A Comparison Study of MOS Temperature Bias - FWC Versus Fan

Tony M. Ansuini
National Weather Service Forecast Office
Indianapolis, Indiana


The statistical guidance message produced from the National Centers for Environmental Prediction's (NCEP), formally the National Meteorological Center (NMC), Aviation (AVN) run of the Global Spectral Model was disseminated to the National Weather Service Offices (NWSO) during the late summer of 1994. The implementation of this AVN-based (FAN) statistical guidance message (Jensenius et al. 1994) gave the NWSFO Indianapolis (IND) another guidance product, along with the NGM-based (FWC) Model Output Statistics (MOS) (Dallavalle et al. 1992) to use for the verification of its temperature forecasts.


The purpose of this study was to focus on how the temperature biases of the two MOS guidance products could be used at the NWSFO IND to produce improved temperature verification results. A goal was to develop some "rules of thumb" or guidelines that operational forecasters could consider when using the MOS temperature guidance.

For the study, the data set was a year's (October 1994 through September 1995) worth of temperature forecasts. The collected data included temperature forecasts for the first three forecast periods (2,190 forecasts) of the FAN MOS, the FWC MOS, and for comparison purposes the Coded Cities Forecast (CCF) for IND. The actual observed temperatures which occurred were retrieved from archived MAPSO data.


The study initially examined the verification accuracy of the individual MOS temperature forecasts. This was done by using the Mean Average Error (MAE) for the combined first three periods of the forecast (Figure 1). It was shown that the NWSFO IND forecasters were able to consistently improve over both the FWC MOS and the FAN MOS temperature guidance forecasts. The greater improvement came against the FAN MOS.

The FWC MOS also showed a consistent improvement over the FAN MOS. The most significant improvement over the FAN MOS, by both the NWSFO and the FWC MOS, was during the warm season (April through September). A lesser improvement occurred during the cool season (October through March).

Once the performance of the individual MOS guidance was determined, the focus of the study then shifted to determining the individual MOS temperature bias. The results of this work showed that the FAN MOS had a consistent, and in some months, a significant cold bias. In general, a cold bias was also shown for the FWC MOS, as well as the NWSFO IND (Table 1, Figure 2); however, not to the extent of the FAN MOS. Also, the FAN MOS, despite its obvious cold bias, did experience a warm bias, once, during the middle of the cool season (January).

Figure 1. Monthly distribution graph of Mean Average Error (degrees F).

Distribution table of temperature bias by periods
(12 months x 3 periods = 36 possible periods).

Cold 23 20 32
Warm 12 13 4
Neutral 1 3 0

In Figure 2, the more random pattern associated with the FWC MOS, as well as the NWSFO IND, temperature bias is evident. In every month, the FWC MOS had a warmer bias than that of the FAN MOS. The FAN MOS showed a gradual decrease from neutral at the end of the cool season through the warm season with an eventual "cold bias peak" occurring during the middle of the warm season (July). It was also noted, a similar, but opposite trend occurred at the end of the warm season through the cool season bringing the FAN MOS temperature bias back toward neutral and then eventually to a "warm bias peak" during the middle of the cool season (January).

Figure 2. Monthly distribution graph of average temperature bias (degrees F).

As shown from Figures 2 and 3, the cold bias of the FAN MOS was the greatest during the warm season months, especially June through September. During this time, the greatest amount of cold bias was found to occur in the third period forecast, while relatively speaking, the least amount of cold bias occurred in the first period. Interestingly, the exact opposite outcome occurred during the cool season. The FAN MOS third period temperature forecast exhibited more of a relative warm bias, as compared to the first period forecast.

From Figure 3, we can see that the third period temperature forecast of the FAN MOS was most variable. Not surprisingly, less variability occurred in the first period temperature forecast. As seen from Figure 4, the FAN MOS third period temperature forecast (as well as the first and second periods) had a cold bias maximum in the middle of the warm season (July). A warm bias maximum was exhibited in the middle of the cool season (January).

Work was then done to determine if there was a certain bias when either MOS was forecasting a maximum or a minimum temperature. From Figures 4 and 5, a graphic illustration is made of the maximum temperature, and the minimum temperature biases of the two MOS products, as well as the NWSFO IND.

On a monthly basis, the cold bias of the FAN MOS occurred both in its maximum, and its minimum temperature forecasts. The FAN MOS warm bias exhibited in January was seen, both in its maximum and minimum temperature forecasts. Therefore, in every month when a certain bias occurred (cold or warm) in the FAN MOS maximum temperature, the same bias was true of its minimum temperature (to varying degrees), and vice versa. The data showed a similar trend in the FWC MOS (7 months exhibited the same bias in its maximum and minimum temperature forecast).

Figure 3. Monthly distribution graph of FAN MOS temperature bias by periods (degrees F).

The NWSFO IND forecasters actually deviated from this trend (only 4 months had the same bias in maximum and minimum temperature forecasts). As shown from Figure 4, there was an obvious cold bias associated with FAN MOS maximum temperature forecast. Nearly half the months (5) had an excessive cold bias (greater than; 1.0 degree) as compared to its minimum temperature forecast (2 months). The other months had only slight differences in bias between maximum and minimum temperature forecasts.

Figure 4. Monthly distribution graph of average maximum temperature bias (degrees F).

Figure 5. Monthly distribution graph of average minimum temperature bias (degrees F).


The purpose of this study was to give the forecast staff at the NWSFO IND an insight into what the individual MOS temperature biases were. By knowing these statistical biases and then incorporating them with the weather patterns, the forecaster can make appropriate adjustments to the MOS guidance and produce a better temperature forecast.

This work determined the following "rules of thumb" or guidelines associated with MOS temperature biases. The forecaster should consider these when applying MOS temperature guidance in their forecast.




Dallavalle, J.P., J.S. Jensensius Jr., and S.A. Gilbert, 1992: NGM-Based MOS Guidance - The FOUS14/FWC Message. NWS Technical Procedures Bulletin No. 408, NOAA/DOC/NWS/Office of Meteorology, Program Requirements and Development Division, 16pp.

Jensenius, Jr., J.S., J.P. Dallavalle, and S.A. Gilbert, 1994: The AVN-Based Statistical Guidance Message. NWS Technical Procedures Bulletin No. 415, NOAA/DOC/NWS Office of Meteorology.


USA.gov is the U.S. government's official web portal to all federal, state and local government web resources and services.