Remotely sensed carotenoid dynamics improve modelling photosynthetic phenology in conifer and deciduous forests

Detecting the phenology of photosynthesis, which conveys the length of the growing season, is key for terrestrial ecosystem models to constrain total annual carbon uptake and estimate gross primary productivity (GPP). However, some of the vegetation indices that are widely used for modelling GPP lack the ability to represent changes in the magnitude of photosynthesis, leading to errors in detecting phenology and large uncertainties. Crucially, remotely sensed vegetation indices such as the photochemical reflectance index (PRI) and chlorophyll/carotenoid index (CCI) can detect changes in foliar carotenoid composition that represent adjustments in photosynthetic light-use-efficiency (LUE) and changes in phenology. We modeled GPP from remote sensing data using PRI and CCI to represent foliar carotenoid changes as proxies for LUE. GPP values estimated from these PRI and CCI modified LUE models were compared against GPP from eddy covariance flux tower measurements, MODerate Resolution Imaging Spectroradiometer (MODIS) GPP product, conventional meteorological driven LUE-model, and process-based dynamic global vegetation model (ie. JULES) in an evergreen needleleaf and a deciduous broadleaf forest in the Great Lakes region. Overall, and in particular for evergreen needleleaf forests, estimates of start and end of growing season using PRI and CCI LUE-models showed less year-to-year variability than estimates obtained by process-based meteorological models. Although many process-based models provide reasonable estimates of start and end of growing season, our results demonstrate that using regulatory carotenoids and photosynthetic efficiency can improve remote monitoring of the phenology of forest vegetation.