On the Potential of the Future ENMAP Mission for the Multiseasonal Retrieval of Biophysical Land Surface Parameters
Locherer, Matthias; Hank, Tobias; Mauser, Wolfram
Department of Geography, Ludwig-Maximilians-Universität Munich, GERMANY

To ensure the sustenance of a growing world population, traditional farming methods are stretched to their limits. As a consequence, the importance of precision farming, i.e. an optimized and highly mechanized production of agricultural goods, is increasing. Hyperspectral remote sensing provides technology to derive biophysical land surface parameters, which are vital for improved land surface management, more precisely compared to multispectral methods (i). The upcoming satellite mission EnMAP (Environmental Mapping and Analysis Program) will deliver high quality spaceborne hyperspectral data with a spatial resolution of 30 meters (ii). EnMAP will not only allow for the multiseasonal monitoring of dynamic vegetation development, but will also enable hyperspectral monitoring on the regional scale. In order to prepare for an efficient processing of the future satellite data, methods and algorithms for the derivation of agriculturally relevant land surface parameters are developed using airborne spectroscopy data.

For the purpose of evaluating the performance of different retrieval algorithms with respect to the spectral, spatial and temporal characteristics of the future EnMAP satellite, a data set consisting of imagery and corresponding ground measurements is required, which is (a) spectrally continuous (EnMAP = 420-2450nm @ <=10nm resolution), (b) spatially continuous (EnMAP scene = 30 x 30 km) and (c) temporally continuous (EnMAP near-nadir revisit = 21 days). In order to simulate the capabilities of EnMAP, it therefore is necessary to provide at least four regional hyperspectral acquisitions during one growing season. Applying commercially available imaging spectrometers, however, involves high costs and limited availability, which makes it almost impossible to generate a multiseasonal data set for a specific test area. To overcome this limitation, a cost effective series of airborne imaging spectrometers named AVIS (Airborne Visible and Near Infrared Spectrometer) has been developed at the Department of Geography of the LMU Munich (iii). With the third generation sensor AVIS-3, which is equipped with two camera systems (VNIR and SWIR1) covering a spectral range from 470 - 1700 nm, four acquisitions were obtained during the course of the vegetation period of 2012 (April 28th, May 25th, June 16th, September 8th) over a 12 km² large test site in Southern Germany (Neusling, Lower Bavaria). Furthermore, the 2012-campaign was complemented by two additional acquisitions (May 8th, August 14th) from the airborne sensor HySpex, which is operated by the German Aerospace Center (DLR). Parallel to the imaging flights, in-situ data were gathered, resulting in more than 500 measurements of leaf area index, leaf chlorophyll content, soil moisture, plant height and phenological status of different crops (wheat, barley, maize, sugar beet, rapeseed).

The presented study focuses on comparing empirical and physically-based methods for the multiseasonal derivation of biophysical land surface parameters, especially leaf area index and leaf chlorophyll content, as they are important variables for the description of plant and canopy physiology. On sides of empirical methods, more than 60 widely accepted hyperspectral vegetation indices are applied as well as an optimi-zation algorithm for a narrow band index. Furthermore, a dynamic spectral integral is taken into account. While empirical methods depend on the availability of in-situ data, physically-based methods are inde-pendent from calibration measurements. In addition to this limitation, empirical methods are sensitive to anisotropy effects, resulting from a variable sun-sensor-target-geometry within the airborne data set. Physically-based approaches may explicitly account for these anisotropies, so that illumination angle dependent nonlinearities, instead of being an error source, may serve as additional information, which can be integrated into the retrieval strategy thereby improving the overall retrieval quality. For the physically-based retrieval of land surface parameters, curve fitting is applied using the combined leaf level and canopy reflectance model PROSAIL (PROSPECT5/4SAIL) (iv). A library of look-up-tables (LUT), accounting for different illumination and observer angle settings, serves as basis for the model inversion. The parameter specifications behind these LUTs represent expected value ranges for the leaf and canopy parameters that are accounted for by the reflectance model. The comparison not only investigates, if the physically-based method can replace the accurate, but nonetheless in-situ-dependent empirical methods. It furthermore is investigated, if physically-based approaches may operate in a crop-independent fashion and may be applied multiseasonally without loss of quality.

By artificially upscaling the airborne data to the spatial and spectral resolution of the future EnMAP Hyperspectral Imager, the transferability of the evaluated methods to spaceborne data is shown. As EnMAP will also operate in side-looking modes (+/- 30°), variable illumination situations will be of relevance for the processing of the future EnMAP data as well.

The results indicate that, although physically-based methods require high quality data, a sensor specific preparation of algorithms, as well as a significantly longer computing time, they are superior to empirical methods. It is shown that the physically-based estimation of vegetation parameters from simulated EnMAP data may return more precise results compared to empirical regression approaches. The intended consideration of illumination information and the independence of in-situ measurements enable a high transferability and thus strongly contribute to the applicability of globally deployable systems such as EnMAP. Nevertheless, empirical methods still are useful as they may provide fast and accurate results for studies on well known targets. Furthermore, carefully evaluated vegetation indices may be integrated into the reflectance model inversion process as a-priori-information, increasing the model performance with respect to computing time and accuracy.

REFERENCES
i. Staenz K, 2009. Terrestrial Imaging Spectroscopy Some Future Perspectives, In: 6th EARSeL Workshop on Imaging Spectroscopy, edited by E Ben-Dor (EARSeL SIG-IS, Tel-Aviv)
ii. Kaufmann, et al. (2012): Science Plan of the Environmental Mapping and Analysis Program (EnMAP) , Deutsches GeoForschungsZentrum GFZ, Scientific Technical Report , 63 pp
iii. Oppelt N & W Mauser, 2007. Airborne Visible/Infrared Imaging spectrometer AVIS: Design, characterization and calibration. Sensors, 7: 1934-1953
iv. Baret F, O Hagolle, B Geiger, P Bicheron, B Miras, M Huc, B Berthelot, F Nino, M Weiss, O Samain, J L Roujean, & M Leroy, 2007. LAI, fAPAR and fCover CYCLOPES global products derived from VEGETATION - Part 1: Principles of the algorithm. Remote Sensing of Environment, 110: 275-286