NASA airborne visible/infrared imaging spectrometer (AVIRIS) data were acquired in the summer of 1991 over a Norway spruce forested area in southeastern Austria for the purpose of assessing the potential of hyperspectral data to measure changes in canopy biochemical constituents related to tree damage. In support of the aircraft data acquisition, 312 needle samples were collected from 52 damaged tree stands representing low, moderate, and high levels of crown needle loss and analyzed for nitrogen, lignin, and cellulose. Multivariant statistical analysis of the AVIRIS and foliage biochemical data sets produced a set of wavelengths for estimating canopy nitrogen, lignin, and cellulose concentrations in Norway spruce stands that can be used to provide a measure of forest damage.

The airborne visible/infrared imaging spectrometer (AVIRIS) provides images of the Earth surface at the spectral resolution of 10 nm, in the range of 0.4 - 2.5 micrometers , with a pixel size of 20 m. It was designed for geological investigations, nevertheless AVIRIS has been successfully used for grassland monitoring, vegetation inspection, chlorophyll content measurements in inland waters and coastal zone environments and for atmospheric investigations. Particularly, AVIRIS data have been shown to be usable for in-path water vapor content mapping, opening a new application field for the airborne high resolution remote sensing. The data acquired during the MAC-Europe '91 campaign have been analyzed in order to realize at once both atmospheric correction and atmosphere investigation. The use of some molecular absorption bands (e.g.: H₂O features) for the investigation of the atmosphere is shown. In this connection the reliability of different spectroscopic band parameters (e.g.: band FWHM, central residual band intensity, band area, etc.) is discussed and compared. Our work shows that these band parameters can be affected by the scattered path radiance. The use of absorption bands different from those of H₂O is also briefly examined. The need of higher spectral and photometric resolutions is pointed out.

Surface reflectance retrieval from imaging spectrometer data has become important for quantitative information extraction in many application areas. In order to calculate surface reflectance from remotely measured radiance, radiative transfer codes play an important role for removal of the scattering and gaseous absorption effects of the atmosphere. The present study evaluates surface reflectances retrieved from airborne visible/infrared imaging spectrometer (AVIRIS) data using three radiative transfer codes: modified 5S (M5S), 6S, and MODTRAN2. Comparisons of the retrieved surface reflectance with ground-based reflectance were made for different target types such as asphalt, gravel, grass/soil mixture (soccer field), and water (Sooke Lake). The results indicate that the estimation of the atmospheric water vapor content is important for an accurate surface reflectance retrieval regardless of the radiative transfer code used. For the present atmospheric conditions, a difference of 0.1 in aerosol optical depth had little impact on the retrieved surface reflectance. The performance of MODTRAN2 is superior in the gas absorption regions compared to M5S and 6S.

The Italian National Research Council (CNR) in the framework of its `Strategic Project for Climate and Environment in Southern Italy' established a new laboratory for airborne hyperspectral imaging devoted to environmental problems. Since the end of June 1994, the LARA (Laboratorio Aereo per Ricerche Ambientali -- Airborne Laboratory for Environmental Studies) Project is fully operative to provide hyperspectral data to the national and international scientific community by means of deployments of its CASA-212 aircraft carrying the Daedalus AA5000 MIVIS (multispectral infrared and visible imaging spectrometer) system. MIVIS is a modular instrument consisting of 102 spectral channels that use independent optical sensors simultaneously sampled and recorded onto a compact computer compatible magnetic tape medium with a data capacity of 10.2 Gbytes. To support the preprocessing and production pipeline of the large hyperspectral data sets CNR housed in Pomezia, a town close to Rome, a ground based computer system with a software designed to handle MIVIS data. The software (MIDAS-Multispectral Interactive Data Analysis System), besides the data production management, gives to users a powerful and highly extensible hyperspectral analysis system. The Pomezia's ground station is designed to maintain and check the MIVIS instrument performance through the evaluation of data quality (like spectral accuracy, signal to noise performance, signal variations, etc.), and to produce, archive, and diffuse MIVIS data in the form of geometrically and radiometrically corrected data sets on low cost and easy access CC media.

