Frequently Asked Questions
Lexicon of MOS precipitation type terms:
Conditional probability of freezing/frozen/liquid: Assume that some form of precipitation is falling. What is the probability that precipitation will fall as freezing (sleet, freezing rain), frozen (snow), or liquid (rain)? Because the three categories are mutually exclusive, these probabilities will sum to 100%.
Unconditional probability of freezing/frozen/liquid: What is the probability that some type of precipitation will fall and that type will be sleet/freezing rain; something will fall and it will be snow; or something will fall and it will be rain? These probabilities are calculated by multiplying the conditional probabilties for each precipitation type by the probability of precipitation "on the hour" (PoPO). The unconditional probabilities will sum to the PoPO value -- if PoPO is near zero or very small, the unconditional probabilities also will be very small.
Conditional best category: Assume that some form of precipitation is falling. What is the highest impact precipitation type that is likely to occur (freezing, frozen, or liquid)? See below for an explanation of how the best category is determined.
Probability of precipitation occurrence (PoPO): What is the probability that some type of precipitation will occur "on the hour" -- or in other words at the top of the indicated hour? MOS PoPO is used in the computation of the unconditional precipitation type probabilities (see definition above).
Generalized operator equation: The MOS precipitation type system is developed by pooling together data at METAR sites into one large region - this is known as a "generalized operator" approach. This method is often employed when the event being forecast (e.g. freezing precipitation) does not occur frequently enough at individual stations to develop a stable single-station equation. Pooling data into one or more regions helps to increase the number of events in the sample, and the resulting equation is applicable to all stations within the region.
Q: How are the precipitation type categories defined?
A: The conditional precipitation type categories are derived from hourly METAR observations of present weather, considering only cases in which precipitation (of any type or intensity) was observed. For precipitation cases, observations of present weather valid at 00, 03, 06, 09, 12, 15, 18, and 21 UTC were classified into one of three mutually exclusive categories: freezing, frozen, or liquid. The following table lists the present weather observations associated with each category:
Q: Why does the freezing category include freezing rain and sleet together?
A: Neither freezing rain nor sleet by itself occurs frequently enough to get an adequate sample to develop robust MOS equations for each separately. In fact, even combining the two events into one category comprises at most 5% of all precipitation cases over the CONUS. Also, we have found that model composite soundings for freezing rain and sleet cases do not differ significantly enough to be able to distinguish between the two events using the MOS technique.
Q: What time period are the probabilities valid for?
A: The conditional and unconditional probabilities are valid for a point in time -- at the top of each forecast valid hour -- not over a period of time such as 3-, 6-, or 12 hours.
Q: How is the conditional best category determined?
A: Categorical forecasts of freezing, frozen, or liquid precipitation, assuming precipitation occurs, are derived from the MOS conditional probabilities. The probability forecasts are compared to two threshold probabilities calculated from the developmental sample for each forecast projection and model cycle. The thresholds selected were those which maximized the threat score on the dependent sample, while also constraining the bias to within a reasonable range (between 0.98 and 1.02). Once the thresholds are obtained, the best category is determined in order of highest impact to lowest impact using the following procedure:
Step 1: Freezing is the highest impact category so it is assessed first. The conditional probability of freezing is compared to a statistically-derived threshold value. If the probability exceeds the threshold, freezing is forecast, otherwise move to step 2.
Step 2: Frozen is the next highest impact category. The sum of the freezing and frozen probabilities is computed and compared to a statistically-derived threshold value. If the sum exceeds the threshold, frozen is forecast, otherwise move to step 3.
Step 3: If the freezing and frozen categories are eliminated, the default forecast is liquid. The user should keep in mind that a forecast of liquid also includes the possibility of a mix of rain and snow.
Q: For obtaining the best category, is there one threshold value that is used for each precipitation type, or do they vary? What are typical values for the thresholds?
A: The thresholds vary by model cycle and forecast projection, but do not vary spatially (i.e. for a particular cycle and projection the same threshold is applied everywhere on the grid). For the early forecast projections, a typical threshold for the freezing category is around 25% and a typical threshold for frozen is around 45%. The thresholds gradually decrease for longer forecast lead times. At the 180-h projection and beyond, typical values will be around 10% for the freezing category and around 25% for frozen.
Q: Are the MOS forecasts based on one run of the GFS, multiple runs of the GFS, or an ensemble?
A: The MOS precipitation type forecasts are based on a single deterministic run of the GFS, with separate forecasts generated out to 192 hours for both the 00 and 12 UTC cycles. The cycle can be selected in the first drop-down menu on the image display page. The probabilities are not cycle-averaged.
Q: How is climatology factored into the probability calculations?
A: Climatology is one of the most important predictors in the MOS equations. For each METAR site with a sufficiently large record of present wx observations, monthly relative frequencies of freezing, frozen, and liquid precipitation types were calculated. The relative frequencies at the stations were analyzed to the 2.5-km NDGD grid using the Bergthorssen, Cressman, Doos, and Glahn (BCDG) analysis technique, which applies successive corrections to a first-guess field and vertical adjustments to account for variations of a field with elevation. The gridded precipitation type relative frequencies can be viewed here.
Q: What fields are available in the GRIB2 files?
A: See table below (HH=model cycle):
Q: How often does this page update?
A: Each page updates twice per day. The updates occur after the valid time for the final projection (192 hours from original start time) has been reached for the corresponding Gridded MOS run. Images will update around 5:45 UTC for 00z cycle and around 17:45 UTC for 12Z cycle. For precipitation type, images are produced for every 3 hours from 6 to 192 hours out from the original GMOS run.
Q: Why aren't the products on your website always the most up-to-date?