Development of a global annual land surface phenology dataset for 1982–2018 from the AVHRR data by implementing multiple phenology retrieving methods

Abstract

The land surface phenology (LSP) associated with vegetation dynamics plays an important role in influencing the land surface processes and land–atmosphere interactions. Satellite observations have been widely used in studies for the monitoring of LSP across large areas based on different phenology retrieval algorithms and the routine production of LSP from remote sensing data has yet come to fruition. Here we used six phenology retrieval methods, including the amplitude threshold (AT), first-order derivative (FOD), second-order derivative (SOD), relative changing rate (RCR), third-order derivative (TOD), and curvature change rate (CCR), to retrieve the start of the growing season (SOS) and the end of the growing season (EOS) from the Advanced Very High Resolution Radiometer (AVHRR) data. We improved the curve fitting method to reduce uncertainties owing to data preprocessing. The results indicated that both SOS and EOS retrieved by six different methods had similar spatial distribution and the retrieved dates could vary largely at the pixel level. In the Northern Hemisphere, from 1982 to 2018, the trends of SOS retrieved vary across methods and only the EOS extracted by the relative change curvature method had a significant advanced trend. In the Southern Hemisphere, from 1982 to 2018, SOS results derived from four methods (i.e., AT, SOD, TOD, and CCR) showed significantly delayed trends, EOS results extracted by all the methods demonstrated insignificant trends. The phenology retrieval methods were assessed using the field observation data from the Pan European Phenology Project (PEP725) and from time series of leaf area index (LAI) measured at flux towers. The satellite-retrieved dates of both SOS and EOS were positively correlated with field observation and the relationships are largely dependent on how field phenology metrics are defined. We presented longer time series (1982–2018) data of phenology metrics with fewer gaps and multiple phenology retrieving methods as compared to the MODIS land cover dynamics product. Based on our assessments, one might use the SOS generated by FOD and the EOS generated by RCR as they provide results the most consistent with field data among all the tested methods. If studies aim to use the earliest SOS (or latest EOS) in a year, one might use the data retrieved based on TOD or CCR. The global dataset is delivered for uses in studies and applications associated with LSP.