PROSPECT-DP: A coupled leaf optical properties model for improving leaf spectra simulation in the red-edge domain by excluding the ChlF effect

Abstract

Plenty of leaf optical spectra datasets have been employed in the calibration and validation of leaf optical or empirical models. However, no experiment has been conducted that measures the true leaf reflectance and transmittance spectra while isolating chlorophyll fluorescence (ChlF) contributions, which results in a superimposed effect on measured reflectance or transmittance. As a result, to date, the simulated accuracy of leaf spectra in the red-edge domain remains limited due to the absence of true leaf spectra used in calibration of leaf models. This study aims to enhance the accuracy of simulating leaf spectral characteristics in the red-edge domain by combining an existing leaf physical model (PROSPECT-D) with the data-driven method that is typically used to establish leaf empirical models. A new leaf spectra dataset without ChlF contributions, including reflectance and transmittance data for 849 leaves, was first measured and replace the existing leaf spectra datasets in the modelling process. A coupled leaf optical properties model (PROSPECT-DP) was then established, in which a principal component analysis (PCA) approach was employed to model leaf spectra in red-edge domain using the spectral vectors derived from the training dataset, while the coefficients of spectral vectors were determined by leveraging the PROSPECT-D simulated leaf spectra except for the red-edge domain. Finally, validation of the PROSPECT-DP model with an independent validation dataset of 203 leaf samples showed that it performed much better than the PROSPECT-D model for spectral simulation in the red-edge domain. Furthermore, the PROSPECT-DP model exhibited better performance in leaf trait inversions with a closer relationship between measured and PROSPECT-DP-derived leaf chlorophyll a + b and carotenoid pigments compared to the PROSPECT-D model. Therefore, the novel coupled leaf model presented in this study will be highly beneficial for the integration of leaf models into canopy models and for applications in remote sensing to assess plant traits.