4.1 INTRODUCTION
Since 1974, environmental data have been collected using geostationary satellites. Over the years, a large user community with diverse interests has developed a variety of applications and conducted research using these data. These differing interests impose a variety of demands on the data providers. These demands are not restricted only to timely distribution of the data, but extend also to ensuring the continuity of the data. NOAA is responsible for stewardship of this valuable resource, and is committed to preserving both the quality and continuity of environmental data.
As NOAA improves environmental sensors and their supporting systems, continuity of services becomes increasingly important in protecting the investments of the diverse user community. Meeting this objective requires an evolutionary process for incorporation of improvements using modular services. With such a process, improvements do not require adjustments on the user's part to continue existing operations, but only to take full advantage of those improvements.
With the need for continuity of service in mind, this section addresses ground processing considerations for the proposed FPA imager in context of a NOAA ground processing system. This reference ground processing system is described in section 4.2. To support the proposed imager, sections 4.3 through 4.6 describe modifications of the functions of data reception, calibration and normalization, earth location and gridding, preparation for distribution, and sector distribution. Section 4.7 presents several scenarios of operation exploiting rapid image formation by the proposed imager and the earth location capabilities of the proposed ground processing system. Section 4.8 presents a summary and conclusions.
4.2 REFERENCE GROUND PROCESSING SYSTEM
Presently, NOAA operates geostationary satellites via a distributed network of ground systems at the Command and Data Acquisition Station (CDA) at Wallops Island, Virginia and the Satellite Operations and Control Center (SOCC) in Suitland, Maryland. At these facilities, data are received, processed, and distributed to government users in the various forms required. Outboard systems distribute data and products to external users such as the military, private industry, and educational institutions. For real-time data, these systems include GOES-TAP; Satellite Field Service Stations (SFSSs); and, in the future, NOAAPORT. The Data and Product Archives support both government and external users. This operation will continue into the future, with upgrades to support the enhanced capabilities of GOES I-M spacecraft.
Allowing time for development and production, the proposed imager would be ready for deployment well after GOES I-M series spacecraft and their ground processing system are due to begin operation; thus, GOES I-M ground processing is used as the reference system for the following discussion. GOES I-M ground processing will service all of the data streams communicated to and from the spacecraft. These data streams include those generated by meteorological sensors (imager and sounder), space environment and monitoring sensors, search and rescue relays, and Data Collection System (DCS) observations. Also, the ground processing systems receive and transmit telemetry to monitor and supervise spacecraft operation. The discussion that follows considers the imager data stream and addresses other data streams only to the extent that they support processing the imager data.
4.2.1 GOES I-M Ground Processing System
GOES I-M ground processing is accomplished by subsystems grouped as follows:
* The GOES I-M Spacecraft Support Subsystem, including the Operations Ground Equipment (OGE), the Telemetry and Control (T&C) Subsystem, the RF transmit/receive subsystems, and ground communications
* The Product Generation and Distribution (PG&D) Subsystem, including the Central Meteorological Satellite Computer System (CEMSCS), the Multi-Discipline Data Analysis System (MIDAS), the GOES Sectorizer System, and the GOES Ingest and NOAAPORT Interface (GINI).
GOES I-M Spacecraft Support Subsystems reside at both the CDA and at the SOCC. PG&D Subsystems are at Federal Building #4 in Suitland, Maryland, and the NOAA Science Center in Camp Springs, Maryland. System components performing the real-time operations of processing and retransmitting raw instrument data are at the CDA; system components dealing with refined data and less time-critical operation are in Maryland.
4.2.2 Spacecraft Support Subsystems
Figure 4-1, adapted from [GOES91], shows the data flow among the geographically distributed GOES I-M Spacecraft Support Subsystems. Wideband raw instrument data are received by RF equipment at the CDA and routed by the OGE Data Acquisition and Patching Subsystem (ODAPS) to a Sensor Processing System (SPS), where the data are demodulated and processed.
Each SPS (one per spacecraft) ingests imager data and instrument telemetry, and then processes the imager data based on the telemetry and historical data. The SPS calibrates and normalizes imager data, computes and stores calibration statistics, applies reference grids to the image, locates the data samples with respect to earth coordinates, derives data to support orbit and attitude determination, applies time tags, and formats the data for GOES VARiable (GVAR) transmission.
