Ronald J. Holyer

Toward universal multispectral suspended sediment algorithms

Abstract A data acquisition and analysis program has been undertaken to demonstrate the feasibility of remote multispectral techniques for monitoring suspended sediment concentrations in natural water bodies. Two hundred surface radiance measurements (400–1000 nm) were made at Lake Mead with coincident water sampling for laboratory analysis. Water volume spectral reflectance is calculated from the recorded surface radiance and volume reflectance-suspended sediment relationships investigated. Statistical analysis indicates that quantitative estimates of nonfilterable residue and nephelometric turbidity can be obtained from volume spectral reflectance data with sufficient accuracy (based on U.S. Environmental Protection Agency standards) to make the multispectral technique feasible for sediment monitoring. Algorithms exhibit sufficient universality to indicate they can be implemented in many cases with little or no ground truth for calibration.

Coastal bathymetry from hyperspectral observations of water radiance

Abstract Water depth, bottom reflectance, inherent optical properties of the water column (scattering, absorption, and fluorescence), and illumination conditions combine to determine the upwelling spectral radiance of coastal waters. If these complex optical relationships could be quantified, it would be possible to extract coastal information from spectral radiance data. We use data from the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) in a neural network system to establish quantitative, empirical relationships between one of these parameters, depth, and remotely sensed spectral radiance. Data are analyzed for two areas: the western coast of Florida in the Tampa Bay area and the Florida Keys between Upper Matecumbe and Plantation Keys. The neural network approach results in retrieval of reasonable depths from spectral radiance in both cases over a depth range of 0 to 6 m. Retrieved depths for Tampa Bay are accurate to a RMS error of 0.84 m relative to depths in the National Ocean Survey (NOS) Hydrographic Database, and the Keys retrievals have an RMS error of 0.39 m relative to a bathymetric survey conducted to support this study. A neural network trained on a combination of the two data sets results in a combined RMS error of 0.48 m, nearly the same performance as neural networks trained individually. The ability of the neural network to generalize, producing algorithms with some degree of universality among diverse coastal environments is, thereby, demonstrated. The result of the generalization analysis is of practical importance because it indicates that the neural network may not require an extensive training set of water depth data in order to be “tuned” for each location where depth retrievals are desired. While empirical, the neural network is in some sense a model of the inversion of the radiative transfer problem within the marine environment. The neural network approach, therefore, operates on a higher level than more traditional statistical curve fitting solutions for retrieval of remotely sensed information.

Wavelet-based feature extraction from oceanographic images

Features in satellite images of the oceans often have weak edges. These images also have a significant amount of noise, which is either due to the clouds or atmospheric humidity. The presence of noise compounds the problems associated with the detection of features, as the use of any traditional noise removal technique will also result in the removal of weak edges. Recently, there have been rapid advances in image processing as a result of the development of the mathematical theory of wavelet transforms. This theory led to multifrequency channel decomposition of images, which further led to the evolution of important algorithms for the reconstruction of images at various resolutions from the decompositions. The possibility of analyzing images at various resolutions can be useful not only in the suppression of noise, but also in the detection of fine features and their classification. This paper presents a new computational scheme based on multiresolution decomposition for extracting the features of interest from the oceanographic images by suppressing the noise. The multiresolution analysis from the median presented by Starck-Murtagh-Bijaoui (1994) is used for the noise suppression.

Histogram-based morphological edge detector

Presents a new edge detector for automatic extraction of oceanographic (mesoscale) features present in infrared (IR) images obtained from the Advanced Very High Resolution Radiometer (AVHRR). Conventional edge detectors are very sensitive to edge fine structure, which makes it difficult to distinguish the weak gradients that are useful in this application from noise. Mathematical morphology has been used in the past to develop efficient and statistically robust edge detectors. Image analysis techniques use the histogram for operations such as thresholding and edge extraction in a local neighborhood in the image. An efficient computational framework is discussed for extraction of mesoscale features present in IR images. The technique presented in the present article, called the Histogram-Based Morphological Edge detector (HMED), extracts all the weak gradients, yet retains the edge sharpness in the image. A new morphological operation defined in the domain of the histogram of an image is also presented. An interesting experimental result was found by applying the HMED technique to oceanographic data in which certain features are known to have edge gradients of varying strength. >

Comparative study of two recent edge-detection algorithms designed to process sea-surface temperature fields

Two algorithms used for the detection of fronts in satellite-derived sea-surface temperature fields are compared. The two algorithms produced surprisingly comparable results considering the substantial differences in the two approaches: multilevel (Algorithm 1) versus locally based (Algorithm 2). Algorithm 1 offers the advantage of shorter run times. Algorithm 2 can be made faster if one is willing to accept less reliable edge detection. Algorithm 1 also offers the advantage of being adaptive and therefore automatic in its application to different data sets. However, when direct control with regard to detection of the edges is demanded, Algorithm 2 contains two tunable parameters to select the smoothness and the strength of edges, while Algorithm 1 as presently coded does not. >

Noise and temperature gradients in multichannel sea surface temperature imagery of the ocean

