Evaluating the performance of PROSPECT in the retrieval of leaf traits across canopy throughout the growing season

Abstract

Leaf traits and subsequently leaf spectral properties depend on the leaf phenological stage and light conditions within a canopy. The PROSPECT radiative transfer model has been extensively and successfully used to retrieve leaf traits for mature, sunlit leaves at peak vegetation growth, i.e. summer. However, research on the quantification of leaf traits using PROSPECT across the canopy vertical profile throughout the growing season is still lacking. Therefore, this study aims at examining the effect of leaf position on the performance of the PROSPECT model in modelling leaf optical properties and retrieving leaf chlorophyll content (Cab), equivalent water thickness (EWT), and leaf mass per area (LMA) throughout the growing season. To achieve this objective, we collected 588 leaf samples from the upper and lower canopies of deciduous stands over three seasons (i.e., spring, summer and autumn) in Bavaria Forest National Park, Germany. Leaf traits including Cab, EWT and LMA, were measured for all the samples, and their reflectance spectra were obtained using an ASD FieldSpec-3 Pro FR spectroradiometer coupled with an Integrating Sphere. We initially assessed the performance of the PROSPECT model by comparing reflectance spectra generated in forward mode against reflectance spectra measured on leaf samples collected in the field. We subsequently inverted the PROSPECT model to retrieve Cab, EWT and LMA using the look-up-table (LUT) approach. Our results consistently demonstrated that the measured reflectance of leaf samples collected from the lower canopy had a stronger match with PROSPECT simulated reflectance spectra, especially in the NIR spectrum compared to leaf samples collected from the upper canopy throughout the growing season. This observation concurred with the pattern of Cab and EWT retrieval accuracies across the canopy i.e. the retrieval accuracy for the lower canopy was consistently higher (NRMSE = 0.1-0.2 for Cab; NRMSE = 0.125-0.16 for EWT) when compared to the upper canopy (NRMSE = 0.122 - 0.269 for Cab; NRMSE = 0.162 -0.0.258 for EWT) across all seasons. In contrast, LMA retrieval accuracies for the upper canopy (NRMSE = 0.146 - 0.184) were higher compared to the lower canopy (NRMSE = 0.162 - 0.239) for all seasons except for the spring season. For all the leaf traits examined in this study, the range in retrieval accuracy between the upper and lower canopy was greater in summer (compared to other seasons). We report for the first time that although the PROSPECT model provides reasonable retrieval accuracy of Cab, EWT and LMA, variations in leaf biochemistry and morphology through the vertical canopy profile affects the performance of the model over the growing season. Findings of this study have important implications on field sampling protocols and upscaling leaf traits to canopy and landscape level using multi-layered physical models coupled with PROSPECT.