Branch samples were collected from canopies of red spruce (Picea rubens) and eastern hemlock (Tsuga canadensis) on June 28, July 15 and September 6, 1990, and spectrally characterized by needle age class (first- and second-year). Spectral differences between spruce and hemlock first- and second-year branch segments include differences in green peak reflectance features, red edge parameters, and amplitude features of the NIR plateau. Second- year needles of both species exhibit decreased reflectance in the NIR when compared with first-year needles collected at the same time. It is also apparent from these data that phenological change has a dramatic impact on hyperspectral fine features, such as green peak and red edge parameters, as well as NIR plateau features.

The polarimetric radar image analysis is giving a great impulse in understanding the physical mechanisms involved in the interactions radiation-matter. In the summer 1991 the multifrequency polarimetric SAR on board the NASA AIRSAR flew over the Black Forest site in South Germany. The Institute for Remote Sensing Applications of the Joint Research Center is exploiting the SAR capabilities in extracting information on forest spatial structure and water content by implementing a systematic approach in which Particular Polarimetric Configurations (PPCs) characterizing the backscattering of the forest stands are obtained. The PPC is defined as the polarimetric configuration of the transmit and receive channels which produces a particular set of values of the statistical parameters (i.e. mean, standard deviations, maximum and minimum extremes) of the return power of the targets belonging to a variable- size moving window. In this way we could create a thematic information for the imaged scene and try understanding the reasons beyond the found PPC combinations. Preliminary results show that the polarimetric orientations generating maximum and minimum powers are related to the spatial orientation of the trees' branches. Investigations are still on going on the information contained in the ellipticity configurations, possibly regarding the dielectric constant of the forest components.

In this paper, we investigate the use of linear filtering techniques in the exploitation of hyperspectral imagery. In particular, we focus on applications of the simultaneous diagonalization (SD) filter to hyperspectral image analysis. The SD filter can be designed to enhance a particular feature with a known spectral response pattern and suppress undesired features. The filter is used to enhance surface features in two AVIRIS scenes acquired from the Forest Ecosystem Dynamics (FED) flightline on September 8, 1990. A spectral distance algorithm is also applied to this hyperspectral data to enhance surface targets with similar spectral response patterns. Although no detailed ground information was utilized, targets were selected by image interpretation methods and included an urban/disturbed site, and a wetland bog area, and a northern vegetation type (mixed hardwood). Both the SD filter and the spectral distance algorithm are able to effectively separate these classes of ground features from others.

Spectral reflectances were acquired above three different coniferous forest stands. Continuous visible/near-infrared (0.400 - 1.100 micrometers ) spectra were collected at multiple (8 - 12) view angles in the solar principal plane by a spectroradiometer (SE590, Spectron Engineering, Denver, CO) placed on ground-based platforms that extended above the forest tops. Data were collected over a full sun angle range in August 1991 above a 14 m high mixed spruce-hemlock stand at a research site managed by the University of Maine in Howland. Data were also collected at multiple sun angles in June/July 1992 at the Petawawa National Forest Institute in Petawawa, Ontario, Canada above two stands: a 14 m high jack pine stand and a 18 m high red pine stand. Reflectances were calculated from calibrated radiances, and corrected for panel anisotropy per sun angle. As a function of view angle per sun angle, forwardscatter and backscatter spectra from all sites were used to compute vegetation indices and Photosynthetically Active Radiation (PAR, 0.400 - 0.700 micrometers ) reflectance. These unique data sets reveal that these forest stands were highly anisotropic, both in terms of sun angle, view angle, and spectral wavebands. Additionally, the vegetation indices exhibited sun and view angle changes, with maximum values at moderately oblique forwardscatter view angles (approximately 30 - 40 degree(s)).

There is considerable interest in utilizing microwave and visible spectrum imagery for the assessment of tropical rain forests. Because rain forest spans large sub-continental areas, medium resolution (1 - 16 km) imagery will play an important role in providing a global perspective of any forest removal or change. Since 1978, AVHRR imagery from NOAA polar orbiters has provided coverage of tropical regions at this desirable resolution, but much of the imagery is plagued with heavy cloud cover typical of equatorial regions. In contrast, no historical source of active microwave imagery at native 1 - 16 km resolution exists for all the global rain forest regions. In this paper, the authors compare the utility of Seasat scatterometer (SASS) k_u-band microwave data to early-date AVHRR vegetation index products for discrimination of tropical vegetation formations. When considered separately, both the AVHRR imagery and the SASS imagery could be used to distinguish between broad categories of equatorial land cover, but the AVHRR imagery was slightly superior. When combined, the two data sets provided discrimination capability superior than could be obtained by using either set alone.