Abstract A multispectral algorithm is routinely used to correct for atmospheric effects in NOAA AVHRR data to produce multichannel sea surface temperature (MCSST) imagery. However, the resulting imagery are of poor quality (i.e., with increased noise levels and reduced SST gradients) in comparison with the original Channels 4 and 5 imagery. The quality reduction is especially apparent in imagery of low thermal contrast areas. We propose that the increased noise results from the amplification of AVHRR Channels 4 and 5 random noise in the MCSST processing (thus the use of more channels, such as AVHRR Channel 3, will only increase MCSST noise). In support of this hypothesis, a model is presented for predicting MCSST noise levels from Channels 4 and 5 noise levels. The model is evaluated using two sample AVHRR image sets. Documentation of diminished sea surface temperature gradients in the MCSST imagery in comparison with the Channels 4 and 5 imagery is also presented and discussed. However, the exact cause for the reduction is not clear.

Sensitivity of satellite multichannel sea surface temperature retrievals to the air‐sea temperature difference

The advanced very high resolution radiometer multichannel sea surface temperature (MCSST) retrieval technique provides global algorithm accuracy statistics generally showing a bias of less than 0.1°C and an rms error of less than 0.7°C when compared to colocated drifting buoy in situ data in the absence of aerosols. This remaining error is not always random but is shown to be correlated to the air-sea temperature difference. The MCSST technique is modeled and then compared to in situ data to show this dependency. Atmospheric radiative transfer calculations are used to provide a simulation of satellite retrieval sensitivity to air-sea temperature differences. Buoy sea surface temperature (SST) and air temperature observations are then presented as experimental verification of the simulation results. Retrieval errors depend both on the mean air-sea temperature difference conditions present in the data set used to empirically derive the algorithm and on the changes in air-sea temperature difference conditions relative to the derivation data set mean conditions. Retrieval error is found to respond linearly with air-sea temperature difference changes. MCSST retrieval errors of 1.0°C can occur for air-sea temperature difference changes of 7°–10°C from mean conditions when the dual-window (channels 3 and 4) or triple-window (channels 3, 4, and 5) algorithms are used. The split-window (channels 4 and 5) MCSST algorithm is shown to be less sensitive to air-sea temperature differences. Cross-product SST (CPSST) and nonlinear SST (NLSST) algorithms are also examined. These algorithms generate results similar to the MCSST algorithm for the dual- and triple-window equations. However, the CPSST and NLSST split-window algorithms demonstrate greater sensitivity to air-sea temperature difference changes than do the MCSST split-window algorithm. Retrieval errors of 1°C can occur for air-sea temperature difference changes of 10°–12°C from mean conditions. Users of satellite SST retrievals in regions that experience large fluctuations in air-sea temperature difference should be aware of this possible error source.

Velocity vectors for features of sequential oceanographic images

This paper investigates a fundamental problem of determining the position, orientation, and velocity field of the Gulf Stream in time-varying imagery. The authors propose an approximation method to characterize the deformation of these image motions for the purpose of estimating the velocity field of these images. The technique is focused on the interpretation of the change in the extracted features of the Gulf Stream. The underlying technique employs a triangulation of the region by a simplicial approximation of the velocity field on each triangle. A generalized computational framework, an outline of the mathematical foundation, and an implementation are presented.

Two-dimensional variability in porosity, density, and electrical resistivity of Eckernförde Bay sediment

Image-based analysis techniques are employed to compare sediment density heterogeneity as interpreted from X-radiography and electrical resistivity measurements. Assessments of sediment heterogeneity in two dimensions through the use of the two methods, with some exceptions, yield similar results. Autocorrelation functions estimated from density fluctuations reveal an anisotropy between vertical and horizontal sediment structure indicated by horizontal correlation lengths being greater than vertical correlation lengths. In addition to the anisotropy, discrepancies between values of correlation lengths calculated from two statistical methods are indicative of the scale-dependent nature of the sediment structure. This scale dependence is also exemplified by the differences between the images of X-radiography and electrical resistivity. The unequal volume of sediment over which the respective methods are integrated produces results differing by the amount and type of information included in each image. Electrical microresistivity is more sensitive than X-radiography to many smaller-scale variations in sediment density. Combining the results from X-radiography and microresistivity imaging allows the calculation of Archie’s m parameter, an extremely sensitive indicator of sediment microstructure. In addition, electrical microresistivity gives a measurement of sediment tortuosity coincident with sediment porosity and density.

Sediment density structure derived from textural analysis of cross-sectional X-radiographs

Two image texture analysis procedures, autocorrelation and binary run length, quantify the scales, orientation, and isotropy of density fluctuations imaged in X-radiographs of sediment cross sections from the Arafura Sea (Australia) and Eckernförde Bay (Germany). Image texture-based results agree with traditional bulk methods in some cases and disagree in others. Advantages of imagebased techniques over bulk methods are nonintrusiveness of the approach and the ability to produce more detailed characterizations of the spatial variability present in sediment structure. The image texture-based parameterizations of density structure are interpreted with respect to environmental processes.