Richard L. Bankert

A Comparative Evaluation of Sequential Feature Selection Algorithms

Several recent machine learning publications demonstrate the utility of using feature selection algorithms in supervised learning tasks. Among these, sequential feature selection algorithms are receiving attention. The most frequently studied variants of these algorithms are forward and backward sequential selection. Many studies on supervised learning with sequential feature selection report applications of these algorithms, but do not consider variants of them that might be more appropriate for some performance tasks. This paper reports positive empirical results on such variants, and argues for their serious consideration in similar learning tasks.

Cloud Classification of AVHRR Imagery in Maritime Regions Using a Probabilistic Neural Network

Abstract Using Advanced Very High Resolution Radiometer data, 16 pixel × 16 pixel sample areas are classified into one of ten output classes using a probabilistic neural network (PNN). The ten classes are cirrus, cirrocumulus, cirrostratus, altostratus, nimbostratus, stratocumulus, stratus, cumulus, cumulonimbus, and clear. Over 200 features drawn from spectral, textural, and physical measures are computed from the pixel data for each sample area. The input patterns presented to the neural network are a subset of these features selected by a routine that indicates the discriminatory potential of each feature. The training and testing input data used by the PNN are obtained from 95 expertly labeled images taken from seven maritime regions; these images provide 1633 sample areas. Theoretical accuracy of the PNN classifier is determined using two methods. In the hold-one-cut method, the network is trained on all data samples minus one and is tested on the, remaining sample. Using this technique, 79.8% of the...

An Automated Method to Estimate Tropical Cyclone Intensity Using SSM/I Imagery

Abstract An automated method to estimate tropical cyclone intensity using Special Sensor Microwave Imager (SSM/I) data is developed and tested. SSM/I images (512 km × 512 km) centered on a given tropical cyclone (TC), with a known best-track intensity, are collected for 142 different TCs (1988–98) from the North Pacific, Atlantic, and Indian Oceans. Over 100 characteristic features are computed from the 85-GHz (H-pol) imagery data and the derived rain-rate imagery data associated with each TC. Of the 1040 sample images, 942 are selected as training samples. These training samples are examined in a feature-selection algorithm to select an optimal subset of the characteristic features that could accurately estimate TC intensity on unknown samples in a K-nearest-neighbor (K-NN) algorithm. Using the 15 selected features as the representative vector and the best-track intensity as the ground truth, the 98 testing samples (taken from four TCs) are presented to the K-NN algorithm. A root-mean-square error (rmse)...

Improvement to a Neural Network Cloud Classifier

Abstract Examination of various feature selection algorithms has led to an improvement in the performance of a probabilistic neural network (PNN) cloud classifier. Thee algorithms reduce the number of network inputs by eliminating redundant and/or irrelevant features (spectral, textural, and physical measurements). One such algorithm, selecting 11 of the 204 total features, provides a 7% increase in PNN overall accuracy compared to an earlier version using 15 features. This algorithm employs the same search procedure as before, but a different evaluation function than used previously, which provides a similar bias to that of the PNN classifier. Noticeable accuracy improvements were also evident in individual cloud-pipe classes.

The Identification and Verification of Hazardous Convective Cells Over Oceans Using Visible and Infrared Satellite Observations

Abstract Three algorithms based on geostationary visible and infrared (IR) observations are used to identify convective cells that do (or may) present a hazard to aviation over the oceans. The performance of these algorithms in detecting potentially hazardous cells is determined through verification with Tropical Rainfall Measuring Mission (TRMM) satellite observations of lightning and radar reflectivity, which provide internal information about the convective cells. The probability of detection of hazardous cells using the satellite algorithms can exceed 90% when lightning is used as a criterion for hazard, but the false-alarm ratio with all three algorithms is consistently large (∼40%), thereby exaggerating the presence of hazardous conditions. This shortcoming results in part from the algorithms’ dependence upon visible and IR observations, and can be traced to the widespread prevalence of deep cumulonimbi with weak updrafts but without lightning over tropical oceans, whose origin is attributed to sign...

