Christopher C. Wackerman

A two-scale model to predict C-band VV and HH normalized radar cross section values over the ocean

This paper presents a two-scale model that can predict C-band, VV polarized (C-VV), normalized radar cross section values generated from the CMOD4 model and calibrated normalized radar cross section values derived from C-band, HH polarized (C-HH), RADARSAT-1 synthetic aperture radar imagery with coincident buoy observations of the local wind. The model is based on standard composite models that incorporate tilt modulations, with a new approach for incorporating hydrodynamic modulations. It is shown that inclusion of the hydrodynamic term does not significantly impact the fit to normalized radar cross section (NRCS) values, but does allow the model to accurately predict the upwind to downwind C-VV ratios and improves the model fit to simultaneous C-VV and C-HH NRCS observations. The model uses a new wave height spectrum that is a linear combination of the Apel and Romeiser spectra, weighted heavily toward the Apel form, and has modified values within the C-band Bragg wavenumber regime which are close to midway between the two spectra. The final model can represent the CMOD4 C-VV NRCS results with a root-mean-square error (RMSE) of 0.47 dB and can represent the RADARSAT-1 C-HH NRCS observations, without any change to the model parameters, with an RMSE of 1.9 dB. The model can accurately reproduce the upwind to downwind C-VV NRCS ratios from CMOD4 (RMSE = 0.15 dB) and fits simultaneous C-VV to C-HH ratios derived from aircraft observations to within 1 dB (RMSE = 0.65 dB). If the C-HH hydrodynamic term is allowed to be scaled differently than the C-VV hydrodynamic term, it lowers the RMSE for the simultaneous C-VV to C-HH aircraft observations to 0.49 dB while not affecting the fit to C-VV and C-HH NRCS values. Compared to other C-HH models published in the literature, this model provides a better fit to the RADARSAT-1 C-HH NRCS data across a larger range of conditions than any other single model, provides a better fit to simultaneous C-VV and C-HH NRCS data at an incidence angle of 20°, and reproduces the decreasing trend in the C-VV to C-HH ratio with increasing wind speed observed in data.

Phase retrieval and estimation with use of real-plane zeros

Locations at which the Fourier transform F(u, υ) of an image equals zero have been called real-plane zeros, since they are the intersections of the zero curves of the analytic extension of F(u, υ) with the real–real (u, υ) plane. It has been shown that real-plane zero locations have a significant effect on the Fourier phase in that they are the end points of phase branch cuts, and it has been shown that real-plane zero locations can be estimated from Fourier magnitude data. Thus real-plane zeros can be utilized in phase retrieval algorithms to help constrain the possible Fourier phases. First we show a simplified procedure for estimating real-plane zeros from the Fourier magnitude. Then we present a new phase retrieval algorithm that uses real-plane zero locations to generate a simple parameterization of the Fourier phase and uses knowledge about the image to estimate the Fourier phase parameters. We show by example that this algorithm generates improved phase retrieval results when it is used as an initial guess into existing iterative algorithms. We assume that the image is real valued.

The Discrimination Of Sea Ice Types Using Sar Backscatter Statistics

X-band (HH) synthetic aperture radar (SAR) data of sea ice collected during the Marginal Ice Zone Experiment in March and April of 1987 was statistically analyzed with respect to discriminating open water, first-year ice, multiyear ice, and Odden. Odden are large expanses of nilas ice that rapidly form in the Greenland Sea and transform into pancake ice. A first-order statistical analysis indicated that mean versus variance can segment out open water and first-year ice, and skewness versus modified skewness can segment the Odden and multilayer categories. In additions to first-order statistics, a model has been generated for the distribution function of the SAR ice data. Segmentation of ice types was also attempted using textural measurements. In this case, the general co-occurency matrix was evaluated. The textural method did not generate better results than the first-order statistical approach.

Validation of a CFAR vessel detection algorithm using known vessel locations

The National Oeanic and Atmospheric Administration (NOAA)/National Environmental Satellite, Data, And Information Service (NESDIS) is in the second year of a two-year demonstration of Synthetic Aperture Radar (SAR) derived products called the Alaska SAR Demonstration (AKDEMO). This demonstration provides near real-time SAR data and derived products, including wind images and vectors, hard target locations, along with ancillary data, to specific users in the government community. One of the derived products are vessel positions obtained from a constant false alarm rate (CFAR) vessel detection algorithm developed by Veridian ERIM. This algorithm has been tested and validated to maximize the number of ships found while minimizing the number or false alarms on one SAR image of the Red King Crab fishery in Bristol Bay on October 18, 1999. This resulted in using a detection statistic threshold of about 5.5, depending on image resolution used. Until now, this validation has been done with only general knowledge of fishing fleet size and location, but no in situ vessel information. This paper presents the results of a validation of the SAR vessel detection algorithm using observer reported vessel positions along with information on vessel size and local wind speed.

