Methodology comparison of quantitative LAI retrieval using imaging spectroscopy and geo-spatial iterpolation in a soft wood forest

Recent advances in validation of vegetation products have led to new ground based methods to assess the accuracy for small and large spatial footprint sensors. In addition to classical biophysical parameter retrieval (in particular Leaf Area Index (LAI)), current approaches combine photonvegetation interaction based methods for both, the remote sensing instrument and the ground measurements.

An experiment with the airborne imaging spectrometer HyMap was carried out in the Millingerwaard in the Netherlands in summer 2004 in the framework of an initiative of the Belgian Space Office (Belspo). The Millingerwaard is a large flooding area of the river Rhine, close to the German-Dutch border. The area is a natural reserve and significant supporting data exists (vegetation maps, LIDAR data, CASI data, species composition maps, etc.). The Millingerwaard is a managed natural ecosystem with dominant softwood forests comprised of Populus nigra L. Salix alba L. and Salix purpurea and dense understory namely Urtica dioica L. and Rubus species.

We propose an approach using hemispherical camera pictures combined with a stratified random sampling scheme at plot level and systematic sub-sampling of these plots similar to the VALERI approach. Measurements to assess on-ground the gap fraction, leaf distribution angle, and specie reflectance to estimate LAI have been performed as well. 156 points (e.g., 13 plots, each with 12 sub-sampling points) have been measured and processed using a neural network based approach for validation and calibration purposes of LAI.

LAI is derived on the one hand from the simultaneous HyMap overflight of the test site using a quantitative statistical based approach with calibrated gain factors following the combined approaches of Clevers (WDVI) and Chen (LAI bias correction). On ground spatially distributed LAI maps are generated on plot level (each plot covers 20x20m) and used for calibration and validation of HyMap derived LAI. The second approach includes a spatial interpolation approach that is based on simple kriging with varying local means of the ground measured LAI. The HyMap derived fractional cover is used as an exhaustive secondary variable for the geo-statistical approach.

Both, ground and HyMap derived LAI values are subject to a significant bias, as recently reported in literature. We are using the bias correction to compensate for this difference, and derive true LAI for both methods. The two proposed methods are finally compared to measured ground LAI, where some of the plots remain as validation points that previously have not been used for calibration or interpolation approaches. We conclude that the proposed approaches are very well suited to derive LAI on a forest stand scale with unprecedented accuracy. This is due to the fact that both methods are based on the same principles and we use consistent approaches to measure the interaction of light with vegetation.