Ecosystem modeling requires information about canopy chemistry. This is usually obtained through chemical analysis and laboratory spectrometric measurements. the potential of spectrometric remote sensing was investigated with an ISM (Imaging SpectroMeter) airborne campaign (1993, Les Landes, France). This spectrometer operates in the 800-3200 nm range. The study area consists of a mosaic of homogeneous parcels of maritime pines with a wide variety of ages (2-48 years). During the airborne campaign, 21 parcels were sampled and chemically analyzed for lignin, cellulose, and nitrogen. Samples were spectrally analyzed in laboratory with a Technicon InfraAlyser 450 and a NIR 6500 system. Correlations between the similar bands of the two spectrometers were surprisingly low. Predictive equations of nitrogen, lignin, and cellulose were obtained by stepwise regression analysis on spectral data. The stability of predictive relationships from laboratory to remote sensing level was especially analyzed. Technicon-derived predictive equations used with ISM data led to encouraging results for nitrogen and cellulose. Lignin could not be predicted NIR 6500-derived predictive equations were also tested with raw ISM data and data processed to minimize atmospheric effects. Minimization of atmospheric effects improved results for nitrogen and lignin.

The procedure PULREF, which converts optical remote sensing data to absolute reflectance values, is presented. This procedure can be applied to different hyperspectral (AVIRIS, CASI,...) and multispectral sensors (TM, SPOT,...) and for a wide range of atmospheric and flight conditions. It is based on the radiative transfer code LOWTRAN-7, which contains all important absorption features and also includes multiple scattering calculations. PULREF corrects the adjacency effect by considering the environment's influence on the signal measured by a sensor. The easy use of PULREF makes it suitable for application oriented users. The necessity of converting remote sensing data to absolute reflectance values is demonstrated in an example. On the basis of a multitemporal and multisensoral dataset, a regression between the arNDVI (NDVI based on absolute reflectance values) and the plant parameter LAI was determined. The relationship is valid for different vegetation periods and regions in Germany. The regression between the LAI and the arNDVI of airborne and spaceborne sensors coincides well with results from ground based radiometer measurements. The atmospheric correction scheme PULREF can therefore serve as a prerequisite for time and location independent extraction of land surface parameters from optical remote sensing data.

Extracting significant features is essential for processing and transmission of a vast volume of hyper-dimensional data. Conventional ways of extracting features are not always satisfactory for this kind of data in terms of optimality and computation time. Here we present a successive feature extraction method designed for significance-weighted supervised classification. After all the data are orthogonalized and reduced by principal component analysis, a set of appropriate features for prescribed purpose is extracted as linear combinations of the reduced components. The method was applied to 500 dimensional hyperspectral data which were obtained from tree leaves of five categories. Features were successively extracted, and they were found to yield more than several percents higher accuracy for the classification of prescribed classes than a conventional method does.

The characteristics of the atmospheric aerosols can be determined with different techniques. The new aureolemeter Finetec model Pom-01 and the related software are described. Combined experiments on the aerosol properties of the western Mediterranean Sea were carried out in May and June 1994, in Southern Sardinia. The optical properties of the aerosols, such as the size distribution, the optical thickness, and the single scattering phase function, are presented. The aerosols detected in this region clearly present a trimodal volume spectrum, which consists of the typical maritime mode, i.e., background plus sea salt, and possibly coarse particles of anthropogenic origin. An attempt is made to develop an algorithm for the retrieval of the aerosol optical thickness from the AVHRR-2 data.

The so-called topographic effect in remote sensed images is caused by the different orientation of the terrain surface relative to the light source and the sensor position. It produces a variation in the spectral radiances associated with a given cover type and causes problems in image classification and interpretation due to the increased in-band variance for theoretically homogeneous classes. Different techniques have been proposed in the literature to reduce the topographic effect. The method presented here is based on the modeling of the radiances reflected by a Lambertian surface. A normalization coefficient is derived from a simplified radiative transfer equation and is used to reduce, for each pixel of the image, the reflected radiances to the values they would have on a horizontal surface; the unknown atmospheric parameters are empirically evaluated by an iterative process for variance minimization of radiance values in a few significant sites. We apply the method to a Landsat TM image for which we obtained an accurate DEM. To define the value of the unknown parameters, the correction is first applied to a number of training areas, characterized by both homogeneous cover and rugged topography. Using these values the correction can be finally extended to the entire image. Misregistration of DEM and TM images and local changes in atmospheric parameters appear to be the major sources of error, as well as the deviation of the surface response from a purely Lambertian model.