Developed for the new capabilities of GOES I-M, the GVAR format was designed to maintain as much commonality as possible with the formats acceptable to existing Direct Readout earth stations. The largest data fields of this format are reserved for the meteorological data measured by the Imager and Sounder instruments. Also included are parameters associated with the measuring instruments, and provision for auxiliary products. After SPS processing, the GVAR-formatted data are transmitted from the CDA to the spacecraft that relays these data to Direct Readout earth stations, among them the SOCC.
Following the spacecraft relay, the GVAR-formatted data are received by RF equipment at the SOCC, resynchronized, and converted to an intermediate frequency (IF) signal. This signal is relayed to PG&D subsystems at Federal Building #4 and the NOAA Science Center, and to the other Spacecraft Support Subsystems in the SOCC, including the Product Monitors (PMs) and the Orbit and Attitude Tracking System (OATS).
Under normal operating conditions, the SOCC PMs identify and register landmarks in sectors of the imager data. The PMs support the OATS by providing, on request, the earth location coordinates of the landmark data, and Image Motion Compensation (IMC) and servo error data extracted from the GVAR-formatted data. PMs at both the CDA and the SOCC monitor the quality of GVAR-formatted imager and sounder data; display nonimage data contained in the GVAR-formatted data stream; and monitor the performance of calibration and normalization functions, accumulating statistics to support tuning of these functions. Under backup operating conditions, the CDA PM assumes the role of the SOCC PMs to support the OATS.
The principal function of the OATS is to produce the IMC coefficients used by spacecraft systems to meet the sensing systems' image navigation and registration (INR) requirements (as discussed in section 4.5). This process begins when the OATS ingests data forwarded by the GOES I-M Telemetry and Command System (GIMTACS), extracted from the Multi-purpose Data Link (MDL), and requested from the PM. From the star, range, and landmark observations of these data, the OATS determines the orbit and imager and sounder attitudes. On a daily schedule, the OATS produces IMC coefficients by predicting orbit and attitude perturbations following housekeeping operations or repositioning maneuvers. In addition, the OATS produces a schedule of commands to acquire needed data for the next day's processing and to satisfy other requests, including verification and tuning of Mirror Motion Compensation (MMC).
4.2.3 PG&D Subsystems
Each of the PG&D subsystems, whatever its location, shares a common architecture. This architecture includes VAS Interface Electronics (VIE) adapted from earlier GOES systems, a GVAR Ingestor, a GOES Real-Time Database (GRT), and applications. The VIE receives GVAR data and forwards them to a GVAR Ingestor that places the data in the GRT for access by PG&D applications. Located at Federal Building #4, the CEMSCS generates most of the automated products. These are image subsets for Weather Facsimile (WEFAX) distribution by the spacecraft, estimation of low-level winds from cloud motion, and data and product archives.
The remaining PG&D subsystems--MIDAS, the GOES Sectorizer System, and GINI--are at the NOAA Science Center. Most interactive products are generated by meteorologists using the display, processing, and database services of MIDAS. MIDAS maintains a comprehensive database of meteorological data from polar orbiting satellites, National Meteorological Center (NMC) guidance products, conventional observations, and weather radars. Using MIDAS, the interactive products generated are global winds products, cloud top temperatures, atmospheric moisture distributions, rainfall estimates, sounding products (temperature and moisture profiles sampled at different altitudes), and derived imagery products (combined data from several imaging channels). The GOES Sectorizer System selects subsets (known as sectors) from the full earth disk GVAR data, enhances them, and distributes them in facsimile format (GOESFAX) to SFSSs. From the SFSSs these sectors are distributed to NWS Forecast Offices and other government and educational users. Currently in development, GINI remaps imagery into the three map projections required by AWIPS, and forwards these images to the AWIPS contractor for distribution via NOAAPORT.
4.3 DATA RECEPTION
For the proposed imager, ground processing begins with reception of a baseband digital data stream. Forming this digital data stream requires RF reception, detection, and bit synchronization equivalent to that performed by the ODAPS for NOAA-operated spacecraft.
