W.H. Bakker

Cross correlogram spectral matching : application to surface mineralogical mapping by using AVIRIS data from Cuprite, Nevada

Abstract A new approach toward mineral mapping front imaging spectrometer data is presented, using a spectral matching algorithm- based on the cross correlograrn. A cross correlograin is constructed b y calculating the cross correlation. at different match positions, m, between a test spectrum (i.e., a pixel spectrum) and a reference spectrum (i.e., a laboratory mineral spectrum or a pixel spectrum known to represent a mineral of interest) by shifting the reference spectrum over subsequent channel positions. The cross correlogram for perfectly matching reference and test spectra is a parabola around the central matching number (m=0) with a peak correlation. of I. In laboratory .spectra, deviations from this shape indicate differences in mineralogy, whereas, in image data, this may be partly attributed to spectral mixing, noise, changes in atmospheric and illumination conditions, and other scene- and sensor-dependent variables. A cross correlograin spectral snatching algorithm was designed and tested on 1994 data from the airborne visible/infrared imaging spectrometer of the Cuprite mining area. Accurrate mapping of kaolinite, alunite, and buddingtonite was achieved by extracting three parameters from the cross correlograins that were constructed on a pixel-by-pixel basis: the correlation coefficient at match position zero, the rrurrrtent of skewness (based on the correlation differences between match numbers of equal but reversed signs; e.g., m=4 and m=-4), and the significance (based on a Students t-test of the validity of the correlation coefficients).

A back-propagation neural network for mineralogical mapping from AVIRIS data

Imaging spectrometers have the potential to identify surface mineralogy based on the unique absorption features in pixel spectra. A back-propagation neural network (BPN) is introduced to classify A...

CCSM : cross correlogram spectral matching

Cross correlogram spectral matching (CCSM) is a new approach towards mineral mapping from imaging spectrometer data. A cross correlogram is constructed by calculating the cross correlation at different match positions between a test spectrum (i.e., a remotely-sensed spectrum) and a reference spectrum (i.e., a laboratory mineral spectrum). To assess the sensitivity of the cross correlogram as a means of spectral matching, the technique is applied to laboratory spectra. In each experiment, the cross correlogram function was derived, a test of significance of the correlations was conducted and a moment of skewness was calculated to characterize the curve shape of the correlogram. The cross correlogram for a perfect spectral match had a parabolic shape with the maximum correlation of one at match position zero and a symmetry around the central match number. The cross correlation is found to be insensitive to differences in gain and thus allows to compare materials of different albedos. The cross correlogram i...

MERIS and the red-edge position

Abstract The Medium Resolution Imaging Spectrometer (MERIS) is a payload component of Envisat-1. MERIS will be operated over land with a standard 15 band setting acquiring images with a 300 m spatial resolution. The red-edge position (REP) is a promising variable for deriving foliar chlorophyll concentration, which plays an important role in ecosystem processes. The objectives of this paper are: (1) to study which factors effect the REP of vegetation, (2) to study whether this REP can be derived from the MERIS standard band setting and (3) to show what REP represents at the scale of MERIS data. Two different data sets were explored for simulating the REP using MERIS bands: (1) simulated data using reflectance models and (2) airborne reflectance spectra of an agricultural area obtained by the airborne visible-infrared imaging spectrometer (AVIRIS). A “linear method”, assuming a straight slope of the reflectance spectrum around the midpoint of the slope, was a robust method for determining the REP and the MERIS bands at 665, 708.75, 753.75 and 778.75 nm could be used for applying the “linear method” for REP estimation. Results of the translation to the scale of MERIS data were very promising for applying MERIS at, for instance, the ecosystem level.

Thermal Infrared Spectrometer for Earth Science Remote Sensing Applications—Instrument Modifications and Measurement Procedures

In this article we describe a new instrumental setup at the University of Twente Faculty ITC with an optimized processing chain to measure absolute directional-hemispherical reflectance values of typical earth science samples in the 2.5 to 16 μm range. A Bruker Vertex 70 FTIR spectrometer was chosen as the base instrument. It was modified with an external integrating sphere with a 30 mm sampling port to allow measuring large, inhomogeneous samples and quantitatively compare the laboratory results to airborne and spaceborne remote sensing data. During the processing to directional-hemispherical reflectance values, a background radiation subtraction is performed, removing the effect of radiance not reflected from the sample itself on the detector. This provides more accurate reflectance values for low-reflecting samples. Repeat measurements taken over a 20 month period on a quartz sand standard show that the repeatability of the system is very high, with a standard deviation ranging between 0.001 and 0.006 reflectance units depending on wavelength. This high level of repeatability is achieved even after replacing optical components, re-aligning mirrors and placement of sample port reducers. Absolute reflectance values of measurements taken by the instrument here presented compare very favorably to measurements of other leading laboratories taken on identical sample standards.

