Table of Contents 
Previous Chapter 8 Product Generators
An integral part of the AWIPS automated product preparation process is the final formatting of products. Traditional text products can take the form of plain-language sentences grouped into paragraphs, tables, or strings of hydrometeorological symbols and characters. These three types are referred to as free-form, tabular, and coded products, respectively. AWIPS will also create "broadcast-ready" products, prepared for voice generation for NWR. Eventually, as the NWS moves well into its modernized operations, products will also be produced as graphics, images, and gridded data sets. These products will service all NWS program areas including public, aviation, marine, agriculture, hydrology, and fire weather.
Because of the large number of text products to be issued, some as frequently as hourly, the process must be as automated as possible. The forecaster will always have the facilities to review and edit these products before release, but the goal is to minimize the necessity of manual intervention. Further, any such forecaster editing or modification of the machine-generated products will require additional quality control to correct spelling or other typographical errors. This is particularly important for products prepared for NWR which will be converted from text to voice. If the forecaster feels that changes are necessary in the substance of the forecast, then he or she should modify the database and regenerate the text. This will insure the integrity of the verification and forecast monitoring programs.
NWS products include observations, climatological summaries, routine forecasts, watches, warnings, advisories, statements, and outlooks. These product types are described in Table 12 on page 56.
The NWS defines automated product formatting as the transformation of observations and/or forecasts of hydrometeorological elements and events into products according to user-specific content and form. Depending on the product, this process can include the computation of additional parameters from general forecasts and current weather observations. It can also include the insertion of headlines highlighting hazardous weather conditions. Formatting may also involve the addition of certain protocol and routing information which defines the product, valid time, and where it is to go.
Table 13 on page 56 lists the routine forecast products that will be supported by AFPS. Many products which will be automated or semiautomated on AWIPS are not included in this list. For example, watches, warnings, advisories, and statements will be centrally generated or supported by AWIPS IDB software. Observations summaries and climate reports also fall in this category.
Table 12 - NWS Product Types
------------------------------------------------------------------------------------------------------------------
Product Type Description
------------------------------------------------------------------------------------------------------------------
Observations Reports which contain information about observed weather conditions either at a specific
time or over a period of time, generally within the past 24 hours.
Climatological Summaries of weather conditions for a fixed period of time for selected locations. These
summaries cover periods of one day, one month, a season, or the year.
Routine forecasts Products that describe the current forecasts for the next few hours out to several days.
Watches Forecast products issued when hazardous weather threatens but the occurrence is neither
certain nor imminent.
Warnings Forecast products issued when hazardous weather is imminent or occurring.
Advisories Forecast products issued when hazardous weather is imminent, is occurring, or has a high
probability of occurring. Advisories are issued for events that are expected to be less severe
than for warnings.
Statements Products issued as follow-up information to watches, warnings, or advisories; to cancel
watches, warnings, or advisories; to clear watches; and to provide information on non-
severe, but potentially hazardous conditions.
Outlooks Products issued to provide preliminary discussions for potentially hazardous weather. In
general, these are free-form products and their preparation will not be automated.
------------------------------------------------------------------------------------------------------------------
Table 13 - Products to be supported by AFPS.
--------------------------------------------------------------------------------------------------
Public Aviation Marine Agriculture Fire Weather Hydrology
--------------------------------------------------------------------------------------------------
Coded TAF(a) Coastal Marine Agricultural National Fire Danger Quantitative
Cities Rating System Precipitation
Information
Zone TWEB Great Lakes Nearshore Fruit Frost Fire Weather Presup River Recreation
Marine pression Statement
Area Local Marine Fire Suppression
Great Lakes Smoke Management
Alaska Coastal Marine Land Management
Offshore
--------------------------------------------------------------------------------------------------
- (a)
- The TAF is expected to replace the FT for domestic use before is deployed.AFPS
Product formats are governed largely by the product types and the dissemination media. For example, observations and climatological reports distributed as text over NOAA Weather Wire Service (NWWS) are formatted as tables. The best example of this type of product is the hourly weather roundup, which is generated automatically today on AFOS in tabular form and released without forecaster review or intervention.
This same product is also broadcast over NWR after being read into a microphone by a forecaster or technician. In the AWIPS era, this product will be machine-formatted into a paragraph of complete sentences and made ready for text-to-voice conversion by NWR computers.(1)
W/W/A products will be formatted as paragraphs much as they are today. These products will contain information on the area affected, a description of the hazardous weather or flooding, a call-to-action statement, and a basis statement. They will be generated for both long-fuse (several hours away) and short-fuse (within 6 hours) phenomena. The NWWS and NWR products will look much the same because they are in paragraphs. Because the basis statements are manually prepared, these products will require quality control before being released for transmission or broadcast.