Regardless of the origin of the baseband digital data stream, the next step in its processing is frame synchronization. This is the process of isolating and extracting data blocks by detecting synchronization patterns and headers in the data stream that represent the start of a data block and by accumulating the data a bit at a time. This processing is similar to that performed by the SPS, as is the remainder of the data reception processing required by the proposed imager.
Following frame synchronization, the SPS extracts nonimage data from the data stream and decommutates, or separates, the incoming data stream into separate data streams, one for each of the sensor channels. Nonimage data include instrument status and calibration measurements taken on board the spacecraft; these data are required for statistical analysis and subsequent ground processing steps.
Because of differences in scanning technique, the proposed imager data stream requires less data processing for decommutation than does the GOES I-M imager. While scanning latitudinally, the GOES I-M imager simultaneously acquires radiometric data for five channels, one visible and four IR. These data are accumulated in 480-bit data blocks, each corresponding to 64 microradians (urad) of latitudinal scan, and transmitted at a rate of 5460 blocks per second. By sorting these composite data based on their offset locations in the data block, decommutation performs time division demultiplexing to produce a synchronized set of individual sensor data streams.
Like the GOES I-M imager, the proposed imager acquires its five channels of radiometric data simultaneously. However, for a given scan position, the proposed imager arrays capture many more samples for each channel (as many as half a million for the visible channel and 32,000 for the IR channels). This allows forming transmission blocks that exclusively contain data for separate channels, with channel identification included in the block header. As with GOES I-M, time division multiplexing is performed, but requires less frequent computation to produce the individual sensor data streams.
Compared with GOES I-M, the proposed imager requires additional processing during data reception. As discussed in section 2.5, for a 3 minute transmission time from the proposed imager, the raw data rate is 50 Mbps and would require data compression to reduce the transmission channel rate if that was deemed to be desirable. As a consequence, decompression would have to be applied to transmitted data during ground processing to recover the original data.
A low-loss data compression technique that preserves the integrity of the original data while minimizing the encoding and transmission of redundant data would be required. Traditional compression techniques minimize redundancy by examining linear strings of data or scan lines and then replacing substrings that occur frequently with compact codes. Decompression is a process of copying data and replacing the codes with their corresponding substrings.
The array detectors of the proposed imager form two-dimensional images, a format that offers more opportunities for detecting and minimizing redundancy than a linear string. For example, the standard compression technique used in facsimile (FAX) coding is a two-dimensional extension of run length coding (RLC), a linear technique. FAX is a more efficient coding technique than RLC, when applied to a variety of images.
Selecting a suitable compression technique requires additional analysis and investigation of the redundancy in both GOES images and in high resolution images as produced by the proposed imager. The acceptable amount of information loss must be identified and weighed against the transmission rate reduction benefits of the various techniques.
Lossless, or error-free compression techniques can produce compression ratios in a range from 2:1 to 10:1, although generally they yield a maximum compression ratio of 3:1. Lossy techniques introduce distortion, but can produce compression ratios as high as 30:1 with some techniques offering extremely high fidelity at compression ratios ranging from 10:1 to 20:1. Also, the selected technique must be robust, providing the required compression ratio under the variety of diurnal and seasonal variations that are anticipated in the imager scenes. The required analysis is beyond the scope of the present investigation.
4.4 CALIBRATION AND NORMALIZATION
Following reception, these data are available: individual data sets for each of the imager channels, and non-image data including control and timing information from the imager and calibration measurements taken onboard the spacecraft. Calibration and Normalization, the next stage in processing, will use the non-image data to produce radiance estimates from the raw digital counts in the imager channel data sets. Calibration and normalization required for the proposed imager data is similar to that performed by the SPS for GOES I-M imager data, except for visible channel processing, where calibration is required instead of normalization.
Referring to SPS processing, radiance estimates for IR channels are produced from GOES I-M imager data using hardware look-up tables. A sensor's raw digital count is used as a table index to locate and extract a radiance estimate from the look-up table. For GOES I-M, different tables are provided for each of the imager sensors. These tables are generated by a quadratic polynomial defined with calibration coefficients.
The polynomial's basic coefficients, adjusting the function's gain and bias, are derived from space look and blackbody calibration measurements taken on board the spacecraft. Space look is accomplished by exposing the IR detectors to a view of space beyond the edge of the earth, providing a cold reference. Blackbody measurement is accomplished by exposing the IR detectors to a view of an internal blackbody source, providing a 350 degree K reference. Second-order coefficients for the polynomial are developed from the factory-measured response characteristics of the IR sensors. Optionally, additional coefficients may be included; these are based on temperatures measured on the instrument's baseplate.
