Misestimation of Growing Season Length Due to Inaccurate Construction of Satellite Vegetation Index Time Series

Abstract

Satellite-based vegetation index (VI) is widely used in monitoring land surface phenology (LSP). Currently, well-developed VI products utilize the maximum value composite (MVC) algorithm to produce a composite VI time series (TS). Many of these products, however, lack the actual acquisition date (AD) of the VI value. As an alternative, the median or mean date of a composite period is used to reconstruct the VI TS, which might lead to bias on LSP detection. This letter quantifies the LSP bias in the Northern Hemisphere by generating a 15-day composited normalized difference vegetation index (NDVI) TS from the land long-term data record daily NDVI products using the MVC method. The results show that the AD of the NDVI value is usually later than the mean date of a composite period in spring and earlier in fall, effectively leading to a total overestimation of the growing season length of 5.91 days on average across the Northern Hemisphere (north of 30° N). This bias has a significant spatial pattern with high values observed in Northeastern China, Central North America, and high-latitude areas. However, the temporal trend is not largely influenced overall. Accordingly, we suggest the research community using accurate temporal information, whenever possible, in extracting LSP from VI TS.