Statements will be largely manually prepared. W/W/A information will be available to provide "boiler plate" information for the products, however.
Routine forecast products such as service area forecasts (zone forecasts), agricultural forecasts, and some marine forecasts offer the most challenges for automated preparation. The products are created by assembling plain-language phrases that decision algorithms select from tables of expressions. These products' formats have been migrating toward a common, period-by-period layout, making their structure similar. This will simplify software development by allowing much code to be shared. (As with some of the products mentioned above, some routine forecast products contain free-text sections that will have to be added by the forecaster. Nonetheless, much of the text can be machine-generated.)
The techniques developed to generate these products will follow AWIPS specifications being prepared by TDL with assistance from OM and OH. Not all products have been specified, but as an example, we cite the formatting of today's zone forecast by ICWF. AWIPS will be able to produce the forecasts at any time of the day. Further, the software will refer to local geography. This will include automated inclusion of local effects, such as "heavier snow near lakes" or "fog in low-lying valleys." Such phrases and areas will be specified office by office. Likewise, recognition of areas that are "always" 5° cooler, for example, will be largely automated. (This method effectively increases the resolution of the grid to accommodate small-scale forecast issues.) Hazardous weather headlines will be inserted automatically into the forecasts through AWIPS data files containing the status of current W/W/As.
The AWIPS forecast database will contain hourly data. Thus, there will be an opportunity to add more detailed forecasts. For example, a public rush-hour forecast could easily be generated from information in the database.
During Phase I, TDL will concentrate on developing product generators to meet routine forecast requirements. Schedules for this development are given in Section 9.3.2.4.
Many of the routine forecasts issued by forecast offices are in a form which can be generated, at least in part, by machine. These products include the 1- to 2-day public forecast, the agricultural forecast, the fire weather forecast, and the marine forecast. In general, they will be in a plain-language, paragraph, period-by-period format in the AWIPS era. Coded messages, such as some aviation forecasts, can also be machine-generated. Examples are International Aerodrome Forecasts (TAFs) and TWEBs.
The public forecast contains a "body" portion accompanied by detail phrases. The body is designed to capture the character of the day, and the detail phrases add definition. Detail phrases are generally sentence fragments covering temperature, wind, snow accumulation, and probability of precipitation. The following is part of a sample public zone forecast. The detail portion is underlined.
TODAY...SHOWERS AND THUNDERSTORMS LIKELY THIS MORNING. BECOMING PARTLY SUNNY THIS AFTERNOON. SEASONABLE TEMPERATURES. HIGH IN THE MID 60S. VERY LIGHT WINDS. CHANCE OF RAIN 80 PERCENT.
In contrast, agricultural, marine, and fire weather forecasts contain only detail phrases. The detail sentences are in an order determined by NWS operations manuals or the local forecast office. The following is an example of an agricultural forecast.
WEDNESDAY...SOUTH WINDS 5 TO 10 MPH. 10 TO 11 HOURS OF SUNSHINE. HIGH DRYING POTENTIAL. MODERATE DEW WILL DRY FROM VEGETATION BY 800 AM. LOWEST HUMIDITY 45 TO 55 PERCENT. HIGH IN THE MID 90S.
All forecast product generators must accurately represent the forecasts contained in the database as plain language products according to NWS operations manuals. TDL has been developing product generators since the early 1970s. Over the last few years, TDL has substantially rewritten the software to generate public zone forecasts. It is this revised software that will be the basis for several of the above products prepared on AWIPS.
Public, agricultural, marine, and fire weather forecasts have many elements in common. For example, the public forecast includes information on temperature, clouds, weather, and winds, while marine forecasts include information on winds, waves, and significant weather.
For each element (e.g., precipitation, temperature, wind), there are a number of phrase types the generator must produce. Each corresponds to a particular weather situation. For example:
rain changing to snow,
rain tapering off to drizzle,
mostly sunny this morning...cloudy this afternoon,
south winds 10 m.p.h.
TDL's ICWF approach has been to build modules in which the phrase types are assembled in a specific phrase structure based on rules of grammar and NWS regulations pertaining to the specific product. A high-level description of the requirements for formatting each element is given below to give the reader some idea of the approach to be used. The discussion and examples are not intended to be exhaustive.