The process outlined here for calibration of GOES I-M IR imager data is essentially that required for calibration of the proposed imager's IR channel data sets. However, since a calibration table containing 2^(11) entries would be required for each pixel of the sensor arrays, direct computation of the radiance estimate from the calibration polynomial is a more suitable approach than table look-up.
Again referring to SPS processing, radiance estimates for visible channels are produced from GOES I-M imager data using normalization look-up tables (NLUTs). These tables are loaded with values that are periodically adjusted to incorporate current calibration and normalization information. Table values for the eight visible sensors are generated by an analyst at the PM, a non-real-time operation. From the GVAR data stream, the analyst chooses a full-resolution sector of maximum dynamic range, i.e., one containing cloud cover, ocean, and land masses. Then one of the visible sensors known to have long-term stability is selected as a reference and compared with the remaining seven visible channels. From this comparison, table values may be adjusted, if necessary, to provide uniform radiance estimates. The objective of this process is to ensure that uniform radiance estimates are produced from the raw digital counts of all eight visible sensors when viewing a scene of uniform brightness. If this process does not correct nonuniform visible detector response, striping will be observed in the visible images.
As discussed in sections 3.3.1 and 4.5, our approach to the problem of geographically locating pixels in a proposed imager frame begins with a process of "stitching together" overlapping frames. This process requires detecting features (based on contrast gradients) that are common to the overlap regions of overlapping frames and computing shifts to maximize the coincidence of the features in the overlap region.
Overlap regions for successive frames are viewed by different elements of the sensor array: the eastern overlap region of one frame is compared with the western overlap region of its successor. Any nonuniformities in the performance of the sensor elements across the array would degrade this process; consequently, a calibration process is required for the proposed imager's visible channel data sets. We propose the same as that used for IR calibration, discussed above, based on calibration measurements taken on board the spacecraft by scanning dark (no illumination) and light (evenly illuminated) line targets. In addition to improving visible image quality, this approach eliminates the labor-intensive analytical operation of producing NLUTs using the PM. Deciding the allowable level of array sensitivity nonuniformity that will support our proposed image navigation technique will require evaluation of a variety of visible scenes. This evaluation is beyond the scope of the current investigation.
4.5 EARTH LOCATION AND GRIDDING
Longitude and latitude grid lines and geopolitical boundaries are referenced to imager data using a process known as gridding. Similarly, earth location is a process that associates longitude and latitude coordinates with identified pixels of the imager data. Both gridding and earth location information are annotations to the imager data. For GOES I-M data users, these annotations are transmitted in the GVAR data stream, gridding information in Block 0, and earth location information interleaved in the data blocks for every 196th pixel of the first visible line of every 14th scan.
As discussed in sections 3.2 and 3.3, the objective of the GOES I-M Image Navigation and Registration (INR) system is for all images to be consistently produced with the same fixed earth projection. Earth location and gridding for images produced with the GOES I-M INR is a straightforward transformation of the longitude and latitude grids and geopolitical boundaries, taking into account the spacecraft's orbit and attitude. For GOES I-M, this is performed by the SPS using either actual orbit and attitude data from the spacecraft, or reference orbit and attitude data provided by the OATS. As discussed in section 3, there are significant differences between the scanning technique of the GOES I-M flying spot scanner and the techniques described for the proposed imager. These differences require a different approach to INR for the proposed imager than for GOES I-M INR and, consequently, a different approach to earth location and gridding.
For the proposed imager, we consider ground data processing that provides the INR function of locating imager array pixels with respect to earth coordinates as part of the process that produces gridding and earth location annotations. Our approach produces only these annotations and does not alter the pixels in imager frames. This approach is only outlined herein, since the details of its implementation depend on the information content of imager frames under a variety of diurnal and seasonal variations. As previously discussed in sections 4.3 and 4.4, the required analysis is beyond the scope of the current investigation. Also, we limit discussion to the step-stare imaging method, since time did not allow developing an approach for the time delay and integration (TDI) imaging method.