Within the context of a study to model the effects of land use on carbon storage in the Pacific Northwest region of the United States we are processing about 100 Landsat images. The image processing challenges within the study are numerous. To determine the utility of our proposed methods many of them are being tested on a 1.2 Mha area within the region. The major challenges include mapping of forest cover type and seral stage, radiometric and geometric rectification, and change detection. In this paper we present preliminary results.

Our research takes place within the framework of an expert system for automatic remote sensing image interpretation where we have to computer pixel's topographic context [DES 89]. The aim of this paper is to present a new method to locate drainage network from a digital elevation model and to compute its topographical and topological attributes. The resulting drainage network is represented by a pixel width network, such as the gray level at each network pixel represents valley's topographical attributes.

Much information from remote sensing image is fuzzy and uncertain, which greatly influences the recognition and monitoring of landscape pattern. It is more difficult to identify the types of forest landscape, due to its relative homogeneity compared to larger scale landuse type. Moreover, the categories and process of forest landscape classification, in fact, are fuzzy. In this paper, I use a fuzzy modeling approach to deal with the fuzzy phenomena during the process of identification and classification of forest landscape image. This approach also incorporates the knowledge of vegetation ecology and expert experiences into classification model. Using supervised classification of pixel-by-pixel and group-by-group, I classified a tropical forest landscape. The results indicate that the fuzzy modeling approach can effectively deal with fuzzy information from remote sensing image.

The automatic mapping of land cover from satellite imagery requires optimal classification and spatial generalization procedures. Here we describe the use of functional ink neural networks, based on a flat perceptron net with an augmented feature vector, to generate high accuracy classification products. These can then be trained more rapidly than multi-layer perceptrons. The network output is then used to fix land cover class area statistics which control a low-level generalization procedure based on a combined iterative majority filtering and reduced class growing procedure.

An automated method is proposed for terrain height estimation from SPOT stereo pair images. This method consists of two procedures: registration between stereo pair images and terrain height estimation. In the first stage, stereo pair images are roughly rectified from initially selected 4 ground control point (GCP) pairs. Automated process is used for generating GCP pairs in the second stage. For the derivation of epipolar constraint, images are registered by piece-wise Affine transformation (ATF) using these GCP pairs. Accurate rectification is made by dynamic programming (DP) matching process applied to pair of epipolar lines. Terrain height is obtained from corresponding points on the stereo pair images by using a triangulation technique. Mean height error of this method is evaluated as less than 30 m.

A model is developed to relate the attenuation through a vegetation canopy, to the geometric and dielectric parameters of the canopy constituents. The model is designed to operate over a wide frequency band in the microwave region and include both deciduous and coniferous trees. The vegetation canopy is represented by a discrete random layer of cylinders and disks having the same geometric and dielectric properties as the vegetation canopy constituents. The Foldy-Twersky Integral Equation is used to relate the attenuation to the scattering amplitude of the vegetation constituent evaluated in the forward direction. Numerical results are presented to show how the level and frequency dependance of the attenuation depend on type of vegetation constituent and its orientation.

Electromagnetic surveys using a terrain-conductivity measuring device have been used to detect individual fractures along a fault zone in a ground-water discharge area in Ash Meadows, southwestern Nevada. A fracture identified on aerial photographs and confirmed by excavation was surveyed using the conductivity meter to develop techniques in recognizing fractures. Precise coil alignment was critical in detecting the anomalies associated with the fracture. Rough desert terrain required modification of the conductivity meter, the addition of a unipod connected to the transmitter and receiver coils to enable quick, precise coil alignment. Generally, a decrease in the apparent conductivity readings in the vertical dipole configuration were observed when the transmitter coil passed over the fracture. In the horizontal dipole configuration, readings along the survey line at the different exploration depths tended to converge at the fracture, with the degree of convergence dependent upon overburden thickness.

Methods for estimating profile soil moisture using remotely sensed surface moisture data are reviewed. Primary attention is given to four basic approaches: regression, knowledge-based, inversion, and combinations of remotely sensed data and water balance models. Results of ground-based and aircraft experiments using microwave radiometers for soil moisture measurements are discussed to illustrate the methods.

