Precipitation Dataset for Statistical Post Processing and Downscaling

1
Precipitation Dataset for Statistical Post
Processing and Downscaling

• 29 November 2007

2
Goals

• Create a historical precipitation dataset
• Needs to be quality controlled
• Needs to have a high spatial resolution
• Use this climatic precipitation dataset to
  bias-correct GEFS and NAEFS precipitation
  forecasts (1 resolution)
• Use this dataset to downscale precipitation
  forecasts to the high resolution NDFD grid (5 km
  resolution)
• Compute climate statistics for precipitation

3
Dataset
U.S. Domain (downscaling and bias correction)

• Combine information from two datasets
• CPC precipitation analysis
• Specifications
• Daily
• Since 2002
• 1/8 resolution
• 6-7 thousand daily reports
• Methods used to generate
• Uses inverse distance weighting for spatial
  interpolation of gages
• PRISM climatology for incorporating orographic
• Quality controlled (radar, satellite, station
  checks)
• RFC QPE
• Specifications
• Every 6 hours
• Since 2000
• 4-5 km resolution
• Methods used to generate
• Combines rain gage data w/ radar data (Mountain
  Mapper in West)
• Quality controlled by individual RFCs, then
  mosaicked.

4
Dataset
North American Domain (NAEFS)

• Combine information from multiple sources
• CMORPH
• CPC ¼ Precipitation Analysis
• Same as 1/8 dataset, except uses modified
  Cressman scheme to interpolate to grid
• Canadian Precipitation Analysis (CaPA)
• Specifications
• 6-hourly
• 15 km grid
• Covers Canada and the U.S.
• Methods
• 6 hour forecast precip (GEM model) used as
  background
• Adjusted with rain gauges and radar data
• Being developed

5
Dataset
Global Domain (bias correction)

• Combine information from multiple sources
• CMORPH
• Specifications
• Every 30 minutes
• Since late 2002
• 8 km resolution
• Methods used to generate
• Precipitation estimates derived from low orbiter
  satellite microwave observations
• Precipitation features transported using
  satellite infrared data, when microwave
  observations are not available
• Issues
• Poor estimates of precipitation over ice/snow
  cover
• Only extends to 60 N/S
• Will be replaced by RMORPH
• Early next year (March?)
• Incorporate rain gage data
• Back to 2000 (maybe 1998?)

6
Dataset
Global Domain (bias correction) (continued)

• GFS
• Will fill in data missing from CMORPH dataset (gt
  60 N/S)
• Use 12 hour forecast of 6-hour precipitation
  (between 6 and 12 hours after initialization).
• Issues
• Lower resolution
• Not observed precipitation
• Blending global and regional domains
• Precipitation values in U.S. domain may differ
  between CMORPH and RFC QPE-CPC merged dataset
• Should be better with RMORPH
• For now, scale CMORPH toward higher resolution
  RFC QPE-CPC data, and incorporate Mexico/Canada
  CPC rain gage data to help scale CMORPH

7
Comparison
8
Combining Datasets

• Methods similar to work by Ken Mitchell and
  Youlong Xia
• Temporally disaggregated the 1/8 CPC
  precipitation analyses using Stage IV
  precipitation (http//www.emc.ncep.noaa.gov/mmb/yl
  in/pcpanl/stage4/).
• Maintained 24 hour totals from CPC precipitation,
  but added time variability from Stage IV
  precipitation.
• Original Resolutions
• CPC dataset 15 km resolution
• RFC dataset 5 km resolution
• NDFD 5 km resolution
• Interpolate grids to make combining simple
• RFC data ? 5 km NDFD grid
• CPC data ? 15 km grid that fits exactly over the
  5 km NDFD grid

9
New Method First Interpolate Grids
CPC grid
RFC QPE grid
CPC grid fits exactly over RFC QPE grid (which
matches NDFD grid)
10
Scaling RFC QPE Against CPC

• Step 1
• CPC daily precip defined as precip from 12 UTC
  12 UTC
• Sum 6-hourly RFC QPE over same period to obtain
  daily total

12-18 UTC
18-00 UTC
00-06 UTC
06-12 UTC
12-12 UTC
12-12 UTC
12-12 UTC
11
Scaling RFC QPE Against CPC

• Step 2
• Obtain average of all 9 points of RFC QPE grid
  within each CPC grid box
• Obtain scaling factor (1/2 in this example)

12-12 UTC
12-12 UTC
Average 20
Scaling factor CPCval / RFCtot (1/2 in this
example)
12
Scaling RFC QPE Against CPC

• Step 3
• Scale each 6-hourly RFC QPE grid by the scaling
  factor obtained in step 2

12-18 UTC
18-00 UTC
00-06 UTC
06-12 UTC
½
12-18 UTC
18-00 UTC
00-06 UTC
06-12 UTC
13
End Result

• 5-year precipitation dataset
• 5 km resolution
• Confident in values (CPC truth)
• Already on NDFD grid
• Already incorporates quality controls

14
Missing Data

• Obtain average of all N points of RFC QPE grid
  within each CPC grid box, where N is the number
  of gridpoints with data
• Obtain scaling factor (1/2 in this example)

12-12 UTC
12-12 UTC

• Average SumAllGoodPoints / N 140/7 20
• Scaling factor CPCval / RFCtot (1/2 in this
  example)
• Perform step 3 as usual
• For data missing across time dimension (eg. no
  data at 18 UTC), fill gridpoint with time-average
  of remaining points