A snow-free vegetation index for improved monitoring of vegetation spring green-up date in deciduous ecosystems

Abstract

© 2017 Elsevier Inc. Vegetative spring green-up date (GUD), an indicator of plants' sensitivity to climate change, exerts an important influence on biogeochemical cycles. Conventionally, large-scale monitoring of spring phenology is primarily detected by satellite-based vegetation indices (VIs), e.g. the Normalized Difference Vegetation Index (NDVI). However, these indices have long been criticized, as the derived GUD can be biased by snowmelt. To minimize the snowmelt effect in monitoring spring phenology, we developed a new index, Normalized Difference Phenology Index (NDPI), which is a 3-band VI, designed to best contrast vegetation from the background (i.e. soil and snow in this study) as well as to minimize the difference among the backgrounds. We examined the rigorousness of NDPI in three ways. First, we conducted mathematical simulations to show that NDPI is mathematically robust and performs superior to NDVI for differentiating vegetation from the background, theoretically justifying NDPI for spring phenology monitoring. Second, we applied NDPI using MODIS land surface reflectance products to real vegetative ecosystems of three in-situ PhenoCam sites. Our results show that, despite large snow cover in the winter and snowmelt process in the spring, the temporal trajectories of NDPI closely track the vegetation green-up events. Finally, we applied NDPI to 11 eddy-covariance tower sites, spanning large gradients in latitude and vegetation types in deciduous ecosystems, using the same MODIS products. Our results suggest that the GUD derived by using NDPI is consistent with daily gross primary production (GPP) derived GUD, with R (Spearman's correlation) = 0.93, Bias = 2.90 days, and RMSE (the root mean square error) = 7.75 days, which outcompetes the snow removed NDVI approach, with R = 0.90, Bias = 7.34 days, and RMSE = 10.91 days. We concluded that our newly-developed NDPI is robust to snowmelt effect and is a reliable approach for monitoring spring green-up in deciduous ecosystems.