ERS-1 SAR data were used to study the potential to map fungi diseases and other forest damage. ERS-1 SAR images were processed into a multitemporal geo-coded data set. Visual interpretation of color composite images were used to identify areas affected by fungi diseases or other forest damage. Eleven test points were inspected on the ground. All non-peatland test points identified as damaged suffered from some kind of stress (fungi diseases, elk damage, drying, or use as pasture land). The test points on peatland suffered from fungi diseases even though the backscatter had increased. The preliminary results suggest that the C-band SAR aboard the ERS-1 satellite seems to have potential for mapping forest damage on mineral soil.

The capabilities of the ERS-1 SAR for discriminating clearcuts and partial cuts from open fields, wetland, and surrounding conifer, hardwood and mixedwood forest stands were tested. The test site contains both boreal and more temperate species. Two winter, one spring, four summer, and one fall image were registered to a UTM-corrected thematic mapper image. Separation of the backscatter signatures of different cut and forest conditions was analyzed visually and statistically. Discrimination and mapping capabilities varied with ground cover condition and, in some cases, environmental conditions. Overall separabilities were generally poor; consistent detection of the clearcut areas with the ERS-1 image dataset was not possible. Partial cuts were not detectable. The highest clearcut/forest and clearcut/wetland separabilities were observed in summer and winter imagery, although the results were not consistent. For example, two summer images acquired twelve hours apart provided quite different results. Some clearcuts were detectable on an image acquired at 2300 h, nine hours after a heavy rainfall. A second image acquired earlier that day at 1100 h during a light rainfall did not show the clearcuts. Clearcuts with bare sand or sparsely vegetated surfaces were the most likely to be detected. As ground vegetation density increased, discrimination generally became poorer. Clearcuts with rugged topography and rocky soils were not clearly separable from surrounding forests on any of the images.

The aim of this project was to investigate the potential value of ERS-1 SAR data as a tool for monitoring tropical rainforest conversion and land use planning. Furthermore an easy to use data analysis procedure was developed which is operational under conditions found in many countries of the tropical belt. KFA 1000 photographs from the Russian MIR space station and ground truth data acquired during an intensive field survey served as reference. The analysis showed that several land use classes and different forest types can be identified in flat terrain even in a single ERS-1 scene. The results obtained so far give reason to establish the use of ERS-1 data as a tool to monitor deforestation processes and land use in tropical rain forest environments.

In this paper a hybrid technique of image classification, modeling, and inversion algorithm is introduced in order to extract vegetation biomass from polarimetric SAR data. The development of the hybrid technique has evolved from SAR data analysis and the mere fact that the SAR signal, being correlated with vegetation type and moisture content, saturates as the biomass increases. The technique consists of the following steps: (1) classification of the image to land cover map, (2) classification of the SAR image into scattering mechanisms, (3) formulation of a multilayer forest backscatter model, (4) piecewise inversion of the model to estimate the vegetation water content for various components of the forest canopy, and (5) estimation of the forest biomass by combining the inversion and classification results. This technique has been tested over the boreal forest in Canada by using the multipolarization, multifrequency AIRSAR data. The results indicate that the use of the hybrid technique enhances the estimation of the forest biomass. The general form of the technique and its various components can be regarded as a basic approach to resolve the problem of biomass estimation with SAR data. The modification of this technique into more sophisticated forms can improve the biomass estimation in particular over dense tropical forests.

Soil moisture exerts an important control on the transfer of energy and water at the land surface and its availability determines, to a large extent, the development of vegetation. Satellite remote sensing using synthetic aperture radar has the potential to monitor spatial and temporal changes in surface soil moisture provided the effects of, primarily, vegetation and surface roughness are understood. The results obtained from ground validation measurements undertaken during an annual vegetation cycle in contrasting temperate and semi-arid climates enabled the sensitivity of the SAR to be tested over a wide range of soil moisture and vegetation conditions. In the UK, flat, grassland sites were used to simplify the modeling of vegetation over three contrasting soil types. Both manually read and data-logged instruments were used to monitor moisture conditions both at the surface and within the soil moisture profile for modeling their interrelationships over time. Vegetation parameters were determined through destructive sampling. In Niger the soils were of almost pure sand under both the natural vegetation of fallow savannah and the main crop of millet. As a result of the high permeability of the sand, the timing of SAR data acquisition relative to the last rainfall event was believed to be an important factor affecting radar backscatter. For both areas, seasonal variations in radar backscatter were present and evidence to demonstrate the contribution of the vegetation and soil components is given.

Water stored in the soil serves as the reservoir for the evapotranspiration process, thus the interest in trying to map its spatial and temporal variations in experiments studying the soil- plant-atmosphere interactions at the GCM grid scale. During the 8 week intensive observation period (IOP) of HAPEX-Sahel (Hydrologic Atmospheric Pilot Experiment in the Sahel), this was done with two airborne microwave radiometer systems. The five frequency (5 to 90 GHz) PORTOS radiometer on the French ARAT aircraft and the single frequency (1.42 GHz) multibeam pushbroom microwave radiometer (PBMR) on the NASA C-130 were used. These aircraft measurements were supported by ground based observations at the central sites and, because of several rains during the IOP, covered a good range of soil wetness conditions that existed. The PBMR and the 5.05 GHz PORTOS channel in H polarization show a large dynamic range of TB on each day and between different days in response to variations in rainfall and drying conditions ranging from low TBs of 210 to 220 K for the wettest conditions to values of 280 to 290 K for the driest.

Microwave brightness temperatures (T_B) were measured over three soils at frequencies of 5 GHz ((lambda) equals 6 cm) and 1.67 GHz ((lambda) equals 18 cm). Field measurements were made with a small, lightweight radiometer system which was found to be quite portable and easy to maneuver and operate with simple hand constructed platforms. The soil plots were wetted to near field capacity and T_B measurements were made during the dry-down cycle. Soil profile temperature and moisture content were monitored continuously for the duration of the experiment. Laboratory analyses of soil physical properties such as texture and bulk density were performed to aid in making theoretical calculations of soil dielectric and emissivity. Instrument calibrations were performed periodically by making T_B measurements of the sky, over water, and of an absorbing black-body material. Theoretical calculations of the soil dielectric based on measured soil physical properties compared reasonably well to values reported in the literature. Soil emissivities derived from the field measurements were also compared to emissivities based on theoretical calculations.

