Data times and dates
The data cover the period from 1 April 2000 to 30 June 2001. Thus the sample size of spring months is two months and the sample size of others is just one month. The plots will be updated only after collecting data over a substantial period following the last time in the dataset.
No attempt was made to remove available data sources for days when some source(s) were not available. Thus the periods are not exactly identical for all sources, though they are very close. The number of events in each month for each data source is scaled to a complete 30-day month, making all months equal regardless of whether two such months were sampled (April/May/June) or whether any data are missing. Data from forecast cycles at the end of the month with valid periods during the beginning of the next month are recorded in the later month to match against the observed climatology. The annual total is a sum of these normalized monthly totals.
The model cycles used are:
Model cycle |
24 h QPF valid at |
48 h QPF valid at |
00 UTC cycle
|
|
|
06 UTC cycle
|
|
|
12 UTC cycle |
|
|
18 UTC cycle |
|
|
These valid forecast periods were selected because they produce 24-hour and 48-hour amounts valid at 12 UTC, which can be verified against the gauge network.
The MRF comparisons between the old and new precipitation parameterizations used forecasts from the MRFY from 10/1/2000–10/23/2000, 11/13/2000–2/13/2001, and from the MRFX from 2/14/2001–5/15/2001. These dates correspond to the appropriate parallel experiments. Operational MRF data was used for the same model cycles. Precipitation amounts valid at 36, 60, 84, 108, 132, and 156 hours were all used.
Grids/interpolation and notes for each source
The model precipitation forecasts are on standard postprocessed model grids. NCEP precipitation postprocessing is done through a box overlapping method, preserving the area-integrated precipitation amounts.
Observational estimates of precipitation from the RFC gauge network were obtained on a 1/4-degree grid and 16 of these "squares" (4x4) were averaged to give the precipitation estimates on a 1-degree grid. The multisensor data begin on a roughly 5-km grid and are converted to a 15-km grid using the same area preservation technique used with model data. The 90-km data are simply averages over 6x6 squares of the 15-km data.
The plots have no further interpolation or smoothing, and do not represent colored contour fills. Values shown are the frequency of occurrence of precipitation on these grids exceeding the threshold amounts in each grid box. The entire grid box is shaded one color corresponding to the event frequency.
The last Eta model change to the convective parameterization was in March 2000. Check the COMET source for a concise history of Eta model changes. The Eta model was run at 32 km until late September 2000 and at 22 km thereafter. The QPF output is from the 32-km grid #221 during that entire time as well as postprocessed onto the 90-km grid #104. The 90-km grid provides a comparison point against the AVN since the precipitation amounts represent averages over grid boxes covering a comparable area.
The AVN had a major change to both its convective parameterization and grid-scale precipitation parameterization in mid-May 2001. Thus most of the data are from the older version of the model. Output is on a 1-degree grid, which is standard for viewing AVN and MRF output. The MRF-MRFX comparison is also on the same 1-degree grid.
The CPC analysis, produced by the NCEP Climate Prediction Center, is a gridded field of quality-controlled RFC data, from the gauge network. A Cressman analysis is used to interpolate the gauge data to a 1/4-degree grid. For more information, see the Climate Prediction Center's U.S. Precipitation Quality Control System and Analysis Web page (scroll to the bottom).
The Multisensor analysis is a Stage III national mosaic of RFC quality-controlled data. These RFC data are mostly Stage II bias-adjusted NEXRAD precipitation estimates. Gauge data are used to locally adjust the radar estimates through a local bias estimate and through an optimal interpolation. Details of this procedure are given in Briendenbach, et al. (1998). However, the western RFCs use mountain mapper to adjust the gauge data instead of using radar estimates. More information on these datasets can be found on the National Stage II Analyses homepage and FAQ. This product is known to typically indicate smaller amounts than gauge-only analyses and to grossly underestimate precipitation falling as snow. However, the spatial distributions are better in non-mountainous areas than the gauge analysis and coverage is somewhat more uniform than the gauge network. These data are not reliable until at least January 2001.