[ About UW Probcast ]
About UW Probcast
UW Probcast is a web application currently under development at the University of Washington, a partnership between the departments of statistics, atmospheric sciences, psychology, and the Applied Physics Laboratory. The work is funded by a MURI grant supplied by the Office of Naval Resources in support of efforts to develop new methods in creating and providing probabilistic forecasts.
Users who have experience with other popular weather web sites around the web should by and large find UW Probcast familiar. However, some features are likely to seem new to most users; many of these provide information about probability and forecast uncertainty.
No weather forecast is perfect; when a forecaster provides information about tomorrow's high temperature, there is a certain amount of inaccuracy implied in the figure. Variations in elevation, geography and terrain, in addition to the inherently unpredictable nature of the weather, all contribute to these inaccuracies. However, while any well-considered forecast attempts to land as close to the truth as possible, traditional weather forecasts include no information relating to just how accurate they can expect to be, or the likelihood of an outcome that, while not the most probable eventuality, could still have a significant chance of happening. The UW probcast web site attempts to address these omissions.
Using the web site
Choosing a Location
Upon first arriving at the web page, the user will see a forecast for a default location (possibly the location of the web site developer). To view a forecast of a new location, the users should enter their zip code (or city and state in the format "city, ST") into the "Zip Code" box at the upper left of the display, and hit return (for instance, entering "98068" brings up a weather forecast for Snoqualmie Pass, WA). The current location is listed at the top of the forecast table and also shown as a gold star on the forecast map.
It is also possible for users to click the map, selecting the forecast location by using the mouse. The is the only current way for residents of Canada to select a location, as the current zip code data covers US territories only.
Features and display
The web site display has two major elements: a forecast table that provides a forecast of many weather parameters for one location (located at the top of the page) and a forecast map (located below the table) that provides information for a single weather parameter across the Pacific Northwest. The location currently displayed in the table is shown on the map as a gold star.
The forecast table is a feature that should seem familiar to those users who regularly view weather forecasts on the web. It provides conventional forecasts of temperature and precipitation (inclusive of rain and snow at present time) in addition to probabilistic elements, such as an upper or lower possible but unlikely extreme of temperature or precipitation or probability of freezing or probability of precipitation. These forecasts refer to a single location over five 12-hour long periods that can be changed by entering a zip code (or city and state) into the text box at the upper right of the table. Every numeric forecast also links to display the corresponding map of this feature at the associated time, allowing the user to quickly view forecast and probability maps.
The weather map provides a single forecast parameter (high temperature, probability of precipitation, etc) for a single forecast period. The map provides a summary of the particular parameter for the period of time - either predicted total accumulation for precipitation or a predicted maximum or minimum for the period in the case of temperature. The relevant time period is displayed above the weather map along with a description of the forecast parameter currently being displayed. Chance of freezing and chance of precipitation refer to the probability that a given event will occur at any time within the given time period.
The forecast is provided for five forecast "periods" referring to alternating day or night periods. Every period begins at either 0 or 12 hours UTC (0 UTC is either 4pm or 5pm and 12 UTC is either 4am or 5am for the US West Coast, depending on the current status of daylight savings) and covers a period either 12 hours long in the case of accumulated precipitation or 18 hours in the case of temperature. Due to the fact that the lowest temperature often occurs after what would otherwise be the 4/5am day/night cutoff, temperature periods overlap.
Upper and Lower Extremes
Every temperature forecast on the UW Probcast web site includes a possible but unlikely upper and lower extreme (referred to as 'as high as' or 'as low as' in the table). These represent outcomes that, while unlikely, remain possible and can be viewed as reasonable alternate scenarios to the 'most likely' forecast. Specifically, the possible but unlikely upper extreme of the daytime high temperature represents a temperature that, 9 times in 10 (or 90% of the time), will not be reached during the corresponding period. However, 1 time in 10 (or 10% of the time) we expect the daytime high to reach or even exceed this value. Similarly, the possible but unlikely lower extreme of the daytime high is a unlikely prediction for temperature that we expect the actual outcome to exceed 9 times in 10 (or 90% of the time). However, 1 time in 10 (or 10% of the time) we can expect the measured daytime high to be at or lower than this value.
In addition, the predicted most likely precipitation forecast is accompanied by an unlikely but possible upper extreme amount (referred to as 'As much as' on the web site). This figure represents a upper extreme for the predicted amount of rain or equivalent snow fall, a much less likely outcome than the predicted most likely amount, but still an outcome with a reasonable chance - 1 time in 10 or 10% - of occurring.