Precipitation information can be complex; many weather elements are devoted to letting the forecaster express this complexity. These are the majority of the weather elements prepared by the forecaster. A forecaster may select up to three possible precipitation types, and each type is described with a probability and intensity. In addition, thunderstorms of several intensities may be forecast with or without precipitation. There are a finite number of phrase types for precipitation, each corresponding to a particular weather situation. For example
- precipitation (with no qualifier),
- precipitation x changing to precipitation y,
- precipitation x mixing with precipitation y,
- precipitation x mixing with and changing to precipitation y,
- precipitation x beginning as a brief period of precipitation y.
In general, there is a phrase structure corresponding to each type of phrase. Each phrase structure consists of one or more parts describing the precipitation type, intensity, probability, beginning and/or ending times, the time of change from one precipitation type to another, etc. There is a particular ordering (Table 14) of the parts (including punctuation) that form the phrase.
Table 14 - Example Phrase Structure
---------------------------------------------------------
PROBABILITY TYPE TIME INTENSITY
A chance of rain this afternoon heavy at times
A slight chance of snow this evening
Occasional sleet after midnight
---------------------------------------------------------
In addition, for precipitation, there are three groups or families of phrase structures: simple, complex, and thunderstorm. The simple structure does not express any change of state, whereas the complex does.
The range and sequence of cloud categories during the forecast period determine the type of cloud phrase to be included in the product. Little change in cloud cover during the forecast period will result in a simple cloud phrase, such as PARTLY CLOUDY.
If both sunny and cloudy periods occur during the forecast, more description is included in the product. Either there is a trend in the forecast cloud amount or the situation is variable. In either case, an attempt is first made to determine whether the period can be split. The period can be split if cloud conditions in one part of the forecast period differ significantly from cloud conditions in the other part of the period. If the period can be split, an attempt is made to describe cloud conditions in each part of the split period. For example:
SUNNY THIS MORNING...THEN INCREASING CLOUDINESS THIS AFTERNOON.
If the period cannot be conveniently split, a more general cloud phrase results:
VARIABLE CLOUDINESS...
All of these conditions fit into one of five basic structures:
· [CLD1]
· [CLD1] [TIME1]....THEN (BECOMING) [CLD2] [TIME2]
· [CLD1] THROUGH [TIME1]....THEN (BECOMING) [CLD2]
· BECOMING [CLD2] BY [TIME2]
· A PERIOD OF [CLD2] AROUND [TIME2]....OTHERWISE [CLD1]
The modules that build wind phrases for the detail portion of a forecast also produce adjective phrases for the body of the public forecast. The wind direction and speed phrase relays information on the speed and direction of the wind during the forecast period, including significant changes in direction and/or speed and wind gusts.
A complete direction/speed phrase is generated from five basic wind direction/speed phrase structures:
· [DIRECTION] winds [SPEED] miles per hour
· [DIRECTION] winds [SPEED] miles per hour shifting to [DIRECTION] [occur time] (and increasing/decreasing to [SPEED] miles per hour)
· [DIRECTION] winds [SPEED] miles per hour increasing/decreasing [occur time] to [SPEED] miles per hour
· Light [DIRECTION] winds becoming [DIRECTION] [occur time] at [SPEED] miles per hour
· DIRECTION winds SPEED miles per hour becoming light [DIRECTION] [occur time]
The direction is determined first. A significant change in direction results in a change of direction phrase (the second phrase above). If no change of direction is found, then a check is made for a change in speed. If no speed change is found, then the first phrase is constructed.
The wind adjective phrase consists of one or several words describing the speed of the wind. For the public zone forecast, the wind adjective is merged into the body of the forecast with elements like sky cover and precipitation. Examples of adjective phrases include BREEZY, WINDY, VERY WINDY, STRONG, BECOMING WINDY, and DIMINISHING WINDS.
As does the wind-phrase module, the temperature module generates an adjective phrase for the body of the forecast and a detail phrase. The maximum/minimum detail temperature phrase provides information on the maximum or minimum temperature during the forecast period. This phrase is usually a stand-alone sentence or sentence fragment for each forecast period. There are three phrase structures:
- When the high and low temperatures occur at normal times, then simple phrases like HIGH NEAR 90 and LOW NEAR 32 are chosen.
- If the maximum or minimum occurs at a climatologically abnormal time, or temperatures are nearly steady during the period, then one of six special phrases is selected and built, such as TEMPERATURES FALLING TO THE LOW 30S.