All remotely sensed soil moisture data have to be calibrated by reference measurements on ground. The gravimetric sampling (oven drying) is not suitable. It is time consuming, expensive, and destroys the site. Also the neutron probe has less and less acceptance. The remaining possibility is the dielectric moisture determination by rf or microwaves. The optimal measuring frequency for such a device is between 100 MHz and 1000 MHz. Our previously published frost-calibrated in-situ sensor uses 250 MHz. The difference to the used remote sensing frequencies, in L-C- and X-bands is too large. In addition the stationary in-situ device measures only its closest neighborhood. The remotely sensed pixel however is much larger. Therefore, we developed an intermediate sensor for a good compromise in these discrepancies. With frequencies of 540 MHz and 940 MHz it is intermediate to the in-situ sensor (250 MHz) and to the used L-C-X-bands. It is a Goubau-type open-ended surface waveguide, a simple dielectric coated stick, touching the soil surface. The reflected wave on the Goubau-line is a function of the moisture from a well-defined area under the line (0 0.4 m). So, it is intermediate concerning the measuring area, as well. We need approximately 10 minutes for scanning an area of 100 m². This time is short enough. On the other side, this surface moisture sensor will be calibrated by a very accurate frost-calibrated in-situ sensor. An overall system accuracy of 6% can be achieved by this two step indirect ground truth.

This paper reports preliminary results of a study using L-band SAR imagery to estimate soil moisture and surface roughness for bare fields in a mountain watershed. The small perturbation model requires that the surface roughness profile deviates only slightly from that of a smooth surface. This condition is seldom met by most natural surfaces. Thus, the classic treatment of the scattering model has only limited usage. In this study, we show a simplified model derived from the integral equation model for estimation of soil moisture and surface roughness. We then test this model using JPL L-band AIRSAR data.

Research on the analysis of ERS1/SAR data for hydrological applications has been carried out in the Abruzzo region aiming at the integration of microwave (MW) remote sensing (RS) derived data with territorial databases for soil degradation studies. Three different field campaigns were carried out at the same time as the ERS1 overflights collecting dielectric constant, soil moisture content, and surface roughness data. Electromagnetic models (EM) were used to predict the backscattering coefficient of bare fields where ground truth data were collected. Such data were therefore compared with the ones calculated from the SAR images.

Within a project sponsored by DARA, 28 ERS-1 SLC (single look complex) images of an agricultural area in the upper Rhine Valley were processed and integrated into a GIS structure. The GIS contains information on field-boundaries, land-use, soils, and results of continuous soil-moisture and rainfall measurements on several fields in the test site (approx. 50 km²). The GIS is used to establish a relationship between ERS-1 backscatter and measured surface soil-moisture for different land-uses. Since backscatter depends on soil and surface properties, statistical relations between temporal backscatter profiles of different soil- surface complexes are established. They allow us to normalize the multitemporal ERS-1 images to a reference backscatter behavior of corn on loamy sand. Since surface soil-moisture was measured for this soil-surface complex, the ERS-1 images can be converted into reference surface-moisture images. These images are interpreted. Irrigated areas can be identified by their high reference surface-moisture.

Backscattering data obtained from ERS-1 SAR imagery are used to verify the ability to detect field scale spatial and temporal variations of surface soil moisture from spaceborne microwave sensors, and to test sub-field scale detection capability. The investigated area is part of the Sele basin, in southern Italy. For two dates, corresponding to satellite overflights of the area during November, 1993, Time Domain Reflectometry (TDR) measurements were taken over a dense sampling grid in two bare soil fields, in order to acquire ground-truth information for soil moisture in the upper 10 cm soil layer. Statistical analysis shows a good agreement between the two types of data at the field scale; at the sub-field scale, a similarity in the spatial structure ofTDR and radar signal is found through cross-covariance and covariance analysis. In addition, one of the dates gives the possibility of observing the response of the SAR sensor and of the TDR for a field showing quasi-uniform ponding conditions.

This paper discusses the approach and results of using a forest succession model and a microwave vegetation canopy scattering model to determine expected remote sensing signatures as the forest changes through time. We used the forest model to generate characteristics such as species, stem diameter and height for individual trees growing on a 0.01 ha patch over a 500 year period. The simulations were repeated fifteen times under random initialization conditions to provide a measure of variance. These data were used to parameterize a radar backscatter model and radar signatures were calculated for stands at 28 time intervals over the 500 year period. The results are discussed in terms of the radar backscatter response to changes in forest structure and the implications for radar remote sensing of forests. Our results show that a gap type forest growth model can be used to provide inputs for a radar backscatter model. The predicted backscatter can then be checked against actual SAR imagery as a test of both the forest and the radar backscatter models. Successful coupling of these models provides an effective tool for understanding backscatter measurements from forested areas, as well as, a capability to initialize and test forest ecosystem models.

The modeling of microwave backscatter from forests is considered. A simulation is implemented where every tree is viewed as a collection of cylindrical segments whose size and position are known. The total backscatter is then computed by adding the scattered field from each segment. The calculations are carried out using red pine tree data, provided by R. Landry of the Canada Centre for Remote Sensing. By using a surveying instrument, he has generated the coordinates for all segments of a typical red pine tree. The simulation accounts for trunk and branch curvature and tapering which are not usually included in the statistical approach. In addition, array effects at the level of first order scattering are accounted for. The calculations are performed at P band where multiple scattering contributions should be small.

An inversion algorithm based on a nonlinear optimization technique is used to retrieve a subset of forest canopy parameters that can be used in ecosystem modeling. The inversion algorithm uses a parametric model derived from a discrete component forest scattering model which includes all major scattering contributions to radar backscatter. The parametric model, however, is derived for cases where only one mechanism dominates the scattering process, in this case, branch layer volume scattering. The free parameters are taken to be the real and imaginary parts of the dielectric constant of branch layer constituents. The parametric model and the inversion algorithm are validated with synthetic data and applied to two data sets taken at different dates over one of the OTTER sites (Santiam Pass). The inversion algorithm is shown to be a useful tool in the quantitative monitoring of canopy moisture status, as well as a building block of its future versions which will include a more comprehensive set of forest scattering parameters as unknowns.

This paper describes a model that characterizes a forest canopy with three-dimensional structure and considers the spatial position of tree and ground surface components. A forest stand is modeled as an array of cubes. A cube may make up a portion of crown, trunk, gap, and ground surface components as determined by the shape, size, and position of trees. The model uses ray tracing to calculate the scattering from the forest scene as (1) direct crown backscatter, (2) direct trunk backscatter, (3) direct soil surface backscatter, (4) trunk-ground scatter, and (5) crown-ground scatter. An image was produced from elements of the backscatter Stokes matrices calculated for each image pixel. The sensitivity of the model to the spatial pattern of forest stands was tested by modeling radar backscattering coefficients from two simulated 1-ha forest stands with the same trees, but positioned differently. A detailed data set including tree species, size, and position for a 200 by 150 m area of spruce forest in Maine, USA was used to parameterize the model. Model results compared favorably to JPL AIRSAR data over this stand.

