Comparison of Empirical and Physical Modelling for Estimation of Biochemical and Biophysical Vegetation Properties: Field Scale Analysis across an Arctic Bioclimatic Gradient
To evaluate the potential of multi-angle hyperspectral sensors for monitoring vegetation variables in Arctic environments, empirical and physical modelling using field data was implemented for the retrieval of leaf and canopy chlorophyll content (LCC, CCC) and plant area index (PAI) measured at four...
| Published in: | Remote Sensing |
|---|---|
| Main Authors: | , , |
| Format: | Article in Journal/Newspaper |
| Language: | English |
| Published: |
MDPI AG
2020
|
| Subjects: | |
| Online Access: | https://doi.org/10.3390/rs12183073 https://doaj.org/article/7432b33edd0b44f99461cb103eebdd7a |
| Summary: | To evaluate the potential of multi-angle hyperspectral sensors for monitoring vegetation variables in Arctic environments, empirical and physical modelling using field data was implemented for the retrieval of leaf and canopy chlorophyll content (LCC, CCC) and plant area index (PAI) measured at four sites situated across a bioclimatic gradient in the Western Canadian Arctic. Field reflectance data were acquired with an ASD FieldSpec (305–1075 nm) and used to simulate CHRIS Mode1 spectra (411–997 nm). Multi-angle measurements were taken corresponding to CHRIS view zenith angles (VZA) (−55°, −36°, 0°, +36°, +55°). Empirical modelling compared parametric regression based on vegetation indices (VIs) to non-parametric Gaussian Processes Regression (GPR). In physical modelling, PROSAIL was inverted using numerical optimization and look-up table (LUT) approaches. Cross-validation of the empirical models ranked GPR as best, followed by simple ratio (SR) with optimally selected NIR and red wavelengths, and then ROSAVI using its published wavelengths (mean r 2 cv = 0.62, 0.58, and 0.54, respectively across all sites, variables, and VZAs). However, the best predictive performance was achieved by SR followed by GPR and ROSAVI (NRMSE cv = 0.12, 0.16, 0.16, respectively). PROSAIL simulated the multi-angle top-of-canopy reflectance well with numerical optimization ( r 2 = ~0.99, RMSE = 0.004 ± 0.002), but best performing LUT models of LCC, CCC and PAI were poorer than the empirical approaches (mean r 2 = 0.48, mean NRMSE = 0.22). PROSAIL performed best at the high Arctic sparsely vegetated site ( r 2 = 0.57–0.86 for all parameters). Overall, the best performing VZA was −55° for empirical modelling and 0° and ±55° for physical modelling; however, these were not significantly better than the other VZAs. Overall, this study demonstrates that, for Arctic vegetation, nadir narrowband reflectance data used to derive simple empirical VIs with optimally selected bands is a more efficient approach for modelling chlorophyll and PAI than ... |
|---|