Get your maps here!
Climatology of Models and Observations

What these maps show

Factors that could cause model behavior to be different than that shown here

How to use these maps

Click a table entry to see the corresponding map. Data cover the period 1 April 2000 to 1 June 2001. Model changes may have modified model behavior; see below for more. These plots will be updated when a full year of data is available for current versions of the precipitation schemes.

Model

Annual

Jan

Feb

Mar

Apr

May

Jun

Jul

Aug

Sep

Oct

Nov

Dec

Eta
(on 90-km grid)

24h  48h  24heavy  48heavy 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48

Eta
(on 32-km grid)

24h  48h  24heavy  48heavy 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48

AVN days 1-2
(1-deg grid)

24h  48h  24heavy  48heavy 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48

AVN out period
(1-deg grid)

24h  48h  24heavy  48heavy 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48

Obs: CPC
(1/4-degree grid)

24h  48h  24heavy  48heavy 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48

Obs: CPC
(1-degree grid)

24h  48h  24heavy  48heavy 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48 24  48

Obs:
Multisensor – quality problems, improved but "good" data do not go back far enough

                         

Model Update: before/after May 2001 AVN/MRF implementation

October 2000 - May 2001

MRF (old version)

24h  48h  24heavy  48heavy

MRFX (new version)

24h  48h  24heavy  48heavy

Difference

24h  48h  24heavy  48heavy
bigview-24h  bigview-48h  bigview-24heavy  bigview-48heavy

What these maps show

These maps show the number of days per month or per year for which model QPF (quantitative precipitation forecast) or observational QPE (quantitative precipitation estimate) averaged over a postprocessed grid box exceeded threshold values of 0.5",1.0", 1.5", 2.0", 2.5", 3.0", 4.0", 5.0" in a 24-hour or 48-hour period ending 12 UTC for the period from 1 April 2000 to 30 June 2001. The first four thresholds are on four-panel monthly gif images. The last four occur so rarely that only maps of annual frequencies are given ("24heavy" and "48heavy"). The color scheme is the same on all the plots. Note that we expect to see heavier amounts more frequently when the averaging area is smaller, so a fair comparison uses the same size grid boxes.

For model forecasts, the number of days shown is not the same as the number of model cycles. For instance, the Eta has six different model runs with 24-hour QPF valid 12 UTC tomorrow, listed to the right. Thus, if the Eta model is to match an observation of two heavy rain days, it would need to forecast those heavy amounts in twelve different runs. If it only forecast the heavy rain in 3 runs, that would give it a value of 0.5 days in the plot, since that's 3 runs out of a possible 6 per day.
TIME OF MODEL RUN
FORECAST HOUR
VALID TIME
day 1, 00 UTC
60
day 3, 12 UTC
day 1, 12 UTC
48
day 3, 12 UTC
day 1, 18 UTC
42
day 3, 12 UTC
day 2, 00 UTC
36
day 3, 12 UTC
day 2, 06 UTC
30
day 3, 12 UTC
day 2, 12 UTC
24
day 3, 12 UTC

This climatology only measures the frequency of events, not their accuracy. For example, if the model predicts heavy rain just as often as it happens in nature but it always forecasts it on the wrong day, then the model and observed plots will agree because the model has no bias. Basically, these plots allow you to gauge whether a model is predicting a commonplace event or is flagging a potentially unusual event for your geographic area.

The observed precipitation data is primarily over the CONUS; plotted values elsewhere should be considered suspect. Additionally, there are difficulties in estimating observed precipitation. Two different estimates will be provided: the CPC analysis based on gauge data and the multisensor primarily based on radar data. There are some areas of poor data coverage; gauges are known to have collection problems in windy conditions and their point measurements may be unrepresentative of area averages that the models are forecasting, while radar precipitation estimates have many known shortcomings. The difference between these two estimates gives a sense of the uncertainty in estimating "ground truth," though both may incorrectly agree on light amounts where data coverage is sparse. The multisensor data has large amounts of missing data and quality problems during part of the climatology period, but it will be used for comparison when this climatology is shifted to a later period after future model changes.

Factors that could cause model behavior to be different than that shown here:

The AVN/MRF had a major model change 15 May 2001, with an entirely new grid-scale precipitation parameterization and modifications to the convective parameterization. The old version was operational during almost the entire data period. To assess the effect of the model changes, comparison plots are available from the second table. The comparisons cover the time when the current operational versions (except for some minor modifications) were running experimentally in parallel with the then-operational version. The maps show the total number of days that the model predicted rainfall exceeding the thresholds during the roughly 7-month parallel experiment rather than days per month or per year. Overall the new version is similar in its overprediction of heavy amounts over the CONUS but shifts the ITCZ location in the tropics.

A new grid-scale scheme was implemented in the Eta model on 27 November 2001 but the testing sample includes only two months. The convective parameterization is likely to be modified in spring 2002. When enough data is available from these changes, the plots here will be updated or comparison plots made. So far it appears that the Eta model continues to underpredict heavy amounts.

In mid-late summer 2001, the Eta model began making "precip bombs" with over four inches in a 6-hour period, while it seldom produced even two inches in 24 hours during the data period used to make these climatologies. This change appears to have resulted from running at higher resolution and lower numerical diffusion (smoothing) in 2001 than during summer 2000 as well as having a dry regime with little MCS activity in summer 2000. See Spurious Grid-scale Convection in the Eta Model for a case example of one of these events. The November 2001 change in the grid-scale precipitation scheme seems to have reduced the number and intensity of these events. The AVN has a long-standing problem making these unrealistic grid-scale thunderstorms, but that proclivity is already reflected in the plots shown and has not changed since the May 2001 model upgrade.

Credits

By Dr. Stephen Jascourt, UCAR/COMET

Thanks to: