FX-Net and Education

Papers

Presentations

FSL in Review

User's Manual
(external link)

DELIVERY OF METEOROLOGICAL PRODUCTS TO AN INTERNET CLIENT WORKSTATION

Ning Wang and Sean Madine^*
NOAA Forecast Systems Laboratory
Boulder, Colorado

1. INTRODUCTION

FX-Net is an Internet-based meteorological workstation that is currently under development at the Forecast Systems Laboratory. The goal is to use the existing WFO-Advanced workstation (MacDonald and Wakefield, 1996) to create sophisticated products that can then be sent over the Internet to a client workstation. The client application provides an AWIPS-like user interface along with the processing that is necessary to interact with the products. The system is intended to work for users that have only modest communications and computing capability.

The main challenge of the project is to effectively fulfill client requests as quickly as possible. This paper describes the development of FX-Net to accommodate the transfer of meteorological products over the Internet. The appropriate representation of a given product is one of the most critical aspects of FX-Net. In order to transfer a product over the Internet, the size of the product is obviously an important factor. Products can be categorized into four groups: satellite imagery, model imagery, model graphics, and radar imagery. Of these, the satellite imagery is the most difficult to handle because of its large size. Model imagery is also difficult not only because of its size, but also because of the very large number of products that are available. Both of these types of products are compressed through the use of a wavelet transform. For the purposes of FX-Net, it was determined that a small amount of loss of fidelity would be tolerable in exchange for a high compression ratio. Since this is one of the most notable aspects of FX-Net, our implementation of this compression technique is discussed in detail. Model graphics are represented in a standard vector graphics format and radar imagery is encoded in a standard lossless image compression format. The processing that is involved for a given product transmission representation also contributes to the amount of time that is necessary to fulfill a request. Also discussed are the processing strategies, for both the server and the client, which contribute to the optimization of delivery time.

2. PRODUCT REPRESENTATION

1. Representation of Satellite and Model Imagery

Satellite and model images are grouped together because their size demands many common requirements related to compression. After decompression, these types of images continue to show meteorological detail even with the small loss of fidelity. The processing time associated with decompression must not offset the time that is saved during transmission. Further, for processing reasons, it is required that the decompression of a particular image frame is independent of adjacent image frames. In the context of the above requirements, a number of compression schemes were studied. The need for high compression ratios along with the tolerance for some loss of fidelity eliminated any lossless schemes.

The Joint Photographic Experts Group (JPEG) format was tested, and it achieves good compression, but the fidelity of the decoded image is far from satisfactory. JPEG-compressed images usually exhibit some blocky effects at a compression ratio as low as 10:1.

The Moving Pictures Experts Group (MPEG) format was also investigated. An advantage of this scheme is that MPEG hardware is readily available for use by the client workstations. MPEG also achieves a high compression ratio (about 35:1) by exploiting the time continuity of the series of frames. There are, however, two major problems with the use of MPEG. The fact that MPEG takes advantage of time continuity requires that a decoder wait for all frames to arrive before initiating the decompression processing. Additionally, the quality of the resultant image loop is not satisfactory in terms of fidelity. As in JPEG, the blocky effect that is introduced degrades the image far too much for use by FX-Net.

We also tested the fractal (or attractor) coding method, which is a variant of the image vector quantization compression method. Advantages of this approach include a high compression ratio and fast decompression time. The quality, however, of the decompressed image is very low in the context of meteorological use. In fact, because this is a method that does not rely on basis functions as a means of representation, it is difficult to achieve a desired fidelity even at the cost of a lower compression ratio.

Ultimately the wavelet transform was chosen as the approach to image compression for FX-Net. The wavelet transform was introduced in the early 1990s and has remained a cutting edge technology in image compression research (Akansu 1992, Prasad 1997). Like the Fourier transform, the wavelet transform relies on a particular set of basis functions. However, the set of basis functions that the wavelet transform uses is localized in both space and frequency, whereas a Fourier transform only contains frequency information. It is the ability of the wavelet transform to contain some spatial information, in addition to the frequency information, that allows it to achieve excellent compression of meteorological images. Besides providing the necessary compression of images, the wavelet transform also fulfills the other representation requirements. The loss of fidelity is acceptable, the decompression processing is reasonable, and the image frames are individually compressed (Three dimensional wavelet compression which exploits the time continuity of the series of frames might yield higher compression ratio but it also introduces some inter-frame dependencies. We are currently doing research work on that.) The batch generation of the compressed products and real-time decompression of images by the client is discussed in Processing Strategies section of this paper.

The three types of satellite imagery that are available through FX-Net are infrared, visible, and water vapor. Each of these has different characteristics in terms of the wavelet transform. One advantage of the wavelet transform is that it is possible to "tune" the compression with a judicious choice of basis functions. The table below compares compression ratios and PSNRs (Peak Signal to Noise Ratio) for Infrared satellite image when compressed with different basis functions. The compression scheme follows the standard transform-quantization-entropy coding procedure.

Table 1.

Antonini (Antonini, 1992)	10:1, 41.01 20:1, 37.64 30:1, 36.00
Daubechies 6	10:1, 40.14 20:1, 37.04 30:1, 35.37
Quadratic Spline (Cohen, 1992)	10:1, 39.62 20:1, 35.41 30:1, 33.11

compression ratio and PSNR for IR satellite image using wavelet compression.

The model image, which also lends itself well to wavelet compression, contains much less information than a satellite image. In comparison, the compression ratio for the model images is very high (about 100:1). The processing complications associated with the model images are discussed in the Processing Strategies section.

2.2 Representation of Model Graphics

Model graphics are represented in a vector graphics format called Dare Graphics Metafile (DGM). Most important, this format offers a compact representation of the model graphics (e.g. typical CONUS 500-mb Height Contour DGM file is on the order of 15 kilobytes in size). The FX-Net system also takes advantage of the ability of DGM to encode progressive disclosure information about the graphics. This enhances the client?s ability to perform efficient zoom display operations.

2.3 Representation of Radar imagery

Radar imagery must be represented differently than the satellite and model imagery because there is no tolerance for loss in the radar signal. Use of wavelet compression, for example, could introduce an artificial feature or even mask a dangerous storm feature that does in fact exist. Since radar imagery also contains much less information (in terms of signal) than satellite imagery and there is no tolerance for loss, FX-Net compresses radar imagery in a lossless manner.

This was accomplished using a standard format, Graphical Interchange Format (GIF), which was tested and it works well for radar images. The compression ratio varies significantly depending on the amount of information in the radar image. The size of even the largest resultant radar files is manageable using the FX-Net scheme.

3. PROCESSING STRATEGIES

It is important to address the processing associated with the various formats that have been chosen for the FX-Net products. The vector graphics format is inexpensive for both encoding on the server and decoding (displaying) on the client. Similarly, GIF encoding and decoding do not contribute significantly to the total time associated with product delivery. The wavelet transform, however, is computationally intensive, particularly on the server side.

3.1 Wavelet compression on the server

Besides the actual wavelet transform, the compression routine on the server involves a dynamic search for the optimal set of basis function coefficients that are used to represent the image. This dynamic search greatly enhances the ability to compress the image at the cost of increased processing time, especially for satellite imagery, where the compression ratio is critical. The cost in time can be on the order of 20 seconds per image frame. For the set of available satellite imagery, frames are compressed immediately upon arrival from the data ingest system. Any subsequent requests for satellite imagery by a client can be fulfilled without performing the compression processing.

The production of model imagery is significantly different. Because there is a huge matrix of possible products, model image products cannot be generated before an actual request. However, the high compressibility of the model images allows the compression routine to bypass the dynamic search for coefficients. A best guess set of coefficients is used, and the processing time in this case is then reasonable for the necessary on-demand compression.

2. Wavelet decompression on the client

The decompression of the wavelet-compressed files on the client is also computationally expensive. The processing takes about 2-3 seconds per image frame when run on a 400MHz PC. The client takes advantage of the multithreading offered by the Java programming language. By executing decompression and communication threads concurrently, the capabilities of the client hardware are optimally utilized. Each individual image frame is displayed for the user as the decompression completes, thus minimizing the perceived wait for the product arrival.

 

4. CONCLUDING REMARKS

As Internet communications get faster and available client hardware provides better processing capability, the concept of a network meteorological workstation becomes more viable. Meanwhile, many factors go into the design of such a workstation. The effective delivery of the products from the server to the client is a very important consideration in the FX-Net project. This includes not only the representation of the various products, but the processing associated with those representations.

 

5. ACKNOWLEDGEMENTS

The authors would like to thank Renate Brummer for reviewing this paper. Thanks also to Nita Fullerton for technical editing.

 

6. REFERENCES

Akansu, A.N., and R.A. Haddad, Multiresolution Signal Decomposition, Academic Press, 1992

Antonini, M. , Barland M. , Mathieu P., and Daubechies, I. Image coding using wavelet transform, IEEE Trans. Image Processing, Vol. 1, pp. 205-220, 1992.

Cohen, A., Daubechies, I. and Feauveau, J. Bi-orthogonal Bases of Compactly Supported Wavelets, Communications on Pure and Applied Mathematics, Vol. 45, pp. 485-560, 1992.

MacDonald, A.E., and J.S. Wakefield, WFO-Advanced: An AWIPS-like Prototype Forecaster Workstation. Twelfth International Conference on Interactive

Information and Processing Systems for Meteorology, Oceanography and Hydrology, Atlanta, GA., Amer. Meteor. Soc., 190-193, 1996

Prasad, L., and Iyengar S.S., 1997. Wavelet Analysis with Applications to Image Processing, CRC Press.

* Jointly affiliated with the Cooperative Institute for Research in the Atmosphere, Colorado State University, Fort Collins, CO