Link to NWS Louisville Science and Technology Homepage Performance Characteristics and Biases of the Operational Forecast Models
Green Line Separater

I. Biases of the Short-Range Numerical Models (Days 0-3).

  • A. The Aviation Model (AVN)

    • Since this model is generated from the NCEP Global Spectral Model (GSM), as true for the MRF, the AVN tends to have a "cold" bias, mainly across the eastern US. Remember that the AVN and MRF are the same numerical model; the difference is in the data cut off times. The AVN will generally display the same biases as the MRF.

    • Overall, for situations when cyclogenesis is possible, the AVN may be the first short-range operational NWP model to capture the event. However, this model can be too slow with the deepening rates of sea-level cyclones. The AVN model is the best at simulating filling cyclones.

    • The 1200 UTC AVN model can be quite useful for detecting important forecast "trends" for all the operational models. A good example would be comparing the "latest 72-hr AVN" with "last nights 84-hr MRF" model solutions.

    • This model tends to have the smallest position errors for cyclones and anticyclones at days 2 and 3. In general, for sea-level lows, the AVN does have a tendency to "track them" too far northwest for cases of cyclogenesis to the lee of the Rocky Mountains. The latter is especially the case when the mid/upper tropospheric trough (helping to initiate the sea-level response) is still undergoing amplification over the Great Basin. Often the AVN (and NGM) will over develop a lead short-wave trough and allow its unrealistically deep sea-level response to advance too far north and west. Once the mid/upper tropospheric geopotential trough and/or closed cyclone center has advanced to the lee of the Rockies, the sea-level cyclone placement is fairly reasonable (generally a baroclinic mode of development).

    • The AVN often underdevelops surface lows, especially over the oceans.

    • This model tends to underpredict precipitation amounts during the warm season for greater than 0.50" events.

    • There is a tendency to predict 1000-500 mb thicknesses too low over sea-level cyclones.

    • This model can be too slow with deepening rates of baroclinic cyclones, especially during the first 12 hours.

  • B. The Nested Grid Model (NGM)

    • The NGM tends to overdeepen sea-level cyclones over land, and underdevelop them across the oceans. This model can be too slow to fill weakening cyclones.

    • The NGM has a "northern bias" with respect to storm track prediction, similar to the AVN model. In general, the NGM will be farthest north of the 3 short-range operational models discussed here. In the case of the NGM, this northern bias is also true for the eastern US.

    • The NGM tends to predict sea-level pressures to be too high for anticyclones across the western USA.

    • The NGM tends to overpredict the magnitudes of sea-breeze convergence zones, and the precipitation which results.

    • During the warm season, the NGM has a bias to overpredict light precipitation (amounts less than 0.50"), especially areal coverage.

    • Near the Gulf of Mexico, the NGM very often underforecasts 0.50" to 1.00" rains, and GROSSLY underforecasts amounts heavier than that.

    • Like the AVN, the NGM has a "cold bias" with respect to the predictions of 1000 to 500-mb thicknesses over sea-level cyclones, during the winter.

    • This model can be deficient with the deepening rates of cyclones across the Gulf Stream during the winter.

    • This model tends to overdevelop cyclones in the lee of model terrain features.

    • The NGM predicts too much stability through too deep of a layer with temperature inversions.

    • Of all the operational NWP models, this model can be the most deficient with "digging troughs" across the western US (the only exception may be the UKMET).

    • The NGM has a bias to underforecast precipitation amounts onto the west coast of the US in the presence of a "blocking ridge" across the Gulf of Alaska.

    • All models, especially the NGM, forecast too much precipitation across Texas during "pre-frontal squall line" situations. That is, once the squall line has moved east and middle tropospheric winds have veered to more westerly, the rain event for Texas has usually ended. Often the NGM will still predict rainfall. This bias most true during and spring and fall/early winter.

    • The overall errors in the NGM decrease during the warmer months.

  • C. The Eta Model

    • Of the short-range models, with respect to forecasted cyclone storm track, the Eta seems to have the most southern bias, especially from east of the Rocky Mountains to west of the Appalachian Mountains. East of the Appalachians ("east coast storms"), this model can be too far northwest, even more so than the AVN/NGM.

    • The Eta normally is superior to the AVN/NGM with predicting precipitation amounts from deep moist convection.

    • This model is superior to the NGM/AVN in forecasting the southward extent and rate of movement of Arctic airmasses across the middle of the country, east of the Rockies (especially seen by utilizing model predicted boundary-layer static stabilities).

    • The Eta outperforms the NGM/AVN for forecasting precipitation to the lee of the Cascades and Sierra mountain ranges (less "spill over").

    • This model can be too conservative ("underdone") with moisture returns from the Gulf of Mexico, which can result in computations of lifted indices to be too high (too much stability).

II. Biases of the Medium-Range Numerical Models (Days 4-10).

  • A. The Medium-Range Forecast Model (MRF)

    • Shallow cold air is not handled well, i.e., the MRF will likely be too slow with southward transports of shallow cold air masses, especially arctic air to the east of the Rocky Mountains. This problem generally is true for all the operational forecast models.