Radar data from the remote sensing technique have been used in conjunction with theoretical microwave modeling to develop a retrieval algorithm for the root mean square height of the rough surface. The algorithm exploits frequency (L and C band) differences in the radar response from a vegetated rough surface. These differences are related back to the Fresnel reflectivity and surface rms height by using a theoretical modeling approach that is based on a discrete scatter random media technique and uses distorted Born approximation to compute backscatter coefficient from a particular scene. Sensitivity analysis shows that the change in surface reflectivity due to the change in frequency from L to C band is dominated by surface rms height, and, the Fresnel reflectivity stays almost constant over this frequency interval. The inversion algorithm based on these sensitivity differences has been applied to the backscatter model data from a plant canopy of soybean. Calculations show that the technique gives accurate results from a model backscatter data set that is corrupted with 80% of noise. The inversion algorithm is also applied to synthetic aperture radar (SAR) data collected over corn fields during the MACHYDRO'90 experiment in Pennsylvania, USA. There is an excellent agreement between the measured and the retrieved rms surface height.

This paper presents the results from a study on the use of ERS-1 SAR data for snow parameter extraction in mountainous areas. Eight ERS-1 SAR data sets and data from four field measurements from the Kvikne area in south-Norway have been available. The data have been acquired during the snow melt period from April until June 1992. The SAR data have been calibrated and geocoded using Digital Elevation Model (DEM) data, and analyzed in terms of evaluating the relationship between the backscattering coefficient, (sigma) _o, and snow conditions. We have observed a decrease in (sigma) _o of approximately 3 dB between dry and wet snow, which indicates that ERS-1 SAR data can be used for monitoring the extent of a wet snow cover.

An algorithm for geometric and radiometric correction of ERS-1 synthetic aperture radar images by using digital elevation model data is described and applied to images over a mountainous area in Norway. Several images covering the same area are corrected and geocoded into UTM map projection and the results are evaluated against map and in-situ measurements. The map to SAR geometric precision is given by a rms value of 54 m in the east and 45 m in the north. The SAR to SAR geometric precision is given by a rms value of 43 m in the east and 44 m in the north. Estimates of the radar backscatter coefficient are extracted from the geocoded SA backscatter images. Obtained values are for wet snow -13 dB and for dry snow -7 dB which agrees favorably with earlier radar measurements.

With passive microwave snow data the most prominent error feature is the underestimation of snow mass during the winter, especially for North America. The reason why the microwave data underestimates snow mass primarily has to do with the effects of vegetation above snow fields. With the microwave data the emissivity of trees, especially dense conifers, can overwhelm the scattering signal which results when upwelling microwave energy is redistributed by snow crystals. The boreal forests which stretch across the northern tier of North America are perhaps the physiographic region where most of the difference occurs between the snow depth measurements based on climatological data and those based on microwave observations. Forests not only absorb some of the radiation scattered by snow crystals, but trees are also emitters of microwave radiation. So in forested areas the signal received by a radiometer on-board a satellite is produced by a combination of media. Generally, the denser the forest, the higher the microwave brightness temperature despite the type and condition of the media underlying the forest canopy. Furthermore, because the canopy shields the snow from direct solar radiation the deepest snow accumulates in the densest forests. However, if the fractional forest cover of a given microwave pixel can be accounted for in some way then microwave algorithms can be modified by including a forest cover parameter and estimates of snow depth will be improved. In this study we have used a vegetation index, derived from satellite brightness data, as an indicator of forest cover, and preliminary results show that this refined algorithm compares more favorably with climatological snow depths.

A Monte Carlo ray tracing model for canopy birdirectional reflectance is applied to forest canopy architectures consisting of inhomogeneous mixtures of coniferous and both shade- tolerant and intolerant hardwood species. The model assumes a multi-layered canopy of known optical properties, statistical composition and geometric arrangement and, for the results reported here, further assumed Lambertian scattering of canopy elements. Upward and downward direct, diffuse and multiply scattered solar irradiance is traced through the canopy and from a reflecting background. The stand-level architecture for the model was derived from a forest growth succession model which predicted canopy structure height profiles as a function of site condition and environmental driving variables. Fifteen theoretical successional stages consisting of five age class distributions for three different site indices were simulated using the forest growth model. Fifteen forest growth replications for each of the forest plots were generated. Monte Carlo reflectance predictions were then made for each of these theoretical cases with ten canopy reflectance replications per plot. The model was applied to field and aircraft data collected September 8, 1990 from a hemlock-spruce-fir forest canopy at the Northern Experimental Forest study site near Howland, Maine, the location of a NASA sponsored Forest Ecosystem Dynamics Multi-sensor Aircraft Campaign. Model predicted bidirectional reflectance distribution functions are compared to representative samples of AVIRIS (airborne visible/infrared imaging spectrometer) observations corresponding to successional age class distributions of 50, 75, 100 and 200 years for dry mesic and wet sites.

A new model of radiation transfer in complex three-dimensional media has been developed by taking advantage of recent advances in ray tracing computer graphics methods. This model is initially dedicated to the study of radiation transfer problems in terrestrial environments over a variety of spatial scales including: (1) the milli-scale, which represents explicitly the structure and optical properties of scattering elements; (2) the meso-scale, which ranges from the structure of a single plant to that of a vegetation canopy; and (3) the macro-scale, which includes topography and large-scale landscape heterogeneities. This model can be used to generate reflectances of complex targets, where all the geometrical and physical quantities can be controlled explicitly. This opportunity is exploited here to generate and evaluate the performance of bidirectional reflectance (BRDF) models. First, we have checked the accuracy of our code with regard to well-established models. Next, we used it to create realistic synthetic reflectance data sets. Finally, we test the invertibility of a simple BRDF model against those reflectance data.

The Li-Strahler geometric-optical model has been used with some success to simulate the bidirectional and hemispherical reflectance of forest canopies. Although the canopies modeled thus far may have contained trees of widely varying heights, the underlaying terrain has always been represented as level ground. Recently, however, the model has been extended to accommodate sloping terrain in its computation of the forest bidirectional reflectance distribution function (BRDF) and spectral surface albedo. With the model accommodating topography, it has been possible to compute and compare the reflective character of realistic forest canopies on a variety of slopes and aspects.

A new model for the bidirectional reflectance of a vegetation cover combines principles of geometric optics and radiative transfer. It relies on gap probabilities and path length distributions to model the penetration of irradiance from a parallel source and the single and multiple scattering of that irradiance in the direction of an observer. The model applies to vegetation covers of discrete plant crowns that are randomly centered both on the plane and within a layer of variable thickness above it. Crowns assume a spheroidal shape with arbitrary height to width ratio. Geometric optics easily models the irradiance that penetrates the vegetation cover directly, is scattered by the soil, and exits without further scattering by the vegetation. Within a plant crown, the probability of scattering is a negative exponential function of path length. Within-crown scattering provides the source for singly-scattered radiation, which exits with probabilities proportional to further path-length distributions in the direction of exitance (including the hotspot effect). Single scattering provides the source for double scattering, and then higher order pairs of scattering are solved successively by a convolution function. As an early exercise in validation, the model is applied to an open jack pine canopy and ground-level irradiance is predicted with good accuracy.

