A simple and robust method for remote sensing of phenology using green and red reflectance: GRVI-method

Remote sensing of vegatation phenology is important for a regional scale and long-term assessment of the climate change effects on terrestrial ecosystem. Various methods have been already developed, and many of them use time-series NDVI (Normalized Difference Vegetation Index) calculated from visible red and near-infrared reflectance to detect the timing of phenological events. However, the NDVI-methods have some drawbacks: (1) it is impossible to specify the NDVI threshold to detect the autumn phenology, and (2) some NDVI-methods are affected by snow and water on the ground (Nagai et al., submitted). In the study, we propose a new phenology detection method: "GRVI-method". GRVI (Green and Red ratio Vegetation Index) is one of the spectral vegetation indices calculated from visible green and red reflectance. GRVI-method uses time-series GRVI instead of NDVI, and determine the timing of green-up and autumn-coloring with“GRVI=0” as the threshold. To check the effectivity of GRVI-method, we conducted the continuous ground-observation about spectral reflectance and webcam images at some ecosystems in Japan (PEN: Phenological Eyes Network; http://www.pheno-eye.org/). The ground datasets showed GRVI-method has the following advantages compared with other methods: (1) GRVI-method is robust to misleading signals due to snow and water on the ground surface; (2) GRVI-method can detect not only spring green-up but also autumn leaf-coloring. Next, we applied GRVI-method to Terra MODIS datasets of Japan in 2001-2008. Daily webcam images taken at various places were used to evaluate the results. The result was compared with the NDVI-midpoint-method (White et al., 1997) and the MODIS phenology product distributed by NASA (MOD12Q2). As a result, GRVI-method showed consistent results with the actual timing of green-up and autumn-coloring observed by webcam images. NDVI-midpoint-method was not able to detect the autumn leaf-coloring in most parts of Japan. MOD12Q2 product showed the timing of leaf-development earlier than the actual and the timing of leaf-coloring later than the actual. Moreover, there are a lot of no-data pixels in the MOD12Q2 product, especially in snowy areas.