Our approach takes advantage of the proposed imager's principal benefit: a well-correlated sequence of overlapping "snapshot" images, each containing a consistently sampled set of pixels. Working with the collection of snapshot images provides a significant benefit to be realized during implementation of our approach. Many of the processing steps proposed operate within single image frames or localized regions of image frames; they do not depend on data outside their current frame or region. With no dependencies on other data, these operations can be performed in parallel; thus an implementation of our approach can provide high performance through parallel processing. The remainder of this section describes the required data processing, generally outlined as follows:
* For a given channel, each frame of the imager data set is related to its neighbors; a mosaic of these frames is created based on redundant image information in overlap regions common to neighboring frames.
* Landmarks are located in the mosaic image. The landmark locations are interconnected with a network of lines; each landmark location is connected to at least two other landmark locations, each in a different frame.
* Distances between landmarks in the mosaic image are iteratively revised by adjusting the landmark locations within the mosaic until the distances agree with their known values, which are a function of actual spacecraft orbit, attitude, and time of day.
* Landmark location adjustment values are apportioned among the frames of the data set.
* Gridding and earth location annotations are developed for each frame based on its location in the frame sequence (e.g., scan mirror coordinates, frame image exposure event time) and the landmark location adjustment values.
With respect to a complete scene, each frame captured by the proposed imager includes redundant information. As discussed in section 3.3.1, this takes the form of an "overlap" or margin of additional rows and columns of pixels surrounding the targeted frame area. We propose using the redundant image information in these overlap regions for two purposes: first, to control our construction of a mosaic image, representing the entire scene viewed by the imager; and second, to support computation of the relative shifts and rotation of the imager frames that occurred while the imager scan was advanced from frame to frame.
Our mosaic is constructed with a coordinate system of subpixels. Subpixel dimensions are selected to limit any error introduced by the contrast gradient operation (discussed below) to less than one-fourth of a frame image pixel. This requires that the linear dimensions of a subpixel each be selected to be one-eighth the size of the linear dimensions of pixels of the imager frames. The validity of this selection of subpixel dimensions requires evaluation of the information content of imager frames with a variety of diurnal and seasonal variations.
Subpixels are used in the mosaic for two reasons: first, to allow determination of small relative rotations between frames, and second, to enhance the performance of the contrast gradient operation. As discussed in section 3.3.1, for the visible channel, a full earth image is captured with a scan of 46 latitudinal frames and 23 longitudinal frames using a frame of 512 x 1024 pixels. In our mosaic, this region is represented using 8 x 46 x 512 or 188,416 latitudinal subpixels and 8 x 23 x 1024 or 188,416 longitudinal subpixels.
We begin the process by tentatively placing each frame in the mosaic at a nominal position based on its location in the frame sequence. We proceed, examining each frame in the order of the scan sequence, with the frame being shifted and rotated as needed to bring features in its portions of the overlap regions into coincidence with the corresponding features in the overlap regions of its neighbors. As we proceed through the scan sequence, we record the shifts and rotations required for the addition of each frame.
Image processing offers a variety of approaches to detecting features in the overlap region and determining coincidence of features. The processes we require are similar to, but simpler than, those for matching cloud images when tracking their motion to produce estimates of winds aloft. Cloud motion winds are computed by tracking matched images selected from successive GOES images. The matching process is difficult since, during the 30 minute interval between GOES images, scene illumination variations and image deformations are common. For the proposed imager, the corresponding interval is generally not less than 200 milliseconds and never more than 14.4 seconds, so these variations and distortions are less likely.
Selection of optimal techniques requires an analysis of image data that is beyond the scope of the current investigation. For the sake of discussion, we offer a sample technique for detecting features and establishing coincidence in the overlap region at the east-west boundaries of two neighboring frames. Each frame's overlap region (one from the eastern frame and one from the western frame) is supersampled into mosaic subpixels, i.e., each subpixel is assigned the value of the frame image data at its location. A contrast gradient is derived for each supersampled sub-pixel. This is the magnitude of the vector sum of central differences in radiance taken north-south and east-west. This technique operates on the image's contrast gradients instead of image pixel values to minimize bias along preferred north-south or east-west axes, and to eliminate sensitivities to contrast and mean gray levels.