The DART (discrete anisotropic radiative transfer) model simulates radiative transfer in heterogeneous 3-D scenes; here, a forest plantation. Similarly to Kimes model, the scene is divided into a rectangular cell matrix, i.e., a building block for simulating larger scenes. Cells are parallelipipedic. The scene encompasses different landscape features (i.e., trees with leaves and trunks, grass, water, and soil) with specific optical (reflectance, transmittance) and structural (LAI, LAD) characteristics. Radiation directions are subdivided into contiguous sectors with possibly uneven spacing. Topography, hot spot, and multiple interactions (scattering, attenuation) within cells are modeled. Two major steps are distinguished: (1) Illumination of cells by direct sun radiation. Actual locations of within cell scattering are determined for optimizing scattering computation. (2) Interception and scattering of previously scattered radiation. Diffuse atmospheric radiation is input at this level. Multiple scattering is represented with a spherical harmonic decomposition, for reducing data volume. The model iterates on step 2 for all cells, and stops with the energetic equilibrium. This model predicts the bi-directional reflectance factors of 3D canopies, with each scene component contribution; it was successfully tested with homogeneous covers. It gives also the radiation regime with canopies, and consequently some information about volume distribution of photosynthesis rates and primary production.

The model of leaf reflectance PROSPECT was used to test the capability to describe a leaf water stress. Two types of water stress were set up: a strong and sudden stress and a moderate and progressive one. The measurements of chlorophylls a and b contents, water content and equivalent water thickness assessments, and the estimation of the N leaf structure parameter supplied the three parameters of entry necessary for a functioning model. So it was possible to simulate the reflectance spectrum in visible, near, and mid infrared domains. Calculated values were significantly correlated to measured values whatever the type of applied deficit. Nevertheless, correlations were more significant in the strong deficit case. In the progressive water stress case, the analysis showed best correlations for deficient plants (R² equals 0.869) than for the unstressed controls (R² equals 0.716). These results show the problem of the single nonmeasured parameter N. So we tried to determine the value of the N parameter form the anatomical section observation of the limb in the end of a flowering stage.

The reflectance spectrum of green vegetation is mainly determined by the leaf content in chlorophyll and water, the respective spectral signatures being modulated by the structural characteristics of the leaf and the canopy. However, correlation between the reflectance spectra and the leaf content in such constituents as lignin, cellulose, nitrogen, starch, etc. has been demonstrated. To further study this question, we have constituted a data set associating high quality spectra (of single leaves, optically thick stacks, needles, stalks, on fresh and dried material) with a number of physical and chemical measurements (leaf thickness, water content, chlorophyll, carotenoids, cellulose, lignin, proteins, nitrogen, starch). This data set has been used to investigate the link between the optical properties and the composition, focusing on the biochemical components. Two approaches have been followed: the classical regression analysis and modeling based on the Kubelka-Munk formula. In the latter case, empirical specific absorption coefficients have been determined and then used to decompose the infrared spectra in water and other components contributions. This method is successful in retrieving the relative water content but does not yet allow us to estimate the biochemical content. It was applied both on laboratory and AVIRIS spectra.

In previous studies, desert-fringe vegetation densities were assessed in an ungrazed semi-arid rangeland in Utah and in an animal exclosure in the arid Sinai. Applying satellite measurements of reflectances, the dark vertical cylinders (DVC) model for desert-scrub (characterized by predominantly vertical architecture) was inverted. In the present study a new plane-parallel model is presented, which treats the canopy as a layer of small Lambertian spheres (SLS), or small facets (leaves) with a specific distribution of the leaf area, where the Schonberg function of the angle between the solar beam and the view direction specifies the reflectance from the canopy. The SLS model is inverted with the satellite-measured reflectances of the Sinai exclosure and the surrounding overgrazed, practically bare-soil terrain. The SLS-model inversion results are compared with DVC results. Both models provide plausible canopy characterizations, but the SLS model is more realistic when viewing from the zenith with sun at a high elevation. The reflectance ratio of the dark plant-elements to the bright soil is key to assessing the density of the plants. In the inversion of the AVHRR visible and near-infrared data, the plant element reflectances in the infrared are adjusted so that the plant optical density in the infrared matches that determined in the visible spectral region. Early in the dry summer season (after the winter rains), high infrared reflectances are inferred, but sharply lower infrared reflectances, appropriate for the dried out plants, are found in the later stages of the dry season. This result, that the physical changes in the plant conditions can be assessed, is highly encouraging for our SLS modeling effort.

In 1998, two significant components of NASA's Mission to Planet Earth program, Landsat 7 and the first Earth Observing System, morning equator-crossing platform (EOS AM1), are each scheduled to be launched into 705 km, sun-synchronous orbits. Landsat 7 will carry the enhanced thematic mapper plus (ETM+) sensor, while the premier instrument on the EOS AM1 platform will be the moderate resolution imaging spectrometer (MODIS). In addition, NASA is currently assessing an option to fly a Landsat ocean color instrument (LOCI) on the Landsat 7 spacecraft along with the ETM+. The Landsat ETM+ will: (a) cover a ground swath of 185 km, (b) provide 30 m multispectral coverage over six bands in the visible (VIS), near infrared (NIR), and shortwave infrared (SWIR) regions, as well as a 15 m panchromatic (PAN) band and a 60 m thermal IR (TIR) band, and (c) yield global repeat coverage every 16 days. By contrast, the LOCI, would, if flown: (a) cover a usable ground swath of approximately equals 2000 km, (b) provide approximately equals 1.1 km spatial resolution at nadir in eight VNIR bands equivalent to those of the SeaWIFS instrument, (c) yield global coverage every two days at 12 bit quantization and with four available gain settings such that unsaturated radiance data could be collected over land surfaces, and (d) be tiltable +/- 20 degree(s) along-track to minimize sunglint effects over water bodies. The MODIS sensor on the EOS AM1 platform will: (a) cover a ground swath of 2048 km, (b) provide approximately equals 1 km spatial resolution at nadir in 29 bands covering the VNIR, SWIR, and TIR regions at 12 bit quantization, plus two VNIR bands at 250 m and five VNIR/SWIR bands at 500 m spatial resolution, and (c) yield complete global coverage every two days. Plans call for the two spacecraft to be placed in a same-day orbit, spaced 30 minutes apart, i.e., equatorial crossing times of 10:00 a.m. for Landsat 7 and 10:30 a.m. for EOS AM1. Thus, multispectral data having both high (30 m) and medium-to-coarse (250 m - 1000 m) spatial resolution will be acquired repetitively on a global basis under nearly identical atmospheric and plant physiological conditions. It is expected that significant benefits will be realized by utilizing these data in a multistage, multispectral sampling approach. In fact, NASA has recently endorsed the concept of jointly flying an advanced technology TM-class instrument with MODIS on the EOS AM2 platform scheduled for launch in 2004. Much of the following text was extracted from a white paper (Sellers and Williams, 1994) that was instrumental in leading NASA Headquarters to endorse this concept.

