A Semi-Analytical Snow-Free Vegetation Index for Improving Estimation of Plant Phenology in Tundra and Grassland Ecosystems
Wei Yang,Hideki Kobayashi,Cong Wang,Miaogen Shen,Jin Chen,Bunkei Matsushit,Yanhong Tang,Yongwon Kim,M. Syndonia Bret-Harte,Donatella Zona,Walter Oechel,Akihiko Kondoh
DOI: https://doi.org/10.1016/j.rse.2019.03.028
IF: 13.5
2019-01-01
Remote Sensing of Environment
Abstract:Satellite monitoring of plant phonology in tundra and grassland ecosystems using conventional vegetation indices (VIs), such as the normalized difference vegetation index (NDVI), can be biased by effects of snow. Snow free VIs that take advantage of the shortwave infrared (SWIR) band have been proposed to overcome this problem, viz., the phonology index (PI) and the normalized difference phonology index (NDPI). However, the PI cannot properly capture the presence of sparse vegetation, and the NDPI does not account for the influence of dry vegetation. Here, we propose a novel snow-free VI, designated the normalized difference greenness index (NDGI), that uses reflectance in the green, red, and near-infrared (NIR) bands. The NDGI is a semi-analytical index based on a linear spectral mixture model and the spectral characteristics of vegetation, snow, soil, and dry grass. Its performance at estimating the start and end of the growing season (SOS and EOS) was evaluated using simulation datasets, time-lapse camera data at tundra sites, and flux tower gross primary production (GPP) data at grassland sites. Simulation results demonstrated that the NDGI can exclude the influence of snow on estimates of SOS and EOS. At the tundra sites, the NDGI markedly outperformed the NDVI, PI, NDPI, NIRv (near-infrared reflectance of vegetation), EVI2 (two-band enhanced vegetation index), PPI (plant phenology index), and DVI (difference vegetation index plus) for SOS estimation, with a root mean square error (RMSE) of 6.5 days and a Bias of 1.3 days, and for EOS estimation, with an RMSE of 8.3 days and a Bias of 0.11 days. At the grassland sites, the NDGI also outperformed the other VIs at SOS estimation, with an RMSE of 10.3 days and a Bias of 4.9 days. Although its performance was poorer at monitoring EOS than SOS at grassland (GPP) sites, its performance was comparable to that of the PI and superior to that of the other VIs at estimating EOS. These results indicate the potential of the NDGI for operational monitoring of plant phenology in tundra and grassland ecosystems based on satellite observations.