Preliminary MM5 Forecast Assessment

A preliminary assessment of the MM5 forecast system was performed after the 2001 Summer season. For simplicity, the "bias" of the forecasted values was selected for this validation exercise. Bias measures the tendency of the model to over-forecast or under-forecast a meteorological parameter. For example, if the model has a positive temperature bias, on average the forecast temperature exceeds the observed temperature. Bias is a calculated using:

Bias = Forecasted - Observed

The table below lists the values for "good", "fair", and "poor" forecasts based on temperature, wind speed and direction, and relative humidity biases. Criteria were determined from personal communication with Daran Rife (2001) at NCAR/RAP and from White et al. (1999).

Data Set Description

Forecasters at Abu Dhabi International Airport provided validation data of daily values of wind speed and direction, sea breeze onset and strength, maximum and minimum temperatures, maximum and minimum RH, clouds, and precipitation for both Abu Dhabi and Al Ain on 41 days. Upper level winds and temperatures at 12 and 24 hrs, freezing levels, and lifting condensation levels were also provided for Abu Dhabi on 38 days. Forecasters at Dubai International Airport provided hourly values of temperature, dewpoint, wind speed and direction on 23 days.

Results

For analysis of the Abu Dhabi data, an averaging period of one to two weeks was used to assure there are at least three values in any given average. This variability is due to the sporadic nature of the assessment reports from the airports.

With the exception of one or two points, maximum and minimum temperature and wind speed lie within +2. This suggests that, on average, the prototype MM5 model forecasts of these variables were good for the city of Abu Dhabi. Likewise, wind direction falls well within the +45° criteria and both maximum and minimum relative humidity are between +13%. The minimum RH bias is positive, which means the model is predicting higher (or moister) values of RH than are observed. The bias line is close to the upper bound of the "good" criteria, suggesting only a "fair" amount of skill in the MM5 forecast of minimum RH. The maximum RH bias tends to be negative and therefore drier than observed. It is closer to the zero bias line for the first half of the assessment period and then dips toward the upper limit of the "good" criteria. For this reason, the MM5 skill in forecasting maximum relative humidity may be assessed as "fair to good".

Comparison of daily values of maximum temperature for AUH and Dubai indicate there is a high degree of variability (2 to 4o C) in the data. The variability in day-to-day forecast biases as well as the differences between Abu Dhabi and Dubai forecasts from the MM5 model suggests that more model development may be necessary to fine tune the forecasts. It also implies that mesoscale features on scales smaller than the highest resolution domain (3.3 km) may be important.

The trend in forecast bias when averaged over three-hour intervals for all 23 assessment days provided by Dubai show on average, wind speed bias remains fairly constant throughout the day. Temperature bias stays within "good" criteria and tends towards zero between 06 and 08 UTC. Dewpoint bias peaks at 4° C at the same time and then slowly decreases back to 1° C. The relative humidity bias is driven by both temperature and dewpoint biases, and swings from -5% to +6% during this same period. A shift in wind direction bias from 6° between 00 and 02 UTC to -8° between 06 and 08 UTC may be driven by the onset of the sea breeze during this time period. It may also explain the changes in temperature, dewpoint and relative humidity.

The table below provides a mean bias of MM5 upper air forecasts of temperature and wind speed for Abu Dhabi during Summer 2001 and for desert southwest of USA during fall 1997 (White et al., 1999). The comparison indicates that forecasted temperatures and wind speeds for Abu Dhabi have similar or smaller biases than those over a similar region in the U.S. It is encouraging that the prototype MM5 upper air forecasts have similar biases to models in the U.S.

Based on the feedback from the users of the model in the UAE and our own assessment, it is important that the following issues be addressed in any future modeling work in the UAE.

• Better initialization data are necessary to more accurately capture and resolve these phenomena. The UAE has a network of automatic weather stations will be very valuable for this purpose. Other data sources could also be included and a four-dimensional data assimilation system should substantially enhance the performance of the model, as has been indicated in other regions of the world.
  
• Better sea surface temperature (SST) data for initialization of the model, especially over the Arabian Gulf, should be obtained. These have a significant impact on the latent heat fluxes (water vapor) and the sea breeze strength, and thus, particularly in summer precipitation, on the timing and location of convective precipitation.
  
• The soil moisture and temperature parameterization and associated physical parameterizations will also have to be improved to better represent sensible and latent heat fluxes over land. Additional surface flux measurements would be helpful.
  

Read more about the MM5 Realtime Forecasting System

View Current MM5 Realtime Forecast

Read about Clark-Hall high-resolution modeling studies

References