Search

Atmospheric correction plays a crucial role among the processing steps applied to remotely sensed hyperspectral data. Atmospheric correction comprises a group of procedures needed to remove atmospheric effects from observed spectra, i.e. the transformation from at-sensor radiances to at-surface radiances or reflectances. In this paper we present the different steps in the atmospheric correction process for APEX hyperspectral data as applied by the Central Data Processing Center (CDPC) at the Flemish Institute for Technological Research (VITO, Mol, Belgium). The MODerate resolution atmospheric TRANsmission program (MODTRAN) is used to determine the source of radiation and for applying the actual atmospheric correction. As part of the overall correction process, supporting algorithms are provided in order to derive MODTRAN configuration parameters and to account for specific effects, e.g. correction for adjacency effects, haze and shadow correction, and topographic BRDF correction. The methods and theory underlying these corrections and an example of an application are presented.

Hyperspectral imagery originating from airborne sensors is nowadays widely used for the detailed characterization of land surface. The correct mapping of the pixel positions to ground locations largely contributes to the success of the applications. Accurate geometric correction, also referred to as “orthorectification”, is thus an important prerequisite which must be performed prior to using airborne imagery for evaluations like change detection, or mapping or overlaying the imagery with existing data sets or maps. A so-called “ortho-image” provides an accurate representation of the earth’s surface, having been adjusted for lens distortions, camera tilt and topographic relief. In this paper, we describe the different steps in the geometric correction process of APEX hyperspectral data, as applied in the Central Data Processing Center (CDPC) at the Flemish Institute for Technological Research (VITO, Mol, Belgium). APEX ortho-images are generated through direct georeferencing of the raw images, thereby making use of sensor interior and exterior orientation data, boresight calibration data and elevation data. They can be referenced to any userspecified output projection system and can be resampled to any output pixel size.

The accurate mapping of forest species is a very important task in relation to the increasing need to better understand the role of the forest ecosystem within environmental dynamics. The objective of this paper is the investigation of the potential of a multi-temporal hyperspectral dataset for the production of a thematic map of the dominant species in the Forêt de Hardt (France). Hyperspectral data were collected in June and September 2013 using the Airborne Prism EXperiment (APEX) sensor, covering the visible, near-infrared and shortwave infrared spectral regions with a spatial resolution of 3 m by 3 m. The map was realized by means of a maximum likelihood supervised classification. The classification was first performed separately on images from June and September and then on the two images together. Class discrimination was performed using as input 3 spectral indices computed as ratios between red edge bands and a blue band for each image. The map was validated using a testing set selected on the basis of a random stratified sampling scheme. Results showed that the algorithm performances improved from an overall accuracy of 59.5% and 48% (for the June and September images, respectively) to an overall accuracy of 74.4%, with the producer’s accuracy ranging from 60% to 86% and user’s accuracy ranging from 61% to 90%, when both images (June and September) were combined. This study demonstrates that the use of multi-temporal high-resolution images acquired in two different vegetation development stages (i.e., 17 June 2013 and 4 September 2013) allows accurate (overall accuracy 74.4%) local-scale thematic products to be obtained in an operational way.

The most commonly used approach to estimate soil variables from remote-sensing data entails time-consuming and expensive data collection including chemical and physical laboratory analysis. Large spectral libraries could be exploited to decrease the effort of soil variable estimation and obtain more widely applicable models. We investigated the feasibility of a new approach, referred to as bottom-up, to provide soil organic carbon (SOC) maps of bare cropland fields over a large area without recourse to chemical analyses, employing both the pan-European topsoil database from the Land Use/Cover Area frame statistical Survey (LUCAS) and Airborne Prism Experiment (APEX) hyperspectral airborne data. This approach was tested in two areas having different soil characteristics: the loam belt in Belgium, and the Gutland–Oesling region in Luxembourg. Partial least square regression (PLSR) models were used in each study area to estimate SOC content, using both bottom-up and traditional approaches. The PLSR model’s accuracy was tested on an independent validation dataset. Both approaches provide SOC maps having a satisfactory level of accuracy (RMSE = 1.5–4.9 g·kg−1; ratio of performance to deviation (RPD) = 1.4–1.7) and the inter-comparison did not show differences in terms of RMSE and RPD either in the loam belt or in Luxembourg. Thus, the bottom-up approach based on APEX data provided high-resolution SOC maps over two large areas showing the within- and between-field SOC variability.