The NOAA polar low earth orbit operational satellites carry HIRS2/MSU instruments and thus have thermal and microwave remote sensing capability. The measurements have been processed for retrieval of surface and atmospheric meteorological parameters using a version of the Goddard Laboratory for Atmospheres (GLA) interactive forecast-retrieval-analysis system. This interactive approach has properties that make it desirable for derivation of climate data sets, because it provides for accurate treatment of the effects of clouds on the IR radiances and elimination of some systematic errors form the retrieved quantities. The use of the microwave MSU cloud-penetrating wavelengths adds significantly to the remote sensing capabilities for the retrievals, as it allows for use of thermal data under partial cloud cover as high as 80%. Interannual differences of atmospheric temperatures derived from HIRS2/MSU on NOAA 10 show good quantitive agreement with values obtained from radiosonde reports. As part of the TOVS Pathfinder program, the retrieval system will be frozen to produce a 15 year (1979 - 1994) set of satellite-derived surface and atmospheric parameters. Data for the years 1986, 1987, and 1988 based on the current retrieval analysis are now available.

In June 1992 the ASMODE-project was started on the assessment of remote sensing techniques for the monitoring of the extent and progression of desertification in the Mediterranean area. Data of several satellite-sensors (NOAA-AVHRR, METEOSAT and LANDSAT) are researched on their usefulness for the monitoring of desertification. Two approaches to monitor desertification are followed in the ASMODE-project: (1) The energy and water balance approach to monitor vegetation activity. Satellite measured land surface temperatures and albedo's are used to estimate aridity parameters like net radiation, actual evapotranspiration, and rainfall. (2) The direct monitoring of the vegetation cover using the vegetation index. The project consists of a one-year monitoring experiment generating satellite derived datasets of Spain to be analyzed in connection with field survey data and existing datasets in a GIS. KNMI was responsible for the `NOAA-AVHRR' part of the project. In this paper the methodology used by KNMI and final NOAA-AVHRR results are presented.

Spectral reflectance measurements have been made over sugar beet crops from a helicopter during 1991, 1992, and 1993 using a portable multichannel spectrometer system. In 1994 the studies were extended to demonstrate the potential for the measurement of stress in other crops. The observations are made from an altitude of about 150 m over the spectral range 420 nm to 810 nm, with a bandwidth of 5 nm. Downwelling solar irradiance and upwelling reflected irradiance are monitored by the multichannel spectrometer simultaneously. Both the absolute values of the reflectance at each wavelength and the variance of these reflectance values across each plot are shown to be related to the state of the crop. Concurrent agricultural ground truth consisting of fresh leaf weight and dry matter accumulation, is used in defining the crop yield models. The study aims to determine the appropriate radiometrically derived parameters which could be used as alternative model inputs. Although significant spectral differences exist and can be extracted by conventional band ratio or singular value decomposition techniques, the variance in the samples of ground truth data constrain the ability to define meaningful radiometric parameters. Improved experimental procedures are proposed.

The Normalized Difference Vegetation Index (NDVI) has been generated from data from the advanced very high resolution radiometer (AVHRR) for nine rangeland sites in Tunisia. Over one hundred images have been utilized to provide information about the vegetation at approximately monthly intervals for the period September 1983 to August 1992. These data were used to validate a rangeland biomass production model which has been developed for the nine rangeland sites. The model used daily rainfall data, estimates of potential evapotranspiration, and a series of site-specific vegetation parameters in order to estimate actual daily transpiration and an index of growing season yield. Although these outputs are not directly related to NDVI they are functions of the same plant tissues which indices such as the NDVI measure. A significant relationship (p > 0.05) was found between the NDVI and estimates of transpiration for all but one of the sites. The relationship between NDVI, integrated over a growing season, and the model derived yield index were not, however, found to be so strong.

Ten years of NOAA GAC data over the Argentinean Pampa were analyzed in relation with climate and crop production. Correlations between crop yield and monthly NDVI (cumulated or not, weighted by the global radiation or not) reached 0.87 for wheat, 0.85 for soybean and 0.83 for corn, despite the classical limitations of AVHRR data (mixed response, atmospheric and directional noise, sensor calibration), the monthly frequency and the size of the test areas (10,000 km²). The quality of these results was partly due to the extensive character of the Pampa's cropping system since the correlation between final yield and NDVI relies on the following two hypothesis: NDVI can predict biomass and biomass is a good indicator of final grain yield. The best correlations were observed with the NDVI sensed at maximum green biomass, hence permitting yield estimations one to two months before harvest. Standard errors of regression were of 0.22, 0.17, and 0.63 t/ha for wheat, soybean, and maize respectively, for a mean yield around 1.7, 2.2, and 3.8 t/ha, respectively. Last, the complement between NDVI data and crop physiologically based models was examined. Despite the data related limitations, the relationship between CERES wheat predicted LAI and NOAA monthly GAC NDVI appeared as promising.

Models of surface temperatures of two land surface types based on their energy budgets were developed to simulate the effects of environmental factors on thermal radiant exitance. The performance of these models is examined in detail. One model solves the non-linear differential equation for heat diffusion in solids using a set of submodels for surface energy budget components. The model performance is examined under three desert conditions thought to be a strong test of the submodels. The accuracy of the temperature predictions and submodels is described. The accuracy of the model is generally good but some discrepancies between some of the submodels and measurements are noted. The sensitivity of the submodels is examined and is seen to be strongly controlled by interaction and feedback among energy components that are a function of surface temperature. The second model simulates vegetation canopies with detailed effects of surface geometry on radiant transfer in the canopy. Foliage solar absorption coefficients are calculated using a radiosity approach for a three layer canopy and long wave fluxes are modeled using a view factor matrix. Sensible and latent heat transfer through the canopy are also simulated using nearby meteorological data but heat storage in the canopy is not included. Simulations for a coniferous forest canopy are presented and the sensitivity of the model to environmental inputs is discussed.

This paper presents a functional characterization of West Africa's ecological systems using satellite data. It shows the capability of spatio-temporal determination of the hydrological budget well fitted to the chlorophyllian phenology. The method includes two parts: estimation of radiation budget at the earth surface, and actual evapotranspiration and, in addition, estimation of one hydrological budget and its capability to describe and possibly to predict growth process. The estimation of evaportranspiration is made with simplified relationships proposed by Jackson and Seguin. These descriptions need a limited number of input data, and their parametrization can be deduced from the main classes of vegetation. A hydrological budget model is then established using evapotranspiration, rainfall measurement, and the estimation of ground contribution at a suitable scale. In this model, we assume that the phenological state of the plant at a given time depends on the efficient stock of water during the months preceding the state development. The results show that the concept of the estimation of radiation budget at the earth surface and actual evapotranspiration can be applied generally for large areas at fine spatial resolution and the model of efficient water shows a good adjustment with the cycle normalized difference vegetation index.