To construct a feature, we remove low-contrast gradient values, contracting the image, until a skeleton remains. A skeleton feature from one frame's overlap region is used as a mask and compared with the corresponding skeleton feature from the other frame's overlap region. The first frame's feature is shifted and rotated until the differences are minimized. These shift and rotation operations are recorded for each frame. Frames are added to the mosaic in scan sequence; for a given frame, the feature coincidence operation is performed only for overlap regions of the frame's predecessors. In general, the operation is performed once for each frame at the northern and eastern scan limits and twice for each of the other frames.
Matching of features in the east-west overlap regions is accomplished with a high degree of confidence, since the exposure event times of corresponding frames have an interval of less than 200 milliseconds. However, for north-south overlap regions, this interval could be as much as 14.4 seconds for north-south overlaps at the latitudinal extremes of the scan. Image motion during this interval can produce apparent shifting of a feature in one frame's overlap region with respect to the equivalent feature in a neighboring frame. This apparent shifting or "image smear" is defined and discussed in section 3.4.4. We use the landmark locations and known relationships among the landmarks both to earth locate the image and to mitigate errors due to image smear.
To restrict the domain for landmark recognition, approximate landmark locations in the mosaic are determined from the frames' scan coordinates. These locations provide initial estimates for the recognition processes. Considering the proposed imager's rapid scan rate (3 minutes for a full earth image), and its potential for frequently updated imaging (20 full earth images per hour, assuming continuous operation), automated landmark recognition is required. Details of an automated landmark recognition technique are dependent on image characteristics and require further analysis.
Using the focal plane arrays (FPAs) of the proposed imager in the step-stare mode simplifies the task of automated landmark recognition. As discussed in section 3.4.2.2, compared with single-pixel sensors or one-dimensional array sensors, the step-stare approach produces a large area of simultaneously exposed pixels. A two-dimensional frame produced by the step-stare approach is a "snapshot," preserving the topological relationships among all image pixels in the exposed scene (the frame image is subject to geometric distortions, as discussed in section 3.4, besides those resulting from orbit and attitude variations). This frame is large enough to completely contain a landmark image.
With landmark images completely contained in frames, there is no need to assemble target images for landmark recognition as is required for single-pixel sensors or one-dimensional array sensors; thus a common source of errors is eliminated. We suggest automated landmark recognition is practical, given the proposed imager's high resolution and "snapshot" method of image formation. Also, we suggest that the enhanced resolution of the proposed imager's IR channels supports practical automated landmark recognition using IR. As with visible automated landmark recognition techniques, the details of an IR technique are dependent on image characteristics.
With landmarks located in the mosaic image, a network of lines interconnecting the landmark locations is formed so that each landmark location is connected to at least two other landmark locations. In the mosaic, these lines represent projections of great-circle arcs interconnecting the image landmark locations. Our objective is to adjust the landmark locations in the image so that the lengths of their interconnecting lines agree with known values. These known values are derived from the actual great circle arcs interconnecting the real landmarks. Landmark projections on a reference plane oriented by the spacecraft's orbit and attitude coordinates provide the needed values.
Adjusting the image landmark locations to bring their interconnecting lengths into agreement with known values is an application of well-known relaxation techniques. The required landmark location adjustments are recorded for each landmark-bearing frame as vectors encoding the required north-south and east-west shift operations. A significant benefit of the image landmark location adjustment process is that the earth location of an image landmark not only is a function of the landmark recognition process, but also is related to the known distances between the image landmark and its neighbors. Adjusting the interconnected network of image landmarks by matching a known reference takes advantage of the relationships among landmarks to minimize the impact of inaccuracies in the recognition of single landmarks. Using the additional information provided by the known lengths of interconnecting lines, we expect that the pixels of image landmarks can be reliably located to within a pixel of their true earth locations. This provides resolution of 0.5 km for visible landmark recognition and 4.0 km for IR landmark recognition using channels 4 and 5. Precise determination of the resolution limits of this technique requires evaluating the performance of the recognition technique when applied to imager frames with a variety of diurnal and seasonal variations.
