Retrieval of coniferous canopy chlorophyll content from high spatial resolution hyperspectral data

Abstract

The discrete Anisotropic Radiative Transfer (DART) model, coupled with an adjusted version of the PROSPECT model, was used to retrieve total chlorophyll content (Cab) of a complex Norway spruce (Picea abies (L.) Karst.) canopy from airborne hyperspectral data acquired at very high spatial resolution. The radiative transfer models were parameterized by using field measurements and observations collected from a young spruce stand growing at the permanent experimental site Bilý Križ (the Moravian-Silesian Beskydy Mts., the Czech Republic, 18.53863°E, 49.50256°N, 936 m a.s.l.). A set of the hyperspectral images with a pixel-size of 0.4 m was acquired for the test site by an airborne AISA Eagle VNIR system in September 18th, 2004. An operational canopy Cab estimation was carried out by means of a PROSPECT-DART inversion employing an artificial neural network (ANN) and a vegetation index ANCB650-720. Both retrieval approaches used continuum removed reflectance values of six AISA Eagle spectral bands located between 650 and 720 nm. The Cab inversion was only performed for direct sun exposed (sunlit) crown pixels in order to ensure a high quality (noiseless) reflectance signal. Results of both inversion approaches were similar, when validated against the ground measured Cab of nine Norway spruce crowns. Coefficients of determination (R2) between ground truth and remote sensing Cab estimates were 0.78 and 0.76, respectively, with root mean square errors (RMSE) of 2.95 μg cm⁻² for the ANN and 3.36 μg cm⁻² for the ANCB650-720 retrieval. The spatial patterns of Cab values estimated by both inversion methods were consistent with each other. About 80% of the Cab estimated values had an absolute difference smaller than 2 µg cm⁻².