In the field of scientific cooperation between our two institutes started in 1986, a common research program was developed for the analysis of the characteristics and the application of the spaceborne optical sensors operating in high spectral resolution mode for the remote sensing of the Earth. The Priroda module will be launched and attached to the inhabited space platform MIR in 1995. By means of the optical sensors placed on board the Priroda module, we are going to study the geophysical parameters useful for pollution monitoring and for resource evaluation. The principal attention will be paid to the study of the following topics: (a) water quality in the coastal zone and particularly near the river estuaries; (b) vegetation stress due to the anthropogenic activities; (c) geophysical studies in areas of geothermal and volcanic activities; (d) estimation and verification of the atmosphere contributions. In order to perform the goals mentioned above, the Italian side has identified some test sites in the Tuscany region. Taking into account the characteristics of different sensors and utilizing both experimental data available now and coming from similar sensors placed on aircraft or spacecraft and suitable models, we are estimating the possible applications of such new sensors.

The third satellite in the miniature sensor technology integration (MSTI) program includes two infrared imaging sensors with filter wheels, and a visible imaging spectrometer, with a common telescope and gimballed scan mirror assembly. These sensors will provide high spatial and spectral resolution imagery in the visible and 2.5 - 4.5 micrometers spectral regions. These spectral regions can provide information on the atmospheric aerosol optical thickness and size distribution; the vegetation index and crop classification; upper atmosphere temperatures, and tropospheric water vapor. In addition, the highly accurate pointing and tracking capability of the scan mirror permits a long look at particular events such as volcanic eruptions, major oil-spills, desert dust storms, forest fires and air pollution episodes. This tracking capability could also support ground truth calibration of other satellite sensors.

Maps of the spatial distribution of different forest types of humid tropical forests are urgently needed for studies of biodiversity, ecology, deforestation and to provide inputs to global circulation models. Recent attempts to map the tropical forests of South America have relied on NOAA-AVHRR data which, with a 1.1 km spatial resolution, is suitable for mapping vegetation at regional and continental scales. Such attempts have, however, generally focused on identifying the forest/non-forest boundary and on determining rates of deforestation and have failed to differentiate different humid tropical forest types. A new sensor -- the ATSR-2 - - launched in 1994 has advantages over the AVHRR. The ATSR-2 has a similar spatial resolution to the AVHRR but its improved radiometric resolution in the visible and near- infrared regions mean that subtle differences in the spectral response of different forest canopies may be detected. In addition, the ATSR-2 has the ability to near-simultaneously view the target area from two different angles providing the scope for reducing atmospheric effects in the data. This capacity may also afford valuable information about the structure of the forest canopy. This paper describes the work currently being undertaken to utilize these characteristics of the ATSR-2 instrument to identify and map humid tropical forests in South America. The project is, in particular, examining how the differences in the (1) phenology, (2) structure, and (3) size, shape and frequency of gaps in the canopy influence the spectral response of different forest types. To date the project has concentrated on determining these parameters for tropical forests in Bolivia and extensive ground observations have been made for different forest plots.

Models using NOAA AVHRR data for estimating the areal distribution of biomass over the Boreal coniferous zone are developed. The method uses corresponding models that are estimated using Landsat TM data as input, and includes a mosaicing scheme for AVHRR images to cover very extensive areas. The estimation method may be adjusted to some other forest characteristics too.

A regional Sahelian grassland model has been designed with the objective that satellite data acquired in different regions of the electromagnetic spectrum can be used complementarily to drive the model. The evolution of the main processes and vegetation structural variables over a growing season are simulated by using standard meteorological data and a limited number of site-specific parameters. In order to analyze relationships between vegetation characteristics and satellite derived parameters, simulated vegetation variables are used as input parameters into physical models of reflectivity to compute VIS/NIR reflectances and radar backscattering coefficients. Reflectances corresponding to NOAA/AVHRR channels and NDVI profiles are simulated on a weekly basis for the entire growing season. Radar backscattering coefficients corresponding to the ERS-1 configuration are also computed. In both cases, area-weighted contributions from vegetation and bare soil are combined to compose the responses of heterogeneous surfaces. Simulated satellite data profiles are in good concordance with NOAA/AVHRR and ERS-1 wind scatterometer data. The complementarity of these two types of satellite for driving the grassland model data are then indicated.

The purpose of this study was to analyze the synergy between optical (Landsat TM) and microwave (ERS-1 satellite SAR) data sets, used singly and in combination, to discriminate and identify Eucalyptus stands and other cover types. The study site of interest is located in the vale do Rio Doce, State of Minas Gerais, Brazil. Reference data used in this study included detailed map and stand information (forest inventory data and management records). From the individual stand records, critical information such as cutting and planting dates were determined which show the age of the stands at the time both Landsat TM and ERS-1 radar data were obtained. The data sets were geometrically corrected, registered, and resampled to a 30 meter grid using cubic convolution interpolation. A maximum likelihood classifier was used in this study. The evaluation of the classifications was both qualitative and quantitative. Results indicated that the Landsat TM used alone was the most effective sensor system to classify Eucalyptus stands (accuracy of 94.8%), followed by the combined data set (accuracy of 85.9%). The combination of TM and ERS-1 data was generally not as effective as the use of Landsat alone, but did enable areas under clouds to be classified. The ERS-1 C-band radar data alone did not provide satisfactory results, with a classification accuracy of only 35.7%. This low accuracy was due, in part, to topographic effects. The use of ERS-1 SAR data appears to be severely limited in areas of rolling terrain, due to the variations in backscatter caused by the topographic effects.

Statistical relationships between site fertility type, proportion of deciduous trees, tree stem volume and multisource image data have been studied. The image data consisted of ERS-1 SAR images, Landsat TM image, NOAA AVHRR LAC image, airborne gamma radiation measurement data, digital soil map data, and geochemical data analyzed from soil samples. The 40 km by 30 km study area, centered at 61 degree(s)10' N, 25 degree(s)08' E, was in Southern Finnish Boreal forest. The ground truth data were boundaries of forest stands and the stand characteristic values for each stand. The analysis methods were linear regression analysis, discriminant analysis, and t-test. In regression analysis, the best model with one predictor variable, two predictor variables and so on were computed up to a model with ten predictor variables. At every step, all the potential predictor variables were tested. The best single image feature was the 1.5 micrometers channel 5 of Landsat TM. The best image variable type after Landsat TM data was the SAR image that had been acquired in March during snowing. Generally, winter SAR images were better than the summer image. SAR data produced a marginal utility of approximately 5 percentage units in estimation of tree stem volume. Use of several SAR images simultaneously increased the performance of estimation only little. On mineral soil lands, simple models gave results almost as good as more complicated models. On peatlands the performance of the models clearly increased when the number of predictor variables increased. The benefit of complicated models as compared to Landsat models only varied from 5 to 15 percentage units.