Image landmark location adjustments are apportioned among non-landmark-bearing frames using weighted averages. For a non-landmark-bearing frame, apportioned landmark adjustments are a weighted vector average of landmark adjustments from landmark-bearing frames using weights that are an inverse function of the distances between frames. In section 3.4.4, image smear limits are described by a function of scan sequence width, a spatial measure of scan sequence timing. This is a sound basis for an alternative for apportioning landmark adjustments.
Since the latitudes and longitudes of the landmarks are known, the earth locations of the adjusted image landmark locations in the mosaic are similarly known. Consequently, the earth locations of all frame pixels in the mosaic can be found. For any frame pixel, this requires computation based on the enclosing frame's scan mirror coordinates, the shifts and rotations noted when the frame was incorporated in the mosaic, the recorded or apportioned landmark adjustments, and the orbit and attitude of the spacecraft. Image field rotation, as discussed in section 3.4.5, is a function of the scan mirror coordinates, and is readily corrected as part of this computation. Similarly, channel-to-channel coregistration is accommodated by this computation, based on factory measures of optical system performance.
With earth locations established, geophysical boundaries and gridding lines can be associated with frame pixels. For each frame, annotations are produced conveying the gridding and earth location information. Gridding information is conveyed using a bitmap with an entry set for each frame pixel covered by a geophysical boundary or gridding line. Earth location information is conveyed by providing precise earth coordinates for the frame's most northwestern and most southeastern pixels, and parameters for a function that derives earth location for the remaining frame pixels. These parameters are computed from those listed above as required to determine earth location for any frame pixel. Also included in the frame's annotation are nonimage data, including frame exposure event time, mirror scan coordinates, and scan sequence coordinates.
4.6 PREPARATION FOR DISTRIBUTION AND SECTOR DISTRIBUTION
The NWS Modernization Program has set a goal of improving the forecasting and warning of locally severe weather events. This goal will be achieved by developing and fielding the AWIPS. AWIPS systems at Weather Forecast Offices (WFOs) will produce timely, high-resolution forecasts and warnings for their areas of responsibility from a database containing a detailed description of the state of the atmosphere in their local areas. This description of the state of the atmosphere will be formed as a gridded database integrating numerical guidance products, observations, weather radar products, and satellite imagery. Key to this operation is high resolution of mesoscale events, i.e., those with a time scale of less than 1 hour and a distance scale of less than 200 km.
Broadcast transmission of the annotated frames of proposed imager data is a suitable source of satellite imagery data for AWIPS, or other systems that select satellite imagery data and resample using specialized coordinate systems. Each frame's earth location annotations describe a bounding box for the frame in earth coordinates, simplifying selection of frames that cover a region of interest. Also, these annotations support rapid and precise calculation of earth locations for the frame pixels. Resampling to local or specialized coordinate systems can be rapidly accomplished with a nearest-neighbor technique. This technique sets a pixel value in a resampled image from the value of the frame pixel with the closest earth coordinates to those of the resampled pixel. In overlap regions, where frame pixels are available from two frames, the frame pixel with the most recent exposure time is used.
Broadcast of the annotated frames, with resampling performed locally for specific uses, will provide improvements in image quality. This approach eliminates the cascade of errors that result from resampling generic products that are already resampled during their production. Detection and analysis of mesoscale events are enhanced by improved image quality and the rapid update rate of the proposed imager, which can produce a full earth image every 3 minutes.
With broadcast of annotated imager frames on this schedule, a WFO could revise its description of the state of the atmosphere for some region of interest every three minutes, a suitable rate for detecting and tracking mesoscale events. Also, an operational benefit derives from the broadcast; from the same broadcast transmission, frames can be selected to form either a full earth image or a multiplicity of images for regions of interest. This means the proposed imager can be operated in a free-running mode, with no special scanning modes or schedule interruptions for special events such as increased surveillance of locally severe weather conditions.
Broadcast of annotated frames, as outlined above, is suitable for advanced systems such as AWIPS and sector-generating systems capable of resampling data, but we must address the requirements of near-term product-generation systems. As discussed in section 4.2.3, imager data are transmitted from the CDA in GVAR format, and processed by GVAR Ingestors to populate the GRT database. All product generation is performed by applications that operate either on this database or on the GVAR data stream. We propose that, over time, the GVAR Ingestors be modified to ingest the annotated frames, and that additional GRT access subroutines be developed for the proposed imager data. For continuity of service during this transition, the formatted imager frames would be resampled using the GOES Standard Projection, reformatted as GVAR, and transmitted to the PG&D systems for the normal course of product generation and distribution.
