Identifying the main drivers of the seasonal decline of near-infrared reflectance of a temperate deciduous forest.

• Understory, leaf angles and sun angle were the main drivers of near-infrared reflectance. • The influence of leaf albedo and leaf area on near-infrared reflectance was negligible. • Simulated seasonal forest canopy reflectances matched Sentinel 2 observations. • Corresponding sun angle effects depended on the viewing angle. • The link between GPP and NIRv may be weaker than expected. The physical mechanisms behind correlations of earth observations and remote sensing products are of vital importance. The so-called 'near-infrared reflectance of vegetation' (NIR V) and gross primary production (GPP) show high correlations among different ecosystems and temporal scales but the underlying relationship is still poorly understood. NIR V is defined as the product of normalized difference vegetation index (NDVI) and near-infrared (NIR) canopy reflectance (R NIR ). We examined this relationship in the case of a temperate deciduous forest in Germany. GPP, R NIR and NIR V all exhibited a strong rise during leaf development in spring and a continual decline after the maximum in early summer. The decline of NIR V in late summer was mainly driven by the decline of R NIR , since NDVI remained saturated. Here we tested the R NIR decline attributions to changes in leaf area index, leaf optical properties, canopy structure, sun-sensor geometry, or understory vegetation by measuring seasonal variations of those factors of the temperate deciduous forest. Leaf area was nearly constant between May and mid September, leaf albedo decreased slightly, leaf angles increased over time towards more vertical leaves, and understory reflectance decreased considerably. We simulated the seasonal R NIR decline of the forest using the radiative transfer model FRT and quantified the sensitivity of the decline to variations in the measured parameters. FRT captured well the observed seasonal R NIR decline by Sentinel 2 using the measured optical and structural properties. Decreasing understory reflectance alone explained 43% of the simulated R NIR decrease, while leaf angle variations explained 31%, the solar zenith angle (SZA) 21%, leaf albedo 7%, and LAI 0%. The effect size of the SZA depended on the viewing angle and would hence be different for different satellites and for local instruments. The results may help to better understand and help to track seasonal changes in forest structure and leaf optical properties using remote sensing techniques. They also suggest that the proposed link between the seasonal evolution of GPP and NIR V may be weaker than expected. [ABSTRACT FROM AUTHOR]