This paper is presenting the results of a three year program of the Belgian Ministry for Science Policy dealing with land degradation and remote sensing. In arid and semi-arid climates the soil salinity problem is becoming more and more important. In this paper two types of simulation models are presented, allowing the prediction of future risks for soil salinity and the interconnected problem of water logging. The method was elaborated in Egypt. One model is dealing with the delta soils, the other one with the desert soils in the newly reclaimed areas.

Exploration and exploitation of local planetary resources will require the acquisition of subsurface information. Continuous profiling subsurface imaging technology will permit the rapid acquisition and interpretation of information on board automated rover vehicles. The overall aim of this research is to provide base technology for an automated ground penetrating radar GPR vision system for interpretation of geophysical data. Image processing algorithms are generally classified as low-, intermediate-, or high-level to reflect both the nearness of the algorithm to the bit image, and the sequence of algorithms that are likely to be implemented on an image. This research primarily deals with the symbolic manipulation of GPR tokens at the highest-level of processing, with minor emphasis being directed to the lower-level task of determining target shape from input consisting of the slope, length, and shape of the GPR pattern signature arms. Major emphasis was directed towards determining the real world objects corresponding to these target shapes.

The Fast Fourier Transform, (FFT) inversion method is newly introduced for discrete multilayer inversion. The method has the advantages of simplicity, short computation time and being robust with respect to noise. In this paper, the FFT inversion method is applied to a discrete homogeneous multilayer dielectric model using oblique incidence and a limited bandwidth. For a given model, a uniform transverse electric, (TE) plane wave is obliquely incident on the top of its surface. The corresponding reflection coefficient is sampled over a finite frequency range. The spectrum of the reflection coefficient is obtained by applying the FFT algorithm to its samples. An algorithm is given for the deduction of the height and permittivity of each layer. A software package is prepared for automatic execution of the algorithm during the whole inversion process. The algorithm is tested by using simulated data of a three-layer half space dielectric model.

Processing of Synthetic Aperture Radar (SAR) data can be implemented either in space domain or in transformed (spatial) frequency domain. The latter is the usual technique, due to availability of Fast Fourier Transform codes. However, for one bit coded signals direct space domain processing may be a very convenient alternative. It is noted that one bit coding does not impair amplitude and phase information, it the signal to noise ratio is small; and that processing of one bit coded sequences can be conveniently implemented by simple digital integrated circuits. A processor for one-bit coded SAR data has been designed for real-time operations: simulation and expected performance of this processor is presented.

It is well known that Synthetic Aperture Radar (SAR) images are degraded by speckle noise. Multilook processing is the traditional and popular technique to reduce it. Usually, multilook is implemented in one dimension, i.e., with reference to the azimuthal spectrum of SAR data. In this paper we extend multilook techniques to two-dimensional SAR spectra. Two different designs of parallel architectures for multilook implementation are presented and discussed.

In this paper we investigate the space-variance of the SAR system transfer function. An extremely efficient procedure to compensate this space-variant effect is provided. A number of experiments confirming theoretical results are presented.

Lossless compression is never as profitable, in terms of compression ratio, as lossy compression of the same data. However, lossless techniques that produce significant compression of geophysical waveform data are possible. A two-stage technique for lossless compression of geophysical waveform data is described. The first and most important stage is a form of linear prediction that allows exact recovery of the original waveform data from the predictor residue sequence. The second stage is an encoder of the first-stage residue sequence which approximately maximizes the entropy of the latter, while allowing exact recovery during decompression. We review the overall two-stage technique, which has been described previously, and concentrate in this paper on some recent performance examples and results using the technique. To obtain the latter, a seismic waveform data base is introduced and made available. We conclude that lossless compression of seismic data can save significant amounts of storage in seismic data bases and archives, and significant amounts of bandwidth in real- time communication of instrumentation data.