4.7 OPERATING SCENARIOS
With GOES I-M, NOAA plans to support three basic modes of operation:
* Normal mode--the routine mode of operation. Full earth or near full earth images are provided every half hour.
* Watch mode--enacted when the onset of severe weather is suspected. A sector covering the northern hemisphere and part of the southern hemisphere is provided every 15 minutes.
* Warning mode--enacted when the onset of severe weather is imminent. Regional sectors covering severe storm areas are provided; any or all sectors can be in a warning mode independently.
In addition, NOAA has international commitments for a full earth image once every 3 hours. As discussed in section 4.6 above, the proposed imager does not require special scheduling considerations to support all of these operating modes.
The enhanced resolution and rapid update rate of the proposed imager provide for early detection and accurate forecasting of mesoscale events, an objective of the NWS Modernization Program. NESDIS Office of Research and Applications Staff have researched and developed techniques supporting this endeavor. These techniques either require or benefit from data of high spatial and temporal resolution, like that of the proposed imager.
Applying these techniques to data of this quality, operational meteorologists can detect and better understand phenomena such as the following:
* Triggering mechanisms important in thunderstorm genesis
* Thunderstorm intensity (estimated through measurement of cloud top temperatures)
* Atmospheric motion (cloud drift winds, water vapor gradient changes, and thermal gradient winds)
* Atmospheric cloudiness (cloud cover emissivity, and height)
* Tropical cyclone tracks (steering currents approximated by deep layer mean wind vectors)
* Clear air turbulence
* Fog detection and dissipation at airport locations
* Potential thunderstorm development (determined from early morning cloud cover)
4.8 SUMMARY AND CONCLUSIONS
The above discussion can be summarized in the following conclusions:
* During data reception, demultiplexing of the data stream from the proposed imager requires less frequent computation than is required for GOES I-M; decompression requires additional processing beyond that required for GOES I-M.
* Automated landmark recognition, in both visible and IR channels, is feasible for the proposed imager's frames.
* The landmark location adjustment process can provide reliable earth location while accommodating image smear. As for GOES I-M, gridding is a straightforward process.
* The frames produced by the proposed ground processing contain imager pixel data that have been radiometrically calibrated, but not altered in any other way. Earth location and gridding information is provided as annotations for these frames.
* The proposed imager does not require special scheduling considerations to support all of the operating modes required by NOAA for GOES I-M imaging. From the same broadcast transmission, frames can be selected to form either a full earth image or a multiplicity of images for regions of interest. This means the proposed imager can be operated in a free-running mode, with no special scanning modes or schedule interruptions for special events.
* The enhanced resolution and rapid update rate of the proposed imager provide for early detection and accurate forecasting of mesoscale events, an objective of the NWS Modernization Program. With broadcast of annotated imager frames on a 3 minute schedule, a WFO could revise its description of the state of the atmosphere for some region of interest every 3 minutes, a suitable rate for detecting and tracking mesoscale events.
* Broadcast transmission of the annotated frames of proposed imager data is a suitable source of satellite imagery data for AWIPS, or other systems that select satellite imagery data and resample using specialized coordinate systems.
* Broadcast of the annotated frames, with resampling performed locally for specific uses, will provide improvements in image quality by eliminating the cascade of errors that result from resampling generic products produced with resampling.
We identified several areas requiring further analysis, which time and resources did not permit us to address:
* Selecting a suitable compression technique requires additional analysis and investigation of the redundancy in both existing GOES images and modeled high-resolution images as might be produced by the proposed imager. The acceptable amount of information loss must be identified and weighed against the transmission rate reduction benefits of the various techniques. Also, the selected technique must be robust, providing the required compression ratio under the variety of diurnal and seasonal variations that is anticipated in the imager scenes.
* Deciding the allowable level of visible channel array sensitivity nonuniformity that will support our proposed image navigation technique will require evaluation of a variety of visible scenes.
* Selecting suitable techniques for detecting features in the overlap region and determining coincidence of features will require an analysis of image data.
* Determining details of automated landmark recognition, in both visible and IR channels, will require additional analysis of image characteristics