Variability and evolution of global land surface phenology over the past three decades (1982-2012)

Monitoring land surface phenology (LSP) is important for understanding both the responses and feedbacks of ecosystems to the climate system, and for representing these accurately in terrestrial biosphere models. Moreover, by shedding light on phenological trends at a variety of scales, LSP provides the potential to fill the gap between traditional phenological (field) observations and the large-scale view of global models. In this study, we review and evaluate the variability and evolution of satellite-derived growing season length (GSL) globally and over the past three decades. We used the longest continuous record of Normalized Difference Vegetation Index data available to date at global scale to derive LSP metrics consistently over all vegetated land areas and for the period 1982-2012. We tested GSL, start- and end-of-season metrics (SOS and EOS, respectively) for linear trends as well as for significant trend shifts over the study period. We evaluated trends using global environmental stratification information in place of commonly used land cover maps to avoid circular findings. Our results confirmed an average lengthening of the growing season globally during 1982-2012 - averaging 0.22-0.34daysyr(-1), but with spatially heterogeneous trends. About 13-19% of global land areas displayed significant GSL change, and over 30% of trends occurred in the boreal/alpine biome of the Northern Hemisphere, which showed diverging GSL evolution over the past three decades. Within this biome, the Cold and Mesic' environmental zone appeared as an LSP change hotspot. We also examined the relative contribution of SOS and EOS to the overall changes, finding that EOS trends were generally stronger and more prevalent than SOS trends. These findings constitute a step towards the identification of large-scale phenological drivers of vegetated land surfaces, necessary for improving phenological representation in terrestrial biosphere models.