Since 1978, AVHRR imagery from NOAA polar orbiters has provided coverage of tropical regions at this desirable resolution, but much of the imagery is plagued with heavy cloud cover typical of equatorial regions. Clearly a medium resolution radar sensor would be a useful addition to AVHRR, but none are planned to fly in the future. In contrast, scatterometers are an important radar component of many future earth remote sensing systems, but the inherent resolution of these instruments is too low (approximately equals 50 km) for monitoring earth's land surfaces. However, a recently developed image reconstruction technique can increase the spatial resolution of scatterometer data to levels (approximately equals 4 to 14 km) approaching AVHRR global area coverage (approximately equals 4 km). When reconstructed, scatterometer data may prove to be an important asset in evaluating equatorial land cover. In this paper, the authors compare the utility of reconstructed Seasat scatterometer (SASS), K_u-band microwave data to reconstructed ERS-1 C-band scatterometer imagery for discrimination and monitoring of tropical vegetation formations. In comparative classification experiments conducted on reconstructed images of Brasil, the ERS-1 C-band imagery was slightly superior to its reconstructed SASS K_u-band counterpart for discriminating between several equatorial land cover classes. A classification accuracy approaching .90 was achieved when the two scatterometer images were combined with an AVHRR normalized difference vegetation index (NDVI) image. The success of these experiments indicates that further research into reconstructed image applications to tropical forest monitoring is warranted.

A dynamic neural network is used to obtain the resistivity information of geologic structures. Based on synthetic data several simulations are made to train and test the designed neural network. Error figures are reported to evaluate the performance of the network.

Toxic waste has been deposited in a wide variety of containment structures. Leakage creates concerns and environmental risk. Risk assessment requires site characterization of the underlying geology and monitoring of fluid pathways to the biosphere. Subsurface imaging will play an increasing important role in site characterization and monitoring. Imaging will play an even greater role in remediation. Radio wave energy propagation has been applied in crosswell data acquisition and tomography inversion algorithms have reconstructed images of electrical conductivity variations in the geologic zone of concern. The variation in conductivity delineated structural geology and mapped the contaminant plume. This paper describes the application of the radio wave tomography imaging in chemical waste landfill and in situ mine site characterization and remediation.

A mathematical evaluation of the potential of active microwave sensors for monitoring underground permafrost is carried out. For this purpose, the Helmholtz Integral is used to relate the scattered field to the field inside the permafrost. A mathematical formation is derived for the scattered field by estimating the inner field through a generalized Rayleigh- Gans approximation. Numerical calculation show that the freezing thawing process is the dominant factor determining the relative level of the backscattered signal.

An inference of landslide areas using digital elevation data and Landsat TM band 6 data has been performed on the basis of the assumption that the occurrence of landslides is closely related to the amount of underground-water and the topographic features of watersheds. It is shown that it is possible to distinguish between dangerous landslide areas and non-landslide areas by using spatial features of watersheds, such as the area, mean slop and shape factor, and the ground surface temperature obtained from Landsat TM data.

The Ku-band (14.6 GHz) Seasat scatterometer (SASS), which flew for 3 months in 1978, was designed to measure the normalized radar backscatter coefficient ((sigma) ⁰) in order to determine the near-surface wind over the ocean. While SASS made measurements of (sigma) ⁰ over land and ice regions, the application of this data has been limited due to the low resolution (50 km) of the measurements. Recently, however, we developed a new technique to generate enhanced resolution (to as fine as 3 - 4 km) images of the surface radar backscatter characteristics from SASS measurements. In this paper we report some results of a study of the seasonal response of the Greenland ice sheet using this technique. Using SASS data, we have generated a time-series of enhanced resolution images of vertically-polarized radar images of Greenland during July - Sept. 1978. The effects of the summer melt along the ice sheet periphery are clearly evident in the time series. In central Greenland, which exhibits a very high radar backscatter at 14.6 GHz, very little change was observed over the three month data set.

A novel sensor is described in this paper for measuring high frequency weak E-fields. The sensor employs a bulk crystal EOM inside the laser cavity by which the optical frequency is mixed to obtain a digitized signal. This E-field sensor has high sensitivity and large dynamic range.

The problem of determining the internal structure of the earth from the frequencies of free oscillation is considered. The simplified case of a spherically symmetric fluid sphere is examined. An asymptotic analysis implies that a velocity profile that is a small perturbation from a homogeneous fluid is determined, up to a finite number of parameters, by two infinite spectra of appropriate angular orders. A numerical study of the problem for finite spectra indicates what types of spectral data allow a stable reconstruction.

The use of tomography methods for data observation, processing, and fiber-optic methods for distribution measurement network design will have great influence on the development of real- time MSP methods. It makes it possible to decrease the information channel number and to improve the signal-to-noise ratio of the information channel due to light carrier use and a combination of sensing and transmitting function in a common fiber-optic tract. The object of this paper is development and investigation of a fiber-optic distribution measuring network using tomography methods of data observation and processing. This network could solve the problem of measurement and reconstruction of the parameter distribution function of non- stationary physical fields.