EARSeL eProceedings 3, 3/2004 405 DERIVATION OF FOREST LEAF AREA INDEX FROM MULTI- AND HYPERSPECTRAL REMOTE SENSING DATA Martin Schlerf, Clement Atzberger, Michael Vohland, Henning Buddenbaum, Stephan Seeling and Joachim Hill Trier University, Department of Remote Sensing, D-54286 Trier, Germany; schlerf(at)uni-trier.de ABSTRACT This study evaluated systematically linear predictive models between vegetation indices (VIs) de- rived from radiometrically corrected airborne imaging spectrometer (HyMap) data and field meas- urements of leaf area index (LAI) (n=40). Ratio-based and soil-line related broadband VIs were calculated after HyMap reflectance had been spectrally resampled to Landsat TM channels. Hy- perspectral VIs involved all possible types of 2-band combinations of RVI and PVI. Cross- validation procedure was used to assess the prediction power of the regression models. Analyses were performed on the entire data set or on subsets stratified according to stand age. A perpen- dicular vegetation index (PVI) based on wavebands at 1,088 and 1,148 nm was linearly related to leaf area index (LAI) (R 2 =0.67, RMSE=0.69 m 2 m -2 (21% of the mean); after removal of one forest stand subjected to clearing measures: R 2 =0.77, RMSE=0.54 m 2 m -2 (17% of the mean)). The study demonstrates that for hyperspectral image data, linear regression models can be applied to quan- tify LAI with good accuracy. Best hyperspectral VIs in relation with LAI are typically based on wavebands related to prominent water absorption features. Such VIs measure the total amount of canopy water; as the leaf water content is considered to be relatively constant in the study area, variations of LAI are retrieved. Keywords: Hyperspectral, vegetation indices, forest, leaf area index INTRODUCTION The majority of studies for extracting biophysical variables from remotely sensed data have used empirical techniques to relate the spectral measurements to biophysical parameters (1). While much work has been done with agricultural crops, relatively little research has been done on inves- tigating the relationships between forest leaf area index (LAI) and satellite data (2). Most of the studies on forests have used broadband VIs (e.g. NDVI, RVI) to derive LAI of coniferous forest stands; however, with varying success (3;4,5,6,7,8;9). Few studies looked at the suitability of high spectral resolution remote sensing data to derive the LAI of coniferous forests. (10) tested CASI data using univariate and multivariate regression and a VI based algorithm and found strong rela- tionships with reasonable low errors. (11) tested if the red edge inflection point (REIP) is primarily controlled by forest canopy LAI using helicopter-borne spectroradiometer data and found a strong non-linear correlation between plot LAI and the REIP for Sitka spruce (Picea sitchensis). For the same tree species, forest LAI was recently related to the canopy REIP computed from imaging spectrometer data (CASI) with success (12). Despite the research undertaken it is often unclear whether the high spectral resolution data offer advantages over broadband data (13). The overall aim of the work was to evaluate the information content of hyperspectral remote sens- ing data for the estimation of forest LAI in comparison with broadband data. More specific objec- tives were (i) to determine spectral vegetation indices (VI) that are best suited for characterising LAI, and (ii) to compare and contrast traditional broad-band and hyperspectral VIs in terms of ba- sic statistical characteristics of the predicted relative to the observed LAI. The research was re- stricted to Norway spruce as only forest stands of this single species occur in sufficient numbers within the selected test site. Besides, coniferous forests do not show a saturation of VI with in-