The objective of the present study is to assess the capability of the ERS-1 wind scatterometer to monitor vegetation at regional and global scales. To this end, multitemporal profiles of scatterometer data, acquired over the main vegetation types of the Africa continent, are compared to GVI data for a period extending from April 1992 to November 1993. Illustration of the effects of vegetation parameters and soil characteristics on the scatterometer response are given. Results show a good correlation between GVI and scatterometer data acquired at large incidence angles, particularly around 40 degree(s). The largest seasonal dynamic is observed over semi-arid areas like the Sahelian zone, indicating the suitability of the ERS-1 wind scatterometer for vegetation monitoring in such areas.

To test the synergy between optical and microwave remote sensing data sets for vegetation analysis, a comparison was carried out between the results of vegetation land cover classification using multitemporal landsat thematic mapper (TM) alone, and then in conjunction with a Canadian airborne C-band synthetic aperture radar (SAR) image gathered as part of the South American Radar Experiment (SAREX'92). These data sets cover the Tapajos National Forest area of the Brazilian Amazon (Para State). Occurring within the area are many land use and cover types, including extensive tracts of undistributed humid tropical forest, large pastures, small scale agriculture, abandoned plantations and secondary forest growth on old agricultural fields. The addition of radar backscatter and texture information (HH and VV polarizations) to optical data sets significantly increased the separability of classes. For instance, VV backscatter was much higher in areas of permanent agriculture versus those of smaller rotational fields. However, the complexity of the radar backscatter information requires sophisticated analytical capabilities that are only now in development. The synergistic use of active and passive sensors holds a broad promise of solving some of the analytical needs for the global change and carbon modeling communities that cannot be solved with optical data without intensive field validation and/or extensive multitemporal data sets.

In this comparative study, the Bayesian and a neural network (the HLVQ) approach are used to classify multispectral LANDSAT images. The studied area contains several agricultural classes (wheat, flax,...). Some classes are found to be non homogeneous and thus are divided in this study into several subclasses. The Gaussian assumption needed by the Bayesian classifier is thus justified by this division. The main result obtained in this study is that the Bayesian classifier and the neural network considered here provide equivalent solutions for the classification of agricultural multispectral images.

A method is defined and tested for the classification of forest species from multi-spectral data, based on singular value decomposition (SVD) and key vector analysis. The SVD technique, which bears a close resemblance to multivariate statistic techniques has previously been successfully applied to the problem of signal extraction from marine data. In this study the SVD technique is used as a classifier for forest regions, using SPOT and landsat thematic mapper data. The specific region chosen is in the County Wicklow area of Ireland. This area has a large number of species, within a very small region and hence is not amenable to existing techniques. Preliminary results indicate that SVD is a fast and efficient classifier with the ability to differentiate between species such as Scots pine, Japanese larch and Sitka spruce. Classification accuracy's using this technique yielded excellent results of > 99% for forest, against four background classes. The accuracy's of the individual species classification are slightly lower, but they are still high at 97 - 100% for the SPOT wavebands. When the Landsat TM bands 3, 4, and 5 were used on their own, accuracies of 95 - 100% were achieved.

The purpose of this study was to assess the potential for using multi-date Landsat TM for monitoring major changes in Eucalyptus forest plantations and other cover types. Landsat TM data obtained in 1989 and 1992 were used to assess changes in a forested study site in the Vale do Rio Doce, State of Minas Gerais, Brazil. Both Landsat TM data sets were normalized using the Backward Radiance Correction Transformation Model (BRCTM), geometrically corrected, registered, and resampled to a 30 meter grid prior to the analysis. The data sets were also put through a histogram equalization process, so that both data sets would have an equivalent range of digital number values. The spectral/temporal change classification technique was used in this study and the results were evaluated both qualitatively and quantitatively. Results indicated that the test polygon accuracy using the twelve band (raw data) data set was 90.6%, whereas the test polygon accuracy using six bands (PCA) was 90.4%, indicating that both provide a good potential for evaluating forest change.

Groundwater management nowadays is mainly governed by hydrogeological aspects of groundwater occurrences. However, focussing on groundwater as a natural resource of the landscape, there is a deficit in spatial information about the ecological factors influencing the water balance. It becomes more important to assess the ecological role of groundwater in accordance to its quality and quantity as well as to predict risks by changes in the landscape such as lowering groundwater level and nitrate leaching. Due to high spatial diversity in the pattern of land use, relief attributes and soil properties, ecological data have to distinguish different homogeneous site units. Remote sensing data with high resoluted geometry would have the potential to improve the spatial information about landscape and its water balance. In many regions an aggravating deficit between the groundwater consumption and the recharge of groundwater can be observed. This situation is characterized by two effects: (1) Increasing groundwater consumption. (2) Decreasing recharge of groundwater. A higher level of surface sealing can be found in agriculture as well as in urban areas. Evapotranspiration in plant production has increased. By intensification in agriculture plant production has generated more biomass and crop yield. But for every unit of yield and biomass, the plant has to transpirate additional soil water. Therefore, higher intensity in farming is related to higher water consumption by the process of evapotranspiration.

The analysis and quantification of deforestation and forest degradation require diffuse ecological and spatial concepts such as forest fragmentation and forest/non forest interfaces. Anthropic factors associated with natural geophysical conditions are leading to a wide range of fragmentation patterns: their characterization and measurement represent the objectives of this study. Based on visual concepts, a qualitative approach is undertaken to propose a topology of fragmentation patterns: four types (massive, linear, insular and diffuse) are identified and illustrated in a catalogue with 34 samples selected all over the tropical belt. In a second step, some mathematical indicators, based on the number of forest/non forest pixels interfaces and the number of edge pixels, are developed and tested to quantify forest fragmentation. In the TREES Project, AVHRR classifications have to be ultimately calibrated with TM data by linear regressions between the forest cover percentages. Within and in between regions, the regression parameters present a high variability which is expected to be significantly reduced by stratifying the AVHRR classifications. This study further investigates on the most sensitive fragmentation indices affecting the regression slopes.

A methodology for using Landsat TM images from different imaging dates for mapping European forest ecosystems is described. The methodology has been applied to a forested area in central Ireland. The multi-date images are first corrected for the effects of varying sun position, topography and atmospheric conditions. The imagery is classified using a database of pre-established spectral-phenological profiles for the different forest types. Classification accuracy compares favorably with existing ground inventory data. The radiometric correction successfully reduces variation in the spectral signatures caused by changes in topography and atmospheric conditions. This research is being carried out as part of the FIRS (forest information for remote sensing) program at the Joint Research Centre, Ispra.

