uniform increase in NDVI throughout the year, and was primarily associated with forest biomes. Class 2 was predominantly associated with grassland and shrubland biomes and exhibited increases in the amplitude of the annual seasonal signal without a net annual change. Class 3 exhibited increases in the annual summer peak in NDVI and was found primarily in the Taiga and Tundra biomes. While no single attribution could be established for each of these classes, they point to a widespread greening of vegetated surfaces associated with climate-related ameliorations of growing conditions.