Optimized Processing of Airborne Hyperspectral Data for Forest Studies

In order to produce accurate and meaningful results from airborne hyperspectral data analysis, it is essential to have exact knowledge about the quality of the image data itself. Small differences in reflectance already produce diverse results particularly when it comes to vegetation analysis. Especially within the complex environment of forested areas having strong shading effects and large inter- and intra-class spectral variability, there are many naturally occurring aspects influencing the accuracy of image-retrieved parameters. Also in addition to the sensors signal-to-noise ratio and atmospheric conditions during data acquisition, the quality of the data analysis is also dependent on the variable viewing and illumination geometry. In the case of implicating multi-temporal data sets, the co-registration issue is another major quality indicator. The Earth Observation Center (EOC) of the German Aerospace Center (DLR) has developed a processing chain which generates standardised data products automatically, allowing the data to be reproduced easily at any time. This includes systematic and radiometric correction, direct georeferencing of the data and atmospheric correction. These processing steps as well as further necessary pre-processing of the data (e.g. spectral filtering, BRDF correction, masking, band selection) will be investigated to allow efficient and appropriate data extraction for forest applications. In this contribution, we will present the results of the pre-processing optimization exemplary for the case of tree species classification. This involves the mentioned pre-processing steps of radiometric correction, ortho-rectification including the image-to-image matching of the VNIR and SWIR cubes, atmospheric correction and class-specific BRDF correction, mosaicking with fine-adjustment between flightlines, as well as the corresponding data quality control issues. In addition to the airborne data itself, training endmember selection and field validation will complete the good practice of “Optimized Processing of Airborne Hyperspectral Data for Forest Studies”. This study is based on a large scale multitemporal, hyperspectral survey flown in 2013, 2015 and 2016 over the Bavarian Forest National Park in the south-eastern part of Germany. The National Park is a unique area of forest stands that developed with low anthropogenic interference into a landscape with remnants of a primeval forest. The park covers an elevation ranging from approximately 600 m to 1450 m above sea level. It therefore represents a heterogenic and complex study area. The data was acquired by DLR’s OpAiRS service using a HySpex system, comprising two imaging spectrometers with spectral ranges of 400-1000 nm and 1000-2500 nm and up to 416 spectral channels