Snowband Probability Prototype Page

Available Data

Ensemble Resolution Runs Per Day Membership Run Length
HREF ~3-4 km 2 10 (5 time-lagged) 10 mem to 36 h, 7 mem to 48 h
FV3 LAM ~3 km 2 3 60 h
HRRR TLE ~3 km 6 3 (2 time-lagged) 36 h

Additional Details

  • The HREF consists of 5 current and 5 time-lagged members and updates at 00 and 12 UTC. The models that make up the HREF are the current HRRR, the 6-hour time-lagged HRRR, the current ARW, the 12-hour time-lagged ARW, the ARW2, the 12-hour time-lagged ARW2, the current FV3, and the 12-hour time-lagged FV3. Note that the ARW2 is designed to mimic the NSSL model.
  • The FV3 LAMs consist of three experimental models run at EMC (FV3LAM, FV3LAMx, and FV3LAMda), and are run out to hour 60 two times a day at 00 and 12 UTC.
  • The HRRR is a Time Lagged Ensemble (TLE), but consists of only the extended runs (i.e. the runs that go out to forecast hour 48), which are performed every 6th hour.
  • MODE-Time-Domain (MTD) Info

    The Method for Object-Based Diagnostic Evaluation-Time-Domain (MODE-TD) tool from the Model Evaluation Tools v10.0 software developed by the Developmental Testbed Center (DTC) is used to identify and track ensemble member Quantitative Precipitation Forecast (QPF) objects. 1-hour accumulated QPF from all of the members within each ensemble listed above are used as input into MODE-TD. To identify snow-only QPF, all models are masked using the categorical snow precipitation type grid.

    After applying the snow-only mask, MODE-TD is used to identify and track modeled snowbands through space and time. MODE-TD initially applies a convolution filter (5 grid spaces) to smooth the data, before masking and removing all data below a certain hourly precipitation amount. This simplifies the process of identifying the desired objects by removing small, noisy objects. Thereafter, MODE-TD is run to identify modeled objects and their attributes (e.g., centroid location, intensity, orientation, area). The methodology for identifying objects is complex and involves the following two processes illustrated in the figure below:

      1. Identify coherent space/time objects of a certain size
      2. Merge nearby space/time objects that are likely the same object

    The above figure illustrates an example of this process where raw precipitation is input into MODE-TD in (a), resulting in two identifiable objects in (b), with only the raw QPF within these objects being retained in (c). For more information, please see the DTC description about MODE-TD at https://dtcenter.org/met/users/index.php.

    About the Graphics

    As discussed above, MODE-TD outputs a variety of object attributes, including object centroid location, velocity, orientation, area, and intensity. Only some of these attributes may be informative to forecasters looking at ensemble data. To create visually appealing graphics of the most useful attributes, only the outline (i.e. object area) and intensity of the object are displayed. The intensity is calculated as the 90th percentile of QPF within each object (see panel c in the figure above), and is represented by the color of the outline of the object.

    About dModel/dt

    dModel/dt is a tool to look at the forecast evolution or trends for a particular valid time. This is a function of both how far out a particular model forecast extends and the how often a new model cycle is available. For example, dModel/dt of a 1-h hourly-updating HRRRv3 forecast will have 14 of the previous forecasts for a fixed valid time beginning with the 15-h forecast from the oldest model initialization. However, looking at dModel/dt for a 12-h forecast will only show the previous 3 forecast cycles valid for that time from an hourly-updating model. For a model updating less frequently like the HREF that is issued every 12 h out to 48 h, no dModel/dt is available after forecast hour 36 because that extends beyond the forecast range of the previous cycle.