Publications

Wang, C; Chen, J; Tang, YH; Black, TA; Zhu, K (2018). A Novel Method for Removing Snow Melting-Induced Fluctuation in GIMMS NDVI3g Data for Vegetation Phenology Monitoring: A Case Study in Deciduous Forests of North America. IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, 11(3), 800-807.

Abstract
The normalized difference vegetation index (NDVI) has been widely used in recent decades to monitor vegetation phenology. However, interference from snow cover introduces a high degree of uncertainty in interpreting NDVI fluctuation, because snow melting increases NDVI value in a manner similar to vegetation growth, leading to false detection. In this study, we present a novel methodology to smooth out data noise caused by snow in the third generation NDVI dataset from Global Inventory Modeling and Mapping Studies (GIMMS NDVI3g). This method is developed to replace small values with a pixel-specific snow-free background NDVI estimate, based on the assumption that the existence of snow decrease NDVI value and the patterns of NDVI fluctuation after snow melting and that after initiation of vegetation growth are different. Using the daily gross primary production (GPP) data of 111 site-years from FLUXNET in nine North American sites and the GIMMS NDVI3g dataset, we found that the green-up onset day (GUD) derived from raw NDVI is 42.2 days earlier than that of GPP, on average. This difference decreases to 4.7 days when applying the newly developed method. Additionally, the root mean square error and Spearman's correlation coefficient between NDVI-derived GUD and GPP-derived GUD are improved from 46.8 to 12.8 days and 0.22 to 0.64, respectively. Our results indicate that this method could effectively improve the ability to monitor the vegetation phenology by NDVI time series in areas with seasonal snow cover.

DOI:
10.1109/JSTARS.2017.2778076

ISSN:
1939-1404