Retrieving global leaf chlorophyll content from MERIS data using a neural network method

Abstract

Leaf chlorophyll content (LCC) is an indicator of plant physiological function and is an important parameter in estimating the carbon and water fluxes of terrestrial ecosystems. Spatiotemporally continuous LCC products are therefore needed at scales from the site level to the globe. In this study, we developed a neural network model for LCC retrieval from ENVISAT MERIS data based on radiative transfer model simulations. By considering the influence of canopy non-photosynthetic materials and the co-variations between LCC and biophysical parameters, a synthetic database was generated using the PROSAIL model with a good approximation to the canopy reflectance collection of MERIS data. Using a neural network trained from the synthetic database, we derived more realistic seasonal patterns of LCC than those using neural network models trained from synthetic databases generated without considering the influence of canopy non-photosynthetic materials or the parameter co-variations. A new global LCC product (GLOBMAP MERIS LCC) at 300-m resolution in 2003–2012 was generated using the neural network. It shows an improvement over the previous MERIS LCC product in capturing LCC seasonal variations in different plant functional types, and is potentially useful in improving the integration of physiological information within terrestrial ecosystem modeling and ecological monitoring across a range of spatial and temporal scales.