Hyperspectral indices for estimating leaf biochemical properties in temperate deciduous forests: Comparison of simulated and measured reflectance data sets

This study aimed at finding efficient hyperspectral indices for estimating three leaf biochemical parameters: chlorophyll content (CHL, μg cm−2), leaf water thickness (EWT, g cm−2), and leaf mass per area (LMA, g cm−2) in typical temperate deciduous forests. These parameters are required by most biogeochemical models that describe ecosystem functions. We have identified the most efficient hyperspectral indices (both the index types and the wavelength domains) based on both a simulated data set (produced with the calibrated leaf reflectance model PROSPECT) and with data sets (I, II, and III) from measurement of field-collected leaves. Results indicated that CHL, EWT, and LMA can be estimated with high precision using a two-waveband vegetation index (Double Deference index, DDn) for all parameters, with an overall root mean square error (RMSE) of 6.87 μg cm−2 for CHL, 0.0011 g cm−2 for EWT, and 0.0015 g cm−2 for LMA. The best overall indices for temperate deciduous forests were DDn (715, 185) for CHL, DDn (1530, 525) for EWT, and DDn (1235, 25) for LMA, although these indices were not necessarily the best for every specific data set (especially for the simulated data set). Moreover, discrepancies were obvious when the identified indices were applied to different data sets. Even if the wavelengths of calibrated indices have been accurately determined through the simulated data set, the regressions between the indices and the biochemical parameters must be calibrated with field-based measurements. The indices identified in this study are applicable to various species (data set III), various phenological stages and locations (data set I), and various leaf anatomies (data set II) and may therefore be widely applicable for temperate deciduous forests and possibly for other plant communities.