Data Information Page from ArcticRIMS (http://RIMS.unh.edu)
Title:
DAILY PRECIPITATION FROM STATISTICAL RECONSTRUCTIONS (Serreze)
Description:
Providing rapid updates of gridded precipitation fields is
difficult due to the degradation of the station network.
Starting about 1990, many stations in the Former Soviet Union
(FSU) and Canada closed. Canada is also seeing a trend toward
automation. In recognition, we are providing updated fields
using statistical techniques. This reconstruction technique has
also been used to compile fields back to 1980 to provide a period
of overlap with gridded fields based on interpolation of monthly
station precipitation and daily disaggregation (see DAILY
PRECIPITATION, FROM MONTHLY STATION RECORDS AND DISAGGREGATION).
The reconstructions employ a four-step approach. As part of the
data processing of ArcticRIMS (http://RIMS.unh.edu) based at the
Water Systems Analysis Group, University of New Hampshire all data
sets have been aggregated to multiple temporal and spatial resolutions.
Classification:
Meteorology, Precipitation, Climate
Author/PI:
Vorosmarty, Charles, Richard Lammers and Mark Serreze
Contact Information for original gridded daily time step data:
Mark Serreze
Senior Research Scientist
449 UCB, RL-2, #223
National Snow and Ice Data Center
University of Colorado Boulder, CO 80309-0449
E-mail: serreze@kryos.colorado.edu
Tel: 303-492-2963
Web: http://nsidc.org/research/bios/serreze.html
Contact Information for all spatially and temporally aggregated data in RIMS:
Charles Vorosmarty
Department of Civil Engineering
The City College of New York
Steinman Hall, Rm T-513
140th Street & Convent Ave, NY NY 10031 USA
Email: cvorosmarty@ccny.cuny.edu
Tel: (212) 650-7042
Web: http://crest.ccny.cuny.edu/
Richard Lammers
Water Systems Analysis Group
Institute for the Study of Earth, Oceans, and Space
Morse Hall, Room 211
8 College Road
University of New Hampshire
Durham, NH 03824-3525 USA
Email: Richard.Lammers@unh.edu
Tel: (603) 862-4699
Web: http://www.wsag.unh.edu/
Temporal Coverage
Begin Date (year-month-day):
1979-01-01
End Date (year-month-day):
2001-12-31
Spatial Coverage:
Corner coordinates in Ease Projection (Units: Meters form N.P.)
(Description at http://nsidc.org/data/ease/ease_grid.html)
Minimum X:
-4875633.612 m
Minimum Y:
-4875633.612 m
Maximum X:
4875633.612 m
Maximum Y:
4875633.612 m
Corner coordinates in Geographical projection (Units: Degrees)
(Description at http://en.wikipedia.org/wiki/Equirectangular_projection)
Minimum latitude:
45.0
Minimum longitude:
-180.0
Maximum latitude:
90.0
Maximum longitude:
180.0
Units:
mm
Aggregation Method:
General Methods:
STEP 1 involves interpolating monthly precipitation totals
from the NCEP/NCAR reanalysis [Kalnay et al., 1996] to the EASE
grid and then re-scaling these forecasts via a probability
transformation. The re-scaling procedure uses ranked (i.e.,
sorted) values of NCEP/NCAR and observed precipitation at each
grid cell for 1960-1989. Observed monthly precipitation is
based on interpolation of bias-adjusted station data using the
same techniques described for data set "DAILY PRECIPITATION, FROM
MONTHLY STATION RECORDS AND DISAGGREGATION.
The ranks are ascribed cumulative probabilities. Imagine
that an update of NCEP/NCAR precipitation is obtained for June
2004. We determine where the June 2004 NCEP/NCAR value falls in
the 30-year (1960-1989) NCEP/NCAR cumulative probability
distribution. The June 2004 NCEP/NCAR value is re-scaled by
simply replacing it with the observed precipitation value at the
same cumulative probability. Generally, interpolation is
necessary because the NCEP/NCAR value to be re-scaled lies
between two of the ranked NCEP/NCAR values in the sample (1960-
1989) distribution. If the June 2004 NCEP/NCAR value is smaller
than the smallest value in the NCEP/NCAR rankings, it is
ascribed the smallest value in the observed 30-year distribution.
If the June 2004 NCEP/NCAR value is greater than the largest
value in the NCEP/NCAR rankings, it is replaced with the largest
observed value in the 30-year distribution. The re-scaling
assures that any resulting reconstructed time series has same
mean and standard deviation as the corresponding observed time
series.