Probability of Freezing or Precipitation
In addition to the extremes, the UW Probcast forecast also includes predicted probabilities that certain events, such as freezing temperatures or measurable rain or snow occur during the given time period. These are given as "Chance freeze" and "Chance of precip" respectively. It is important to note that the weather maps of these parameters are given in units of percent probability, as opposed to degrees Fahrenheit or inches of rain.
How the Probcast forecast is created
The forecasts displayed on the UW probcast web site begin at the University of Washington's atmospheric sciences department, where they are generated as part of the MM5 ensemble system hosted there. This systems uses the MM5 model to generate multiple forecasts with different initial conditions in an attempt to capture multiple possible outcomes. After the forecast files are transferred to the UW Applied Physics Laboratory, they are combined with observational data and fed into a Bayesian Model Averaging (BMA) technique developed by the UW department of statistics. BMA uses this training data set to provide the actual calibrated probabilistic forecast parameters in addition applying bias correction. The web site display is then generated by combining a set of model predictions from the ensemble with the BMA results, a process akin to creating a weighted average of probability distributions built around the individual forecasts and then integrating the single resulting distribution.
Location information is currently pulled from 2000 census information. Some more recently created zip codes may not be available from this list.
Current zip codes cover the United States only.
While the total forecast lead time - 60 hours for 5 forecast periods - is shorter than most conventional weather web sites, the UW probcast web site is limited by the range of the forecasts provided by the MM5 model that supply the initial data from which these predictions are assembled.
It can sometimes take the server several seconds to generate a response. We appreciate your patience.
Sometimes question marks appear at the end of the forecast table. These represents times in the future for which no forecast is currently available.
Forecast parameter (Weather parameter) - Refers to a single element of the forecast, e.g. daytime high temperature or probability of measurable rain
Forecast period - A time period over which a forecast is valid; for accumulated precipitation forecasts, UW probcast uses a daytime window that stretches from 12 hours UTC to 0 hours UTC (corresponding to 4am to 4pm or 5am to 5pm local time, depending on daylight savings) and a nighttime window that stretches from 12 hour UTC to 0 hours UTC (corresponding to 4pm to 4am or 5pm to 5am local time). For temperature forecasts, the windows are longer and overlaps by six ours to include all relevant maxima and minima. The daytime temperature window lasts from 12 hours UTC to 6 hours UTC on the following day (a total of 18 hours from 4am to 10pm or 5am to 11pm). The nighttime temperature window lasts from 0 hours UTC to 18 hours UTC (from 4pm to 10am or 5pm to 11am the next morning).
Unlikely but possible extreme high temperature ('As high as') - An unlikely but possible upper extreme for high temperature; 9 times in 10 the high temperature will be less than this value.
Unlikely but possible extreme low temperature ('As low as') - An unlikely but possible lower extreme for low temperature; 9 times in 10 the low temperature will remain above this value.
Unlikely but possible upper extreme for amount of precipitation ('As much as') - An unlikely but possible upper extreme for the amount of precipitation; 9 times in 10 the actual amount of precipitation is predicted to be less than this amount.
Chance of Freezing - chance, typically given in percent rounded to the nearest 5%, that the temperature will dip below freezing at any point and for any length of time during the corresponding time period.
Chance of Precipitation (also: Probability of Precipitation or POP) - The chance, typically given in percent rounded to the nearest 5%, that any amount of precipitation will fall on the given location at any time during the corresponding time period.
Precipitation - In this context refers to any form of water falling from the sky, be it liquid rain, freezing rain, or the equivalent amount of snowfall (typically ~10 times the amount given for liquid rain).
Likely amount of Precipitation (also: Quantitative Precipitation Forecast or QPF) - This figure represents the most likely, or best guess, at the amount of precipitation predicted to fall during the corresponding time period for the specified location. The figure is typically displayed in inches or hundredths of an inch.
Bayesian Model Averaging - A technique used to create a single forecast and estimations of uncertainty from a set of multiple predictions.
(From Chris Volinsky) Bayesian Model Averaging is a technique designed to help account for the uncertainty inherent in the model selection process, something which traditional statistical analysis often neglects. By averaging over many different competing models, BMA incorporates model uncertainty into conclusions about parameters and prediction.