The leaf optical properties model PROSPECT has been widely used to retrieve foliar chemistry in reverse mode from directional-hemispherical reflectance factor (DHRF) spectra measured with integrating sphere equipped spectrometers. With bidirectional reflectance factor (BRF) spectra, some researchers attempted to invert PROSPECT after a modification to the latest version of the model. However, the retrieval accuracy varies greatly with chemical constituents and can be low for some of them, such as dry matter content. This paper proposes a new approach called PROCWT by coupling PROSPECT with continuous wavelet transform (CWT) to suppress the surface reflectance effect and enhance the absorption features of chemical constituents. Instead of the reflectance spectra, the wavelet coefficient spectra generated after CWT were used to construct the merit function for model inversion. Given that the multi-scale decomposition of CWT enables enhancement of chemical-specific absorption features, the use of PROCWT at different scales of wavelet decomposition could lead to improved retrievals of biochemical parameters. The performance of PROCWT was evaluated for estimating foliar chemicals of wheat and rice crops from BRF spectra measured with a leaf clip equipped spectrometer over a two-year field experiment. PROCWT was also compared with the standard PROSPECT inversion (STANDARD), the PROSPECT inversion with the subtraction of surface reflectance (PROREF), and the simplified PROCOSINE (sPROCOSINE).Our results demonstrated that the contribution of surface reflectance component was significant for BRF spectra and the effect of surface reflectance could be suppressed by PROCWT as well as PROREF and sPROCOSINE. Compared with STANDARD, PROCWT and the two traditional methods significantly improved the retrieval accuracies for pigments and leaf water content, but only PROCWT produced significant improvement for dry matter content with a decrease of 14.79g/m2 in the root mean squared error (RMSE) (30% of the mean) over the entire experimental dataset by enhancing dry matter absorption features. High scales of wavelet decomposition were favorable for the estimation of carotenoid and water contents and low scales for the estimation of chlorophyll and dry matter contents. The difference in optimal scale revealed the separation of overlapping absorption features attributed to various chemical constituents. In addition, the newest PROSPECT-D outperformed PROSPECT-5B in the retrieval of chlorophyll content but not for carotenoid. This new physically-based approach could be beneficial to analysts attempting to retrieve leaf chemicals from BRF spectra alone and close-range reflectance imagery of crops and even other vegetation types.
Read full abstract