Leaf Area Index derivation from hyperspectral and multispectral remote sensing data in heterogeneous grassland

Kurzfassung

The Leaf Area Index (LAI) is a key parameter controlling biophysical exchange processes in the vegetation canopy. Automated LAI derivation from remote sensing data is only feasible based on physical modeling as it is independent from field measurements, and the increasing number of spaceborne, airborne and UAV-based imaging systems provides manifold opportunities for high spatial and temporal resolution and potentially more accurate LAI estimates in the future. However, the use of hyperspectral remote sensing data in an inverted radiation transfer model has only been analyzed in a small number of studies. Further, physical LAI derivation in heterogeneous, semi-natural ecosystems such as grasslands has been largely neglected. In this study, LAI is derived for heterogeneous alpine grasslands from airborne hyperspectral data using an inverted radiation transfer model. HySpex data have been recorded on August 13, 2012 in the Bavarian alpine upland south of Munich (Germany), covering about 40 km2. The HySpex data comprise the spectral range of 400 - 2500 nm in 336 bands. Contemporaneous grasslands LAI in situ measurements (n = 22) have been conducted in the study area for validation purposes. LAI is derived from the HySpex data using the PROSAIL model and a look-up table inversion approach as implemented in the ARTMO toolbox. The model is parameterized based on field measurements of the chlorophyll, water, and dry matter contents, as well as of the leaf angle distribution. For LAI derivation, a range of optimization strategies is evaluated in order to increase the accuracy and robustness of the LAI mapping algorithm. First, the number of spectral bands used and the distribution of these bands across the spectrum determine the amount of available spectral information as well as the uncertainty potentially biasing the result. Further, the level and structure of noise added to the simulated spectra should account for signal disturbances and simplifications of the PROSAIL model. During inversion, the type of cost function and the number of model results used in a multiple solution sample influence the performance of LAI derivation. The combination of these regularization strategies achieving the highest accuracies is determined. Additionally, a RapidEye scene acquired on the same day over the study area is used for LAI derivation using the same parameterization and inversion settings in order to assess the accuracy loss that could be attributed to the reduced spectral and spatial resolution.