The aim of this study is to present a method for improving the geological interpretation of remotely sensed data in areas which are never entirely free of cloud cover (temperate or tropical climatic zones). The method is based on the production of ortho-images and pseudo- stereoscopic pairs by means of a digital elevation model (DEM). The interpretation is done under a mirror-stereoscope on geometrically corrected images. The images could be interpreted in terms of geomorphology (volcanoes, drainage patterns, etc.), petrography (volcanic differentiates), and structure (faults, annular anomalies). By using stereoscopy it is possible, on a TM 4 5 2 color-composite to distinguish three-dimensional geological features using their spectral responses. These features are not clearly visible without stereoscopy. ERS 1 radar stereoscopy improves the hierarchization of the lineament pattern. The combination of backscattering and stereoscopic view emphasizes geological structures characterized essentially by their texture.

This paper deals with the application of a modified linear mixture model (MLM) to the satellite image classification for a precise evaluation of the landscape unit surfaces in the lagoon environments where transitional zones between continent and sea waters are marked by clusters of mixed pixels. The importance of a precise classification for these border-pixels is evident since satellite observations could become a very precious tool in the monitoring of erosion/sedimentation rate of wetlands. The study area is located in the lagoon of Venice (Italy) which has been subjected to a slow but continuous sinking since the beginning of this century causing a remarkable loss in the extension of wetlands. A data set of three Landsat Thematic Mapper passages was used, 3-year intervalled one another, and covering the period from 1984 to 1990. Validation of the adopted methodology was made by the support of aerial color photos, taken the same days of the 1987 satellite overflight.

The objective of this study is to find areas where landslides may occur in the near future by using satellite remote sensing data and thematic-map data related to landslides. We used Landsat TM data, geological maps, and inclination angles to predict landslide areas. As a result, we conclude that NVI data which was calculated from satellite data, geological types, and inclination angles are important factors to extract areas where landslides may occur.

In a series of radiative transfer calculations the Stokes vector is derived to simulate radiance and polarization measurements in the visible range for different aerosol types. The used radiative transfer model is based on the matrix operator method. The information content of the calculated multiangle Stokes vectors is extracted by a principal component analysis in order to determine the number of independent parameters that describe the system. A correlation analysis is performed between the input variables of the radiative transfer calculations and the scores of each component to obtain the link between them. The objectives of the analysis are: (1) to examine if there is additional information in polarization measurements; (2) to find out which input parameters are derivable from polarization measurements; and (3) to find the optimal viewing geometry either to use polarization effects for the estimation of aerosol types or to avoid the influence of polarization effects on measurements.

The Earth Radiation Budget Experiment consists of an array of radiometric instruments placed in Earth orbit by the National Aeronautics and Space Administration to monitor the longwave and shortwave components of the Earth's radiative budget. Presented is a high-level dynamic electrothermal model of the nonscanning Earth-viewing active cavity radiometers used to measure the Earth's total radiative exitance. High accuracy is obtained by integrating an optical and thermal radiative model with both a transient conduction model of the Suprasil^R filter dome and a dynamic electrothermal model of the active cavity. The instrument optical and thermal radiative performance is characterized by a Monte-Carlo ray- trace representation. Thermal diffusion in the Suprasil^R filter dome is modeled by a three-dimensional transient finite difference analysis. Dynamic electrothermal behavior of the active cavity is described by a two-dimensional transient finite element model. The paper emphasizes the capabilities of the current models, including the ability to perform simulated calibration runs and to simulate the so-called `solar blip' phenomenon.

A simulation system has been developed that allows computer experiments with specific sensor configurations. Application areas for such an approach are: (1) design and optimization of optical sensors for specific (well known) applications; (2) sensitivity of different data products to position and calibration errors; (3) test of retrieval algorithms; and (4) mission support. The simulation system consists of mathematical and physical models to simulate the passage of electro-magnetic radiation from the source of the emission to the sensor. The key part of this simulation system is a ray tracer interacting with a digital terrain model around an ellipsoid. The simulator consists of three parts for the following tasks: (1) With a ray-tracing procedure in a digital terrain model (DTM) a reflection image is generated. The reflection model is Lambertian. The sensor geometry can be a CCD line or a matrix camera. (2) The calculation of the radiation intensity in front of the CCD. The transmission and scattering in the atmosphere, the reflection properties of the surface, spectral filters and angular dependences of the optics have been considered. (3) The evaluation of the analog part of the camera. A statistical noise is added to each pixel. The approach is outlined with some examples.

