Detecting interannual variation in deciduous broadleaf forest phenology using Landsat TM/ETM + data

Abstract

Observations of vegetation phenology provide valuable information regarding ecosystem responses to climate variability and change. Phenology is also a first-order control on terrestrial carbon and energy budgets, and remotely sensed observations of phenology are often used to parameterize seasonal vegetation dynamics in ecosystem models. Current land surface phenology products are only available at moderate spatial resolution and possess considerable uncertainty. Higher resolution products that resolve finer spatial detail are therefore needed. A need also exists for data sets and methods that link ground-based observations of phenology to moderate resolution land surface phenology products. Data from the Landsat TM and ETM+ sensors have the potential to meet these needs, but have been largely unexplored by the phenology research community. In this paper we present a method for characterizing both long-term average and interannual dynamics in the phenology of temperate deciduous broadleaf forests using multi-decadal time series of Landsat TM/ETM+ images. Results show that spring and autumn phenological transition dates estimated from Landsat data agree closely with in-situ measurements of phenology collected at the Harvard Forest in central Massachusetts, and that Landsat-derived estimates for the start and end of the growing season in Southern New England varied by as much as 4weeks over the 30-year record of Landsat images. Application of this method over larger scales has the potential to provide valuable information related to landscape-scale patterns and long term dynamics in phenology, and for bridging the gap between in-situ phenological measurements collected at local scales and land surface phenology metrics derived from moderate spatial resolution of instruments such as MODIS and AVHRR.