Spatial scale variations in vegetation indices and above-ground biomass estimates: Implications for MERIS

The Medium Resolution Imaging Spectrometer (MERIS) is one of the sensors carried by Envisat. MERIS is a fully programmable imaging spectrometer, however a standard 15-channel band set will be transmitted for each 300 m pixel (over land while over the ocean the pixels will be aggregated to 1200 m spatial resolution) covering visible and near-infrared wavelengths. Since MERIS is a multidisciplinary sensor providing data that can be input into ecosystem models at various scales, we studied MERISs performance relative to the scale of observation using simulated datasets degraded to various spatial resolutions in the range of 6-300 m. Algorithms to simulate MERIS data using airborne imaging spectrometer datasets were presented, including a case study from DAIS (i.e. Digital Airborne Imaging Spectrometer) 79-channel imaging spectrometer data acquired on 8 July 1997 over the Le Peyne test site in southern France. For selected target endmembers garrigue, maquis, mixed oak forest, pine forest and bare agricultural field, regions-of-interest (ROI) were defined in the DAIS scene. For each of the endmembers, the vegetation index values in the corresponding ROI is calculated for the MERIS data at the spatial resolutions ranging from 6 to 300 m. We applied the NDVI, PVI, WDVI, SAVI, MSAVI, MSAVI2 and GEMI vegetation indices. Above-ground biomass (AGB) was estimated in the field and derived from the DAIS image and the MERIS datasets (6-300 m spatial resolution). The vegetation indices are shown to be constant with the spatial scale of observation. The strongest correlation between the MERIS and DAIS NDVI is obtained when using a linear model with an offset of 0.15 ( r =0.31). A Pearson correlation matrix between AGB measured in the field and each spectral band reveals a modest but significant ( p <0.05) correlation for most spectral bands. When mathematical functions are fitted through the NDVI and biomass data, an exponential fit shows the extinction and saturation at larger vegetation biomass values. The correlation between biomass and NDVI for DAIS as well as for the MERIS simulated dataset is modest. Further research is required to analyse the scale effects that limit the correlation between field and image AGB estimates.

Combining spectral signals and spatial patterns using multiple Hough transforms: An application for detection of natural gas seepages

Object detection in remote sensing studies can be improved by incorporating spatial knowledge of an object in an image processing algorithm. This paper presents an algorithm based on sequential Hough transforms, which aims to detect botanical and mineralogical alterations that result from natural seepage of carbon dioxide and light hydrocarbons. As the observed alterations are not unique for gas seepages, these halos can only be distinguished from the background by their specific spatial pattern: the alterations are present as halos that line up along geological lineaments in the shallow subsurface. The algorithm is deployed in three phases: a prior spectral classification followed by two serialized Hough transforms. The first Hough transform fits circles through spectrally optimal matching pixels. Next, the centers of the detected circles are piped into the second Hough transform that detects points that are located on a line. Results show that our algorithm is successful in detecting the alteration halos. The number of false anomalies is sufficiently reduced to allow an objective detection based on field observations and spectral measurements.

Storing and handling vague spatial objects

This paper deals with the storage and manipulation of vague spatial objects: points, lines, and regions. Vector data format is used to implement vague objects. We store vague objects by extending GRASS vector format capabilities. A point is represented by a triple, containing the location and the membership value of the point. A line is represented by a sequence of triples, each containing the location and the membership value of a point on the line. A region is represented by a set of lines, which depict important characteristics of the region. Data structures are defined to put constraints on the stored data, so that desired properties for vague objects are satisfied. To complete (approximate) information about lines and regions, a linear interpolation and a triangulation-based interpolation method are used, respectively. Modules are built to edit and retrieve information of vague objects. A union operator on vague objects is also implemented.

Scaling to the MERIS Resolution: Mapping Accuracy and Spatial Variability

Abstract Within the framework of ESAs Earth Observation Program, the Medium Resolution Imaging Spectrometer (MERIS) is developed as one of the payload components of the ENVISAT‐1. MERIS is a fully programmable imaging spectrometer, however a standard 15 channel band set will be transmitted for each 300 m. pixel (over land) covering the visible and near‐infrared wavelength range. Since MERIS is a multidisciplinary sensor providing data that can be input into ecosystem models at various scales, we studied MERIS performance relative to the scale of observation using simulated data sets degraded to various resolutions in the range of 12m. to 300m. Algorithms to simulate MERIS data using airborne imaging spectrometer data sets are presented, including a case study from GERIS 63 channel data over a agricultural site in central Spain (the Almaden test site). Through a pixel purity analysis, end members are derived from the MERIS‐type data and subsequently used as input to a spectral unmixing analysis yielding fraction of end member (abundance) images. The original data as well as the abundance images are spatially analyzed using variogram surfaces and mapping accuracy is modeled at various spatial scales. We observe differences between the sampling resolutions (i.e., pixel size) found to be optimal for the different ground cover types. The optimal scale for observing different components of spectral mixtures varies depending on the type of mixture, however, the best possible resolutions in all cases of mixtures studied is below the envisaged 300 m. field of view for the MERIS sensor. The analysis of semivariogram surfaces demonstrates that the spatial distribution of the variance of the mixtures is invariant with scale, thus the observed mapping discrepancies are not related to the data processing but to the observations themselves.

Implementation and performance of a general purpose graphics processing unit in hyperspectral image analysis

Abstract A graphics processing unit (GPU) can perform massively parallel computations at relatively low cost. Software interfaces like NVIDIA CUDA allow for General Purpose computing on a GPU (GPGPU). Wrappers of the CUDA libraries for higher-level programming languages such as MATLAB and IDL allow its use in image processing. In this paper, we implement GPGPU in IDL with two distance measures frequently used in image classification, Euclidean distance and spectral angle, and apply these to hyperspectral imagery. First we vary the data volume of a synthetic dataset by changing the number of image pixels, spectral bands and classification endmembers to determine speed-up and to find the smallest data volume that would still benefit from using graphics hardware. Then we process real datasets that are too large to fit in the GPU memory, and study the effect of resulting extra data transfers on computing performance. We show that our GPU algorithms outperform the same algorithms for a central processor unit (CPU), that a significant speed-up can already be obtained on relatively small datasets, and that data transfers in large datasets do not significantly influence performance. Given that no specific knowledge on parallel computing is required for this implementation, remote sensing scientists should now be able to implement and use GPGPU for their data analysis.