In STEP 2, we compile a separate set of monthly
reconstructions using a suite of predictor variables from the
NCEP/NCAR reanalysis in a multiple linear regression scheme. The
predictors include 1) forecasts of precipitation; 2) vertical
motion at 500 hPa; 3) precipitation minus evaporation (P-ET) (see
PRECIPITATION MINUS EVAPORATION); 4) zonal and meridional vapor
fluxes; 5) an index of lower-tropospheric stability; 6) sea level
pressure; 7) precipitable water. Regression models for each
month and EASE grid were developed using data over the period
1960-1989. A forward-screening approach was used for variable
selection. Generally, the best and most frequently used
predictors are precipitation, P-E and vertical velocity. The
model slopes and intercepts are then applied to updates of
NCEP/NCAR data. To eliminate systematic biases in the
reconstructions, we again use a re-scaling approach. In this
case, the re-scaling uses the distributions of observed
precipitation (1960-1989) and of regressed precipitation from the
same period.
Correlations between time series of observed and regressed
(STEP 2) precipitation are generally (but not always) higher
than that those between observed and re-scaled (STEP 1)
precipitation. In STEP 3, a decision-tree approach is adopted to
merge results from the re-scaling and regression, based on 1)
relative predictive skill; 2) interpolation biases; 3) mean
observed precipitation.
In STEP 4, the monthly reconstructions are disaggregated
into daily values, using the same basic technique outlined for
data set "DAILY PRECIPITATION, FROM MONTHLY STATION RECORDS AND
DISAGGREGATION"
Comments:
Linear regression assumes that the observed precipitation
time series at a given grid cell represents "truth". Most of the
Arctic is characterized by a low station density. As a result,
time series for a given grid based on interpolation often poorly
reflect the "true" time series structure for the cell. Hence, one
may often be regressing against noise. The problem is compounded
in areas of very low precipitation, where even small measurement
errors can greatly impact on the interpolated time series. By
contrast, the statistical distributions (mean and variance) of
precipitation tend to be reasonably well preserved [Serreze et
al., 2003]. Our reconstructed precipitation product recognizes
these problems. In the decision tree algorithm (STEP 4), the re-
scaled values are used instead of the regression-based values in
areas of low mean precipitation and where the station network is
especially sparse (where interpolation errors are large). In
these areas, the re-scaling is on better statistical footing as
it relies only on the statistical distributions.
A new data set is under development. Details are provided
by Serreze et al. [2003]. It is based on: 1) re-scaling the
NCEP/NCAR precipitation forecasts; 2) re-scaling P-E; 3)
performing tests at each grid cell to determine which re-scaled
variable provides the higher skill, and basing the reconstruction
on the better of the two values, provided that it beats
climatology; 4) further improving the reconstructions through
assimilating any available updates of observed precipitation.
The key differences, other than the station data
assimilation, are: a) elimination of the multiple linear
regression; b) provision of the reconstructions at a coarser
spatial resolution (but "nestable" within the 25 km EASE grid)
employing a "tighter" interpolation of the station data. This
will provide a more realistic assessment of precipitation
variability at the chosen grid size.
References:
Kalnay, E., M. Kanamitsu, R. Kistler, W. Collins, D. Deaven, L.
Gandin, M. Iredell, S. Saha, G. White, J. Woolen, Y. Zhu, M.
Chelliah, W. Ebisuzaki, W. Higgens, J. Janowiak, K.C. Mo, C.
Ropelewski, J. Wang, A. Leetma, R. Reynolds, R. Jenne, and D.
Joseph, 1996: The NCEP/NCAR 40-year reanalysis project. Bull.
Amer. Meteorol. Soc., 77, 437-471.
Serreze, M.C., M.P. Clark and D.H. Bromwich, 2003: Monitoring
precipitation over the terrestrial Arctic drainage system: Data
requirements, shortcomings and applications of atmospheric
reanalysis. J. Hydrometeorology (in press).
Arctic RIMS Contact:
Richard Lammers
Water Systems Analysis Group
Institute for the Study of Earth, Oceans, and Space
Morse Hall
University of New Hampshire
Durham, NH 03824
Phone: (603) 862-4699
Fax: (603) 862-0587
Email: Richard.Lammers@unh.edu
Web: http://wsag.unh.edu
Data Archiving:
This ArcticRIMS data set has been permanently stored to the ARCSS Data Archive at NCAR/EOL (http://www.eol.ucar.edu/projects/arcss) with the support of National Science Foundation grants (NSF) OPP-0230243 and
Humans and Hydrology at High Latitudes (NSF) ARC-0531354