- If the maximum or minimum does not occur during the forecast period, then a generalized temperature description is selected and constructed, such as TEMPERATURES RISING THROUGH THE 70S.
The 1-hour resolution of temperature data will allow for more detail in these phrases. (It also will be useful for forecasts requiring more temporal detail, such as agricultural forecasts, which also need the full dynamic resolution of the data in the database.)
Temperature adjective phrases consist of one or more words describing the temperature relative to local climatology or a previous temperature. In the public zone forecast, the temperature adjective phrase is merged with the body of the forecast to be part of one or two sentences or sentence fragments describing the weather in general.
The appropriate adjective phrase is retrieved after checking temperature thresholds such as the forecast maximum/minimum temperature, the normal maximum/minimum temperature, and yesterday's maximum/minimum temperature.
Examples of adjective phrases include VERY HOT, UNSEASONABLY WARM, MUCH COOLER, MILD, BITTER COLD, and TURNING COLDER.
The above phrase types appear in the public zone forecasts and in some other routine forecasts. For other forecast products, there are additional phrase structures for
- QPF and snow amount,
- drying potential/conditions,
- maximum/minimum relative humidity,
- dew intensity and dry-off time,
- hours of sunshine,
- lightning activity level,
- 10-hour time lag fuel moisture,
- free air wind, etc.
Phrases referring to these elements are constructed as freestanding or detail phrases and are not typically part of any forecast body. As generators are developed for additional products, program modules will be written to construct these phrases.
Each phrase-formatting module provides the user with two levels of control over phrase construction. Control constants are thresholds that specify local climatological conditions and phrase selection. These constants relay to the formatters local forecast office conventions such as how cold is "cold" or what wind speed constitutes "windy."
The user also may change the detail level, which controls the descriptive content of the phrases. More or less detail may be desired, depending on the forecast period and the other weather elements. For example, the temperature may be "unseasonably cold," it may be "cold," or the detail phrase may be omitted altogether.
Aviation forecasts, including TAFs and TWEBs, are coded messages. Generators for these products require cloud height and amount forecasts for a maximum of three individual layers, visibility, weather types (including obstructions to visibility), and winds, all for projections from 1 to 24 hours. For TAFs, cloud types will also be needed. In addition, forecast information will be needed for prevailing (likely), occasional, and chance conditions. The information for occasional and chance will be used only when clarification or expansion of the prevailing conditions is required.
To create the individual sentences of the TAF, the generator will test the prevailing forecasts for significant changes. Only ceiling height, visibility, weather type, wind speed, and wind direction will be tested. Two sets of criteria will be used to determine significant changes: official criteria (the NWS TAF amendment criteria) and user-defined criteria. The latter will be forecaster-selectable and will allow each office to fine-tune how its terminal forecasts are prepared.
When the prevailing forecast sentences have been determined, the generator will summarize the prevailing conditions within each group, starting with the cloud values and followed by visibility, weather types, and winds, in that order. Then the occasional and chance phrases will be summarized in the same way. Summarization will be performed only for conditions that differ from or clarify the prevailing conditions.
Although the details of TWEB generation have not yet been worked out, the general approach will be to summarize information from the gridpoints along the specified routes.
Footnotes
- (1)
- As part of their work on text generators, TDL will prepare products suitable for NWR voice generation.
- (2)
- The discussion and examples in this section are based on current ICWF practice, and are intended to illustrate the approach. Detailed forecast-formatting requirements will be prepared in the course of development.
- (3)
- See Cammarata, M. W., and J. M. Kosarik: The automatic generation of precipitation phrases for the Interactive Computer Worded Forecast. TDL Office Note 92-9, 27 pp.
- (4)
- See Cammarata, M. W., and J. M. Kosarik: The automatic generation of cloud phrases for the Interactive Computer Worded Forecast. TDL Office Note 92-4, 17 pp.
- (5)
- See Kosarik, J. M., and M. W. Cammarata: The automatic generation of wind phrases for the Interactive Computer Worded Forecast. TDL Office Note 92-3, 17 pp.
- (6)
- See Kosarik, J. M., M. A. Przybocki, M. W. Cammarata, J. F. Wantz, J. Lang, and M. R. Peroutka: The automatic generation of temperature phrases for the Interactive Computer Worded Forecast. TDL Office Note 92-8, 20 pp.
Table of Contents 
Next Chapter