New techniques for information extraction from multispectral data require physical modeling to understand the energy transfer at the atmosphere/surface interface and to develop appropriate inversion procedures, in combination with advanced processing techniques. A multi-step procedure is proposed in this work: the first step implies a binary decision about the second step to be applied in each case. If the pixel is considered as being a `pure' pixel, through a spectral/spatial classification procedure based on multiresolution techniques, then numerical inversion techniques, based on a multiple-scattering reflectance model, are used to extract parameters representing specific surface properties, which can only be defined for homogeneous pixels. If the pixel is considered from the first step as being a mixture of different elements, a linear unmixing approach is used to extract the amount of each component. For such heterogeneous pixels, effective values for the parameters are defined. Multiresolution techniques are applied to improve information extraction algorithms. The relation between spectral and spatial integration effects is discussed. Broad-band spectral integration is analyzed taking into account spatial integration over heterogeneous areas. The problem of the spatial scale is also considered, as the development of these techniques is closely related to the spatial/spectral resolution of the data. Applications of these algorithms to AVIRIS/TMS and Landsat TM/NOAA AVHRR data are an illustration of the capabilities and limits of such approaches.

Spectral mixture analysis (SMA) has been applied to radiometrically corrected Landsat-TM and AVIRIS data for estimating green vegetation fractional cover in a test area of Southern France, using reference endmember spectra taken from a library of ground high spectral resolution measurements of vegetation elements and various soil and rock surfaces. Results are expressed as relative contributions of shade, green vegetation and background on a pixel by pixel basis. Results of SMA appear similar to the calculation of a simple spectral vegetation index (NDVI), but can give more interpretable results in terms of scaled fractions of green vegetation or soil and rock backgrounds.

The paper presented here is an assessment of the performance of the linear mixture modeling. It has already been tested over a spatially complex area in Burkina-Faso (West Africa) to retrieve temporal profiles of reflectances of the different cover types of the land use. Using simulated coarse resolution data by the degradation of SPOT images gives the opportunity to check all the parameters of the model and to testify to its reliability and its sensitivity. The results show that linear mixture modeling enables the radiometric values to be estimated to an acceptable degree of accuracy. The robustness of the inversion of reflectances has been evaluated at different qualities of registration. It was also shown that the model is invariant with scale which is very important for its adaptivity.

In order to reduce the problems of clouds, atmospheric variations, view angle effects, and the soil background variations in the high temporal frequency AVHRR data, a compositing technique is usually employed. Current compositing techniques use a single pixel selection criterion of outputing the input pixel of maximum value NDVI. Problems, however, exist due to the use and to the imperfection of the pixel selection criteria of the algorithm itself. The NDVI was found to have the maximum value under an ideal observation condition, while the single pixel selection criterion favors the large off-nadir sensor view angles. Consequently, the composited data still consist of substantial noise. To further reduce the noise, a new set of pixel selection filtering was proposed for compositing NDVI data. It consists of taking into account the biological characteristics and eliminating the temporal variations of the signal which are false. For this we used a mixed method of physical threshold and geometrical operations (GAPF: geometrical and physical filter). The association of the two criterion leads to a mixed `physical and biological' method. The alternative approach showed that the high frequency noises were greatly reduced, while more valuable data were retained.

Linear mixture modeling was applied to NOAA-AVHRR data (bands 1, 2, and 4 X two dates) over the Pampa region, Argentina, to derive fraction images of winter crops, summer crops and pastures. First, the signatures of the three classes were extracted on sets of 3 km wide calibration windows by regressing the mixed NOAA response on the class proportions, estimated from classified Landsat TM data. The class signatures were found consistent with the cover types, but a variability depending on the set of windows was noted. Subpixel classifications of the NOAA data were then performed using the different sets of class signatures. Although some discrepancies between NOAA and Landsat estimates were observed at the window level, the classification results for winter crops, summer crops and pastures compared well at regional scale with Landsat estimates, hence confirming the potential of linear mixture modeling applied to coarse resolution data. However, the extension of the classification beyond the calibration zone was hampered in our case by the poor preprocessings of the NOAA data.

Desertification of the Damagaram region (Sahelian zone of Niger) is represented by several processes: decrease or modification of the vegetation, deflation of sandy soils by the wind and exhumation of lateritic crusts, modification of the pattern of dunes, increase of erosion by water. A field survey was performed by a pluridisciplinary team in order to observe and define the expressions of desertification. The spectral signatures of various soils, rocks and vegetation were measured using a high resolution field spectroradiometer. The measured spectral properties of well defined objects were used to validate the calibration of the satellite data and to define the endmembers in models of spectral unmixing. Landsat multi spectral scanner (MSS) scenes of March 11, 1973 and January 10, 1976, and a thematic mapper (TM) scene of January 4, 1987, were made compatible, and several techniques of image processing were tested and compared: ratios (leaf area index, normalized difference vegetation index, redness, clay index), principal components analysis and spectral unmixing. The interpretation of the resulting images allowed a semi-quantitative monitoring of the evolution of the significant components of the landscape. Local variations of soils and vegetation are particularly well enhanced, whereas the global situation did not change drastically from 1973 to 1987.

Spectral unmixing is referred to in textbooks as a straightforward technique the application of which encounters apparently no problem. Operational applications are however scarce in the literature. The method usually used is based on the least square method of minimizing the error in search of the best fit solution. This method, however, poses problems when applied to real data when the number of end-members increases and/or the composition of end-members is similar. An alternative method based on linear algebra has several advantages: (1) no inversion of matrix is required, no meaningless values are thus generated; (2) not only a condition of the closed system can be introduced, but the end-members remain independent (i.e., the last one is not the complement to 1 of the sum of the other, as in the least square method); (3) a condition of positive value of the weights can be imposed. The latter condition yields a supplementary equation to the system, one more end-member may be taken into account, thus improving both the qualitative and the quantitative aspects of the mixture problem. Examples based on Landsat TM imagery are shown in the fields of vegetation monitoring (subtraction of the vegetal component in the landscape) and spectral geology in arid terrains (end-members being defined through a principal components analysis of the image).

Hydrological modeling is important for a wide range of operational forecasting activities in water resource management. The aim of this paper is to assess the capabilities of Earth observation sensors in relation to modeling data requirements in order to identify future areas of development in both model and sensor specifications. Models range from simple black boxes to distributed physically based models. There is significant variation in the data required and the ways in which these data are used. This range of requirements is compared with the capabilities of existing Earth observation sensors in order to define the current situation. Progress requires the coordinated development of both the sensors and the models, together with a greater understanding of the relationship between measurement and process scales. As a result, existing obstacles to progress in both areas are reviewed with the aid of specific case studies. This analysis leads to a set of recommendations on how to develop the use of sensor data in models.