Airborne hyperspectral radiometers provide the opportunity to obtain calibrated remote sensing data for simulation of the exact spatial and spectral resolutions of current and future satellite systems. This paper describes the use of the NASA advanced visible and infrared imaging spectrometer (AVIRIS) instrument, flown on an ER-2 jet in the lower stratosphere, to simulate shortwave radiances from the future AVHRR-K sensors on the NOAA polar-orbiting satellites. The high spatial resolution (20 m) and discrete 10 nm spectral channels of AVIRIS allow the spectral signatures of individual pixel types to be intercompared. Observations of multiple cloud types have been obtained using the AVIRIS instrument and were used in deriving estimates of cloud optical depth and particle size using visible and near-infrared spectral reflectances. These retrievals were then compared with similar retrievals from simulated AVHRR-K radiances, using spectral weighting and spatial averaging. Coincident cloud physical measurements provided verification of the cloud parameters. The analysis results are used to evaluate the impact of AVHRR pixel resolution and bandwidth on cloud retrievals.

The space borne remote sensing imaging spectrometer MOS will be launched to the Russian MIR-station in autumn 1995. It was developed for monitoring large scaled effects of the oceans, the atmosphere and land such as chlorophyll content, yellow substances, sediments, aerosol parameters, vegetation stress, and vegetation index. Information about such state parameters will be derived from the spectral function of the backscattered sun radiation of the objects under investigation. This spectral function will be measured in 13 wavelength channels from 408 nm to 1010 nm with 10 nm halfwidth for ocean, land, and atmospheric purposes and in 4 channels at the O₂-A-absorption band from 757 nm to 767 nm with 1.4 nm half width for atmospheric purposes only. The reliability of the thematic interpretation depends strongly on the accuracy of the measured data, and it depends on the on-ground and the in-flight calibration methods and procedures. The determination of the spectral sensitivity function of each sensor element, their shape as well as the absolute values on ground and the check of the stability of the scanner properties during the mission time are described.

Beginning in the mid 1960's large aperture scanning radiometers have been used in space to determine spectroradiometric properties of earth scenes in the red and near infrared regions. Panel diffusers as calibration sources for these radiometers were abandoned in favor of internally illuminated integrating spheres because of problems of illuminating the panel diffuser uniformly.1 Since 1970 spectroradiometric instruments used in space for remote sensing of the atmosphere in the ultraviolet for the determination of stratospheric ozone and total column amounts have used spectral radiance calibrations derived from calibrated panel diffusers illuminated by NIST standards of spectral irradiance. An advantage of the panel diffuser technique is simplicity of the experimental set up. Stratospheric ozone profiles and total column amounts are derived from ratios of atmospheric radiances to corresponding solar irradiances incident at the top of the atmosphere in the wavelength region of 250 - 340 nm. An inherent problem associated with measurements for the remote sensing of stratospheric ozone which is not shared with remote sensing measurements of earth scenes at longer wavelengths of the solar scattering and reflective region is the extremely large dynamic range of atmospheric radiances and the steep gradients of radiance with wavelength. For a typical wavelength scan the spectral radiance changes by about lO and the average signal level of a spectral scan can shift by another factor of 25 or more due to solar zenith angle changes between the subsolar point and the solar zenith angle limit of useful scan information which is within a couple of degrees of the terminator. The derivation of spectral radiance calibrations using either the sphere or panel diffuser techniques for ultraviolet remote sensing instruments are single point calibrations at each wavelength. A subsequent linearity calibration of the detector and electronics is made in non dispersed or white light over the entire dynamic range of the instrument of more than six decades. Consequently the derived radiometric calibration constants consist of a radiometric sensitivity term and a signal dependent linearity correction. An initial comparison of spectral radiance calibrations of SBUV-2 instruments using spherical integrator and panel diffuser techniques has been given by Heath et al.2. Subsequent work by Heath et al. describes the results from comparisons of four spectral radiance calibrations derived using panel diffuser techniques with five spectral radiance calibrations derived using spherical integrator techniques for three different SBUV-2 instruments. The comparability of the sphere and panel diffuser spectral radiance calibration techniques is assesed by comparing derived average BRDF values of panel diffusers based upon the sphere technique with laboratory measurements of BRDF of the panel diffusers. The sphere radiances determined relative to NIST standards of spectral irradiance are compared with measurements of sphere radiance relative to a NIST high temperature blackbody source. This work describes the evaluation of the consistency of spectral radiance calibration scales established using panel diffuser and internally illuminated spherical integrator techniques for the SSBUV and SBUV-2 Flight Model 5 instruments using zenith sky radiance measurements as a function of solar zenith angle (UMKEHR) which coincident in space and time. Also described are the spectral radiance calibrations of the Global Ozone Monitoring Experiment (GOME) with the NASA sphere which has been used to intercalibrate SBUV-2 and SSBUV instruments. These spectral radiance calibration constants are compared with those derived using a Spectralon panel diffuser whose BRDF was measured at NASA Goddard Space Flight Center by J. Butler.