Sea Ice Type Classification Of Sar Imagery

Fully automatic algorithms have been developed which can produce sea ice type classification maps and sea ice concentration estimates of synthetic aperture radar (SAR) imagery. The sea ice type classification algorithm uses local statistics to determine ice type boundaries, and the ice concentration algorithm iteratively decomposes the histogram into ice and water histograms. algorithms have been used on both simulated imagery and actual SAR imagery gathered during the 1984 and

Estimating near-shore bathymetry using SAR

Presents two algorithms for estimating near-shore bathymetry from synthetic aperture radar (SAR) data. The first algorithm utilizes the kinematics of the shoaling waves to extract depths under the assumption of no significant current in the direction of wave propagation. The second algorithm utilizes the average radar cross section to estimate the rate of wave energy dissipation due to breaking, and then relates this dissipation to water depth. The two algorithms are complimentary in where they can estimate water depths; the kinematics algorithms works outside of the surf zone while the energy dissipation algorithm works within the surf zone so together they can span the near-shore region. The authors first present the airborne SAR data set that was used to test the algorithms, then they discuss each algorithm, including results from the SAR data set.

Validation of an automatic vessel detection algorithm using SAR data and known vessel fleet distributions

The National Oceanic and Atmospheric Administration (NOAA) National Environmental, Satellite, Data, and Information Service (NESDIS) is conducting a two-year demonstration of synthetic aperture radar (SAR) derived products called the Alaska SAR Demonstration (AKDEMO). This demonstration provides near real-time SAR data and derived products to the U.S. government community working in the waters near Alaska. One of these products is a constant false alarm rate (CFAR) vessel detection algorithm developed by Veridian ERIM International. This algorithm derives vessel positions from SAR data and delivers them within the context of the AKDEMO to users such as the U.S. Coast Guard, National Marine Fisheries Service, and the Alaska Department of Fish & Game (ADF&G), for management and regulatory purposes. The ScanSAR Wide B mode on RADARSAT-1 has an image swath width of 480 km, which covers a much larger area than is practical from ship, aircraft, and helicopter platforms. During the AKDEMO, SAR swaths are taken up to twice a day in both the Bering Sea and the Gulf of Alaska. During events such as fishery openings, extra data are also collected. A validation of the CFAR algorithm is presented using RADARSAT-1 SAR data along with ship fleet locations and sizes collected by the ADF&G at times nearly coincident with the SAR data.

Estimation of wind speed and wind direction from ERS-1 imagery

Synthetic aperture radar images over the ocean have the potential of providing wind vector information over 10 to 50 kilometer resolutions. The authors present preliminary results from an automated algorithm to extract wind speed and wind direction from SAR imagery that indicate direction errors of /spl plusmn/30 degrees and speed errors of /spl plusmn/1.5 meters/second.<<ETX>>

Calculation of the Spatial Distribution of Scatterers in a Diffuse Scene from Sar Data

One practical use of synthetic aperture radar (SAR) images is to segment large areas of land into various geophysical classes; crop types, tree types, or urban versus suburban areas for example. Since large amounts of data are usually involved automatic algorithms are the only practical approach. It would make such algorithms easier to implement if the probability density functions (pdfs) of the various classes could be modeled beforehand; in fact if the pdfs are known optimal detection algorithms can be implemented. In addition, knowledge about the distribution of scattered energy from a given scene can give information about the physical structure of the scattering surface, namely the spatial distribution of scatterers within the scene. For reasons such as these it is useful to develop a parametric model for the pdf of diffuse scenes in SAR images. Much work has been done in this area already [1,2,3] , mainly in attempting to fit pdfs with known analytical forms to histograms generated from SAR data. Unfortunately, as SAR resolutions become better (thus generating more independent samples of a diffuse scene) and SAR calibration procedures become more refined (thus removing uncertainties in backscattered values due to system effects) the analytical forms do a poorer job of fitting the actual histogram data. This indicates that a more complicated pdf model is necessary, one that includes the physical aspects of the terrain being imaged, rather then just continuing to search for other analytical forms. This paper will present a model that essentially divides the SAR pdf into a part due to speckle and a part due to the spatial distribution of the scatterers within the scene allowing a better fit to the SAR histogram values and allowing the spatial distribution function to be extracted from the SAR data. A specific form of this model was presented in ref. 2, but our approach is to generalize that model to multi-looked SAR data and to extract the spatial density function directly in addition to assuming a functional form for it.

Estimating breaker height and type in the surf zone using SAR

In synthetic aperture radar (SAR) images of coastal areas, bright streaks, or smears, can very often be observed in regions that appear to contain breaking waves. In particular, such smears are routinely seen within the surf zone. It is of great interest to determine whether these manifestations of breaking waves within the SAR imagery can be used to characterize any aspects of the surf zone. Standard SAR imaging theory can be used to explain the bright smears by relating the increase in brightness to the increase in turbulence within the breaking region, and by relating the smearing to the fact that the water in the breaking region is moving with a range of velocities which cause varying azimuth shifts in the SAR imaging process. This implies that the smear widths can be used to estimate the range of velocities occurring within the breaking region. What is needed therefore is to be able to relate these velocity ranges to surf parameters of interest, in particular to the height of the breaking wave. This paper summarizes an effort to develop an algorithm that will allow the breaking wave height to be estimated using SAR imagery. A model was developed that relates the range of velocities within the breaking region to parameters that could be estimated using SAR imagery. Procedures were then developed to estimate the model parameters from the SAR imagery. Results of applying these procedures are shown for airborne SAR images collected over a region of known water depth in order to calibrate parameters within the model and to estimate the sensitivity of the algorithm in terms of resulting errors in water depth at breaking. They indicate that it may be possible to estimate breaking wave height using SAR images.