Comparison of GOES Cloud Classification Algorithms Employing Explicit and Implicit Physics

Abstract Cloud-type classification based on multispectral satellite imagery data has been widely researched and demonstrated to be useful for distinguishing a variety of classes using a wide range of methods. The research described here is a comparison of the classifier output from two very different algorithms applied to Geostationary Operational Environmental Satellite (GOES) data over the course of one year. The first algorithm employs spectral channel thresholding and additional physically based tests. The second algorithm was developed through a supervised learning method with characteristic features of expertly labeled image samples used as training data for a 1-nearest-neighbor classification. The latter’s ability to identify classes is also based in physics, but those relationships are embedded implicitly within the algorithm. A pixel-to-pixel comparison analysis was done for hourly daytime scenes within a region in the northeastern Pacific Ocean. Considerable agreement was found in this analysis,...

Estimating Three-Dimensional Cloud Structure via Statistically Blended Satellite Observations

AbstractThe launch of the NASA CloudSat in April 2006 enabled the first satellite-based global observation of vertically resolved cloud information. However, CloudSat’s nonscanning W-band (94 GHz) Cloud Profiling Radar (CPR) provides only a nadir cross section, or “curtain,” of the atmosphere along the satellite ground track, precluding a full three-dimensional (3D) characterization and thus limiting its utility for certain model verification and cloud-process studies. This paper details an algorithm for extending a limited set of vertically resolved cloud observations to form regional 3D cloud structure. Predicated on the assumption that clouds of the same type (e.g., cirrus, cumulus, and stratocumulus) often share geometric and microphysical properties as well, the algorithm identifies cloud-type-dependent correlations and uses them to estimate cloud-base height and liquid/ice water content vertical structure. These estimates, when combined with conventional retrievals of cloud-top height, result in a 3...

Optimization of an Instance-Based GOES Cloud Classification Algorithm

Abstract An instance-based nearest-neighbor algorithm was developed for a Geostationary Operational Environmental Satellite (GOES) cloud classifier. Expert-labeled samples serve as the training sets for the various GOES image classification scenes. The initial implementation of the classifier using the complete set of available training samples has proven to be an inefficient method for real-time image classifications, requiring long computational run times and significant computer resources. A variety of training-set reduction methods were examined to find smaller training sets that provide quicker classifier run times with minimal reduction in classifier testing set accuracy. General differences within real-time image classifications as a result of using the various reduction methods were also analyzed. The fast condensed nearest-neighbor (FCNN) method reduced the size of the individual training sets by 68.3% (fourfold cross-validation testing average) while the average overall accuracy of the testing s...

Automated Lightning Flash Detection in Nighttime Visible Satellite Data

AbstractThe Defense Meteorological Satellite Program (DMSP) Operational Linescan System (OLS) nighttime visible channel was designed to detect earth–atmosphere features under conditions of low illumination (e.g., near the solar terminator or via moonlight reflection). However, this sensor also detects visible light emissions from various terrestrial sources (both natural and anthropogenic), including lightning-illuminated thunderstorm tops. This research presents an automated technique for objectively identifying and enhancing the bright steaks associated with lightning flashes, even in the presence of lunar illumination, derived from OLS imagery. A line-directional filter is applied to the data in order to identify lightning strike features and an associated false color imagery product enhances this information while minimizing false alarms. Comparisons of this satellite product to U.S. National Lightning Detection Network (NLDN) data in one case as well as to a lightning mapping array (LMA) in another c...

Next-Generation Satellite Meteorology Technology Unveiled

AFFILIATIONS: kuciauskas, solBriG, lee, hawkins, surratt, richarDson, anD Bankert—Naval Research Laboratory, Marine Meteorology Division, Monterey, California; miller—Cooperative Institute for Research of the Atmosphere, Colorado State University, Fort Collins, Colorado; kent—Science Applications International Corporation, Monterey, California CORRESPONDING AUTHOR: Arunas Kuciauskas, Naval Research Laboratory, Marine Meteorology Division, Monterey, CA 93943 E-mail: kuciauskas@nrlmry.navy.mil