Estimating 36 Years of Moderate Resolution Land Surface Phenology using a Bayesian Hierarchical Model

Abstract:

Land surface phenology (LSP) is a consistent, sensitive, and integrative indicator of climate change effects on Earth’s vegetation. Existing methods of estimating LSP require time series densities that, until recently, have only been available from coarse spatial resolution imagery such as MODIS (250 m and 500 m) and AVHRR (1 km). LSP products from these datasets have improved our understanding of phenological change at the global scale, especially over the MODIS era (2001-present). Nevertheless, these products may obscure important finer scale spatial patterns and longer-term changes. Therefore, we have developed a Bayesian hierarchical approach to estimate annual LSP from Landsat imagery (1984-present), which has finer spatial resolution (30 m) but relatively sparse temporal frequency. Quantifying phenometric uncertainty has rarely been demonstrated, but is especially important when considering long time series of variable quality and observation density. Thus, our approach uses Monte Carlo Markov Chain (MCMC) sampling process to quantify individual phenometric uncertainty. We observed broad consistency in spring phenometrics and 29 years’ of ground phenology observations at Hubbard Brook (R2 of 0.65). Autumn phenometrics had weaker agreement (R2 of 0.34), likely caused by a decoupling of changes in satellite observed “greenness” and ground observations of leaf coloration and abscission. Extending the moderate resolution LSP record back in time requires methods capable of retrieving phenometrics from sparse time series, and accurately estimating their uncertainty. Our Bayesian hierarchical approach accomplishes both of those goals, and thus represents an important step forward in the production of moderate resolution LSP data.