Estimating Leaf Area Index from Terrestrial Lidar and Satellite Based Vegetation Indices Using Bayesian Inference

Abstract:

Leaf area index (LAI) is an important indicator of ecosystem conditions and a key biophysical variable to many ecosystem models. The LAI in this study was measured by Leica ScanStation C 10 Terrestrial Laser Scanner (TLS) and a hand-held Li-Cor LAI-2200 Plant Canopy Analyzer for understanding differences derived from the two sensors. A total of six different LAI estimates were generated using different methods for the comparisons. The results suggested that there was a reasonable agreement (i.e., correlations r > 0.50) considering a total of 30 plots and use of very different in situ foliage measurements. . The predicted LAI from spectral vegetation indices including WDVI, DVI, NDVI, SAVI, and PVI3 which were derived from Landsat TM imagery were used to identify statistical relationships and for the development of the Bayesian inference model. The Bayesian Linear Regression (BLR) approach was used to scale up LAI estimates and to produce continuous field surfaces for the Oak Openings Region in NW Ohio. The results from the BLR provided details about the parameter uncertainties but also insight about the potential that different LAIs can be used to predict foliage that has been adjusted by removing the wooden biomass with reasonable accuracy. For instance, the modeled residuals associated with the LAI estimates from TLS orthographic projection that consider only foliage had the lowest overall model uncertainty with lowest error and residual dispersion range among the six spatial LAI estimates. The deviation from the mean LAI prediction map derived from the six estimates hinted that sparse and open areas that relate to vegetation structure were associated with the highest error. However, although in many studies TLS has been shown to hold a great potential for quantifying vegetation structure, in this study the quantified relationship between LAI and the vegetation indices did not yield any statistical relationship that needs to be further explored.