Assessment of unified models for estimating leaf chlorophyll content across directional-hemispherical reflectance and bidirectional reflectance spectra

As an important indicator of plant photosynthetic activity, leaf chlorophyll content (LCC) has often been estimated non-destructively in the past decades from reflectance spectra measured with various spectrometers and leaf-holding accessories. Most studies established LCC predictive models with either integrating sphere measured directional-hemispherical reflectance factor (DHRF) spectra or leaf clip measured bidirectional reflectance factor (BRF) spectra. Given the effect of specular reflection on leaf reflectance, it remains unclear how DHRF spectra differ from the corresponding BRF spectra and whether the derived predictive models could be transferred between these two types of spectral data. To fill the gap in spectral examination and model comparison, this study aimed to examine the effect of specular reflection on leaf reflectance properties, the commonly used spectral indices or indicators, and the derived regression models across a broad variety of leaf DHRF and BRF spectra. Specifically, we quantified the difference between leaf BRF and DHRF spectra measurements and evaluated the effect of specular reflection on the estimation of LCC with 20 spectral features in four categories (simple ratio, SR; modified SR, mSR; double difference index, DD and red edge position, REP). Seven measured datasets collected from a combination of species, growing conditions, sites, and years were used to evaluate the model difference and calibrate the unified models across DHRF and BRF spectra. The robustness of those predictive models was validated with independent measured and simulated datasets comprised of both BRF and DHRF spectra. Our results demonstrated that the BRF spectra exhibited systematically higher amplitude than DHRF spectra. The widely used vegetation indices (VIs) in the SR category exhibited the most noticeable sensitivity to specular reflection as a significant proportion in BRF spectra. The adverse effect of specular reflection can be alleviated by the other three categories to different extents. With further assessment of model comparisons, we determined four unified models from the pooled data with two DD indices and two REP metrics. Application of the unified models to the validation spectra yielded low RMSE values up to 4.5 mu g/cm(2) and would not result in a significant loss in accuracy as compared to the BRF-specific or DHRF-specific LCC predictive models derived from the four spectral features. The assessment of unified models with comprehensive measured and simulated data could help us better understand the mechanism underlying the specular reflection effect on the relationship of LCC with sensitive spectral features. It will also facilitate the direct estimation of LCC with the common types of leaf reflectance spectra, which is beneficial to rapid and non-destructive determination of LCC for the plant science and agronomy communities.