    • The model tends to have a wet bias, i.e., areal extents of precipitation often are overforecast, especially light amounts.

    • Often the magnitudes of easterly boundary layer upslope flows along the Front-Range of the Rocky Mountains are overpredicted, i.e., pressure gradients are too strong.

    • The model has a slight cold (heights too low) bias, especially for about the eastern third of the country, with respect to the prediction of mid/upper tropospheric geopotential heights and resultant thickness calculations. Often the MRF will dig troughs in the height field too far south across the Great Lakes and Northeast regions, most noticeably after about Day 3. During the cold season (mainly October-April), the MRF will depress the storm track too far south across the Plains states as a response to this cold bias. Forecasters across especially the Northern Plains/Upper Mississippi Valley areas need to be aware of this characteristic and be careful predicting, for example, snowstorms to "go to their south".

    • Continuing on this "cold bias" problem, the ramifications to the placement of the MRF model storm track may also be regime dependent. However, this is a personal observation and requires considerable objective quantification. For instance, as was discussed during the AVN section, for the Rockies and Plains States, there may indeed be a "northern and western bias" in the presence of a Reverse Pacific/North American Teleconnection (RNA or negative phase PNA -- western North American trough/southeastern USA ridge, mid/upper tropospheric geopotential height field -- other regimes can also project this circulation structure with respect to the USA -- see reference cited below). The "southern bias", especially for areas east of the Appalachian Mountains, may be more likely during a positive PNA (western North American Ridge/eastern North American trough).

    • Somewhat related to this model's cold bias, the MRF often looses too much "energy" too quickly for mid/upper tropospheric cyclones translating northeastward from the southwest deserts. There is often too much "weakening" in confluent flow.

    • The model does have a bias to meridionally couple separate branches of the westerlies (i.e., too much "phasing" occurs between northern and southern stream weather systems).

    • The MRF still has a problem maintaining correct amplitude to low wave number regimes. That is, after about forecast day 3, this model has a tendency to erroneously de-amplify regimes where the planetary-scale is dominate. An example would be the MRF model "crashing" a western North American mid/upper tropospheric ridge in the height field prematurely. In the latter case, the MRF may not be maintaining enough amplitude to the positive projection of the Pacific/North American Teleconnection (PNA; see Barnston and Livezey, Mon. Wea. Rev., June 1987). This problem is important during situations involving Arctic outbreaks. Also, perhaps related to the MRF's problematical prediction of the planetary-scale waves, the MRF can be deficient with "digging troughs" across the western US during "split flow" situations.

    • Still addressing the planetary-scale waves, especially poleward of about 50 deg north, the MRF has a tendency to be too retrogressive, most noticeable beyond day 5. For example, this model "likes" to retrograde mid/upper tropospheric western North American ridges to the point that the response of the flow regime across the US is a "flip-flop", from a western ridge/eastern trough to western trough/eastern ridge.

  • B. The UKMET (United Kingdom Meteorological Office, Bracknell U.K.) Medium-Range Forecast Model.

    • Like the MRF, the UKMET model has problems with shallow cold air.

    • The model tends to be "too zonal", especially after about Day 3.

    • The UKMET tends to progress shorter wavelength features too quickly.

    • Westerlies are often too far south.

    • The model likes to lower sea-level pressures too much and too far south. A result of this problem can be that synoptic-scale fronts are implied to be too far south.

    • This model rarely develops mid/upper tropospheric closed cyclones and anticyclones. This is likely a result of its "zonal" bias.

  • C. The ECMWF (European Center for Medium-Range Weather Forecasts, Shinfield Park, Reading U.K.) Medium-Range Forecast Model.

    • Overall, the ECMWF does well in predicting mid/upper tropospheric heights during the colder part of the year (such as October through April). Also, the ECMWF model tends to perform quite well in predicting amplitudes of planetary-scale regimes such as the Pacific/North American teleconnection (PNA). This model can also perform outstandingly during low to high planetary-scale wave number transition events, and northern hemispheric-scale regime transitions (Berry et al. 1996, CR TM 111).

    • This model does the best of the medium-range forecast models during "shallow cold air situations".

    • There is a bias in the model to "overdevelop" mid/upper tropospheric cyclones across the southwestern U.S. Situations arise where this model will be too slow to predict the movement of cyclones from the southwest deserts.

    • This model has a slight tendency to forecast mid/upper tropospheric heights and the resultant thickness calculations too high (i.e., a "warm" bias).

    • There are some situations, especially during the warmer portion of the annual cycle, when this model has too many closed gyres. This bias may be related to its T216 wave number resolution.

Additional information on model biases is available from the Hydrometeorological Prediction Center of NCEP.

Green Line Separater

Back to Training Documents and Exercises is the U.S. government's official web portal to all federal, state and local government web resources and services.