Including Quality Measures in an Automated Processing Chain for Airborne Hyperspectral Data

Recently the hyperspectral community has shown an increased interest in highly reliable, welldocumented and standardized data products, as shown by initiatives like HYRESSA or VDI guidelines. Also quality control (QC) for remote sensing data is already established for some sensors (e.g., MODIS as the best-known example, or previously for the DAIS7915 sensor). In this paper, an approach for the integration of quality measures in a –by the end of 2007– ISOcertified fully automated processing chain for airborne hyperspectral data is introduced, and examples are given. In the past years DLR has built up a generic processing chain for airborne optical sensors, allowing a fully automatic processing starting from system correction, applying ortho-rectification and atmospheric correction, ending up in a fully calibrated, standardized data product. To ensure and to document a high standard of data quality, measures based on scene and sensor characteristics are applied. In addition, extended metadata is distributed with the imagery for a complete documentation. To conclude, the automated generation and distribution of standardized data reports as well as per-pixel quality flag images with every hyperspectral level 1 and 2 dataset is a step towards a more transparent processing resulting in well characterized data. INTRODUCTION In preparation for the coming ARES sensor of GFZ and DLR (Muller, A. et al., 2005; Richter et al., 2005), the generic processing chain at DLR (Habermeyer et al., 2005) was extended to include an automated assessment of data quality, and to provide a data report with extended metadata. This includes general information on sensor and software, auxiliary data (e.g. DEM used for processing), the particular processing methods, as well as scene-specific processing parameters and scene characterization. Each dataset at every processing level is characterized by scene statistics (e.g. histograms and band covariance matrices). In combination with statistical tests for repetitive patterns, and the evaluation of accuracy and stability of internal calibration sources (e.g. blackbody temperature for TIR sensors), one can provide a first measure of overall data characteristics. Next, reports of clouded and haze-affected areas, saturated and dead pixels, as well as measures of platform stability during data acquisition can be of interest for the data user. Thus selected measures are included in a quality flag image with per-pixel information on data quality. For each flight campaign, additional in-flight measures of sensor characteristics like the overall stability of calibration, typical signal-to-noise performance, co-registration of detector modules, and parameters of the sensor modulation transfer function (MTF) are derived and evaluated. Together with laboratory calibration data, this allows an improved long-term monitoring of airborne hyperspectral sensors. Proceedings 5 EARSeL Workshop on Imaging Spectroscopy. Bruges, Belgium, April 23-25 2007 2 OVERVIEW OF THE PROCESSING CHAIN The first part of the generic processing system (Fig. 1) includes data transcription from system file format to a standardized generic data format to be archived as raw data (level L0). After that, data is processed to at-sensor radiance (L1). For HyMap data, system correction is carried out by HyVista (Cocks et al., 1998). In order to process L2 products, parametric geocoding and / or atmospheric correction can be applied. The software packages used are ORTHO for orthographic rectification (Muller, R. et al., 2005), and ATCOR (Richter & Schlapfer, 2002) for atmospheric / topographic correction. Both programs were developed and customized at DLR for the fully automated processing of large datasets. The necessary digital elevation model (DEM) is automatically retrieved from the digital elevation database W42 at DLR (Roth et al., 2002). The whole processing chain is embedded in DLR’s multi-mission processing and archiving environment DIMS (Mikusch et al., 2000). For L2 data, the processing can be customized (e.g., interpolation method, processing sequence, selection of atmospheric parameters, data smoothing). More details on the processing system are given in Habermeyer et al. (2005). Figure 1: Overview of DLR’s automated processing chain for airborne hyperspectral data. Level 1 Product Level 0 Product Raw Data