The `modular optoelectronic multispectral/stereo scanner' (MOMS-02) is a second generation pushbroom scanner, flown during the D2 mission aboard Space Shuttle flight STS-55. Besides a three line along-track stereo device, the sensor is equipped with multispectral modules that provide data in 4 wavebands covering the visible and near-infrared range. The mission took place during April/May 1993 and provided data of approximately 8 Mio. km2. Data have been acquired in different modes at a mean altitude of 296 km that results in a mean GIFOV of 4.5 m X 4.5 m for the nadir looking panchromatic module and 13.5 m X 13.5 m for the multispectral and tilted panchromatic modules. Centering and widths of all bands for the spectral and panchromatic modules have been newly designed. The narrower and distinct positioned bandpasses allow a more detailed recording of vegetational targets and improve discrimination of various rock/soil types independently of the very high spatial resolution. The results of investigated MOMS-02 data with regard to technical parameters and distinct applications are discussed within this paper.

Most applications of spectral geology require image data to be expressed in the form of reflectance values in order to allow comparisons with field or laboratory measurements. Two problems must be solved in order to pass from the digital number (DN) provided by the satellite operator to a reflectance value: (1) the calibration factors of the DN into a physical value as luminance/radiance at satellite level (Ls), and (2) the computation of a reflectance at ground level (R_g) from this Ls. The calibration in luminance is straightforward as a gain and, sometimes, an offset factor is provided by the operator. The computation of the R_g of a target can be performed if the conditions of illumination and travel of the radiation through the atmosphere are known as well as the nature of the target and its environment. The 5S algorithm of simulation of the solar illumination calculates the radiation at the ground and at satellite level for a given target and environment, through an atmospheric model adapted to the circumstances of image acquisition. Given an environment identical or different from the target, the input variables of the algorithm (nature of the sensor, illumination conditions, type of atmosphere, type and thickness of aerosols) can be adapted for scaling Ls according to parametric R_g values. This relation is approximately linear. A scale of R_g values can thus be matched with a scale of R_s values, which can in turn be converted to DN values, or inversely.

In this paper, we present an image compression algorithm that is capable of significantly reducing the vast amount of information contained in multispectral images. The developed algorithm exploits the spectral and spatial correlations found in multispectral images. The scheme encodes the difference between images after contrast/brightness equalization to remove the spectral redundancy, and utilizes a two-dimensional wavelet transform to remove the spatial redundancy. The transformed images are then encoded by Hilbert-curve scanning and run-length-encoding, followed by Huffman coding. We also present the performance of the proposed algorithm with the LANDSAT multispectral scanner data. The loss of information is evaluated by PSNR (peak signal to noise ratio) and classification capability.

If scientists are to fully exploit the terabytes of remote sensing imagery in present and future libraries, techniques must be developed for efficient and reliable pattern matching. In this paper we investigate two technologies that will play major roles in this large-scale computing challenge. We describe a software neural network algorithm that can be used for pattern matching and test its performance for a multispectral classification task on a single processor workstation and a parallel processing machine, the CM-5. We also look at the impact of a commonly used data compression standard, JPEG, on the accuracy of pattern matching for spectral signatures. We find that accuracy degrades as expected as the compression ratio increases, but that the neural net algorithm is significantly more robust than the statistically based maximum-likelihood algorithm. Empirical results are presented from our experiments and discussed.

We present a novel and mathematically sustained procedure for solving the deconvolution- filtering problem. The technique for obtaining a stable solution to the deconvolution (restoration) problem is based on the regularized functional Newman series of the operator equations and gives more precise results than traditional ones. Several novel rank filtering procedures were proposed for decreasing of noise influences. These procedures have given more accuracy and robust results in comparison with other known ones. The efficiency of the techniques proposed has been proven by numerical simulation analysis and by experimental investigations of different kinds of RS objects such as: (1) rural or vegetation covered areas sensed by three microwave frequencies airborne equipment; (2) forest fire areas, industrial installations at night, electrical structures, etc. sensed by infrared and visible sensors; and (3) tropospheric refractive index height profiles restoration by the satellite navigation system `CIKADA' data.