Assessments of Surface Variables from

1
Assessments of Surface Variables from
Atmospheric Reanalysis
Mark C. Serreze Cooperative Institute for
Research in Environmental Sciences University of
Colorado Boulder CO
2
Introduction One of the strongest benefits of
improved NWP in the polar regions is better
depiction of surface variables. Time series of
even basic variables such as precipitation and
temperature tend to be of relatively short
duration with insufficient spatial coverage.
Data for other key variables, such as fluxes of
latent heat, sensible heat and radiation are
even more scanty. Even with biases, surface
fields from improved NWP will find wide use in
climate studies. With removal of biases and
blending with avialable surface observations,
surface fields from NWP can be used be used to
drive other models (e.g., stand alone LSMs,
sea ice and coupled ice/ocean models)
3
Atmospheric Reanalysis Evaluations of
high-latitude NWP output with an eye toward their
wider use has naturally focused on output from
atmospheric reanalysis. Reanalysis provides long
time series with frozen data assimilation/foreca
st systems NCEP/NCAR (NCEP-1)
1948-present NCEP-DOE AMIP-II (NCEP-2)
1979-present ERA-15 1979-1993 ERA-40
1957-2002 (will be updated?) NCEP NARR
4
There are many examples of the use of surface
variables from reanalysis Clark et al. 1999
Relationships between Eurasian snow cover
variability and 2-m temperature fields from
NCEP-1. Walsh 2000 Circulation relationships
associated with high/low precipitation, using
NCEP-1 precipitation fields Serreze et al.
2003 Blending NCEP-1 precipitation
forecasts with station data. A general
conclusion from evaluations There is useful
skill in reanalysis depictions of many surface
variables, such as precipitation and temperature.
But there are often large biases, and
sometimes, gross misrepresentations.
5
Some Results from Past Evaluations Serreze et
al. 1998 NCEP-1 monthly averaged fluxes of
downwelling solar radiation in the Arctic are
much to high (up to (80 Watts per square meter
in June), apparently due to problems in modeled
cloud cover. Serreze and Hurst 2000 Summer
precipitation rates in NCEP-1 are much too high
over Arctic land areas, mostly from excessive
convective precipitation and evaporation. The
extent to which these problems are conflated
with the excessive shortwave radiation,
the convective scheme and updating soil moisture
by the forecasted precipitation is unclear.
NCEP-2 suffers from the same summer precipitation
biases Serreze et al., 2003. NCEP-1 also
shows a blotchy pattern of precipitation
associated with orography, apparently due to a
problem in horizontal moisture diffusion. The
problem has been addressed in NCEP-2.
6
P-E Aerological estimates of precipitation minus
evaporation (P-E, computed from wind and moisture
profiles) from both NCEP-1 and ERA-15 look
pretty good Cullather et al., 2000 Rogers et
al., 2001. This is a strong point of
reanalysis. The plots at right show mean
January and July P-E (mm) from NCEP-1.
However, P-E computed from the aerological
method and from the surface forecasts of P-E are
not in hydrologic balance. Both models have lower
P-E in the forecasts (but with both P and E being
much too high in NCEP-1).
7
Overall, ERA-15 seems to perform better than
NCEP-1, and the extent to which it has been
examined, NCEP-2. But ERA-15 has a cold bias in
2-m temperatures over boreal forest Viterbo and
Betts, 1999 Walsh and Chapman 1998 looked at
radiation/surface temperature/cloud cover
associations over the Arctic Ocean NCEP-1
slightly exaggerates observed associations
observed associations between cloudiness and
surface air temperature while ERA-15 shows a much
weaker association. Oddly, surface radiation
fluxes in ERA-15 seem to be insensitive to the
presence of clouds.
8
But the Outlook is Encouraging First ERA-40
data are now on line. ERA-40 has
many improvements relative to ERA-15, NCEP-1 and
NCEP-2, many of which are specific to high
latitudes. Second The SEARCH program is
planning for an Arctic System Reanalysis (ASR).
Results from this workshop will feed into
development of the ASR. One activity that can
help is setting performance benchmarks that
should be exceeded, defined through evaluation
of the present state-of-the-art (ERA-40 and
NARR). Here, we focus on ERA-40, with an
emphasis on precipitation a key variable for
which reanalysis could be of great use.
Comparions are made with estimates from NCEP-1,
ERA-15 and satellite retrievals (GPCP).
9
The Issue Understanding the hydrology of the
Arctic regions and monitoring emerging changes
requires (among other things) gridded field of
precipitation of sufficient quality for climate
analysis and for driving hydrologic and land
surface models.
The Problem While the station data base has
always been sparse (see figure at left), it has
seriously degraded since about 1990. For
example, station coverage over Siberia is is now
about half of what it was in the 1980s. The
network is now insufficient to compile gridded
fields. We need alternative strategies.
Blending precipitation forecasts from NWP with
available gauge data may be the best approach.
10
Basic Evaluation Strategy 1) Grid the available
observations to a 100 km grid array down to 45
deg. N (a version of the NSIDC EASE Grid).
2) Translate data from NCEP-1, ERA-15, ERA-40
and GPCP to the same grid. The reanalysis data
and GPCP are provided on a 2.5 deg. lat/long
grid. Full resolution ERA-40 data are
forthcoming. 3) Compare these four estimates to
the gridded observations in terms of mean
annual cycles and temporal correlations. Focus
on the period 1979-1989, for which there is a
common period of overlap. The focus here is
on MONTHLY precipitation Major Caveats The low
station density strongly impacts on the quality
of the gridded observed time series. There is
also the problem of gauge biases. We adjust
using the Legates 1987 coefficients.
11
Mean Annual Cycles for the Major Eurasian
Watersheds Observations show July maxima
and cold season minima. This basic pattern is
reproduced by each model, but with warm-season
precipitation too high in NCEP-1. ERA-40 and
ERA-15 look quite good, and closely track each
other.
12
ERA-40 Mean Precipitation The spatial patterns
of mean monthly precipitation from ERA-40 are in
accord with available observations.
Precipitation is greatest in the north
Atlantic and Pacific, best expressed in winter.
Land areas and the Arctic Ocean exhibit winter
minima and summer maxima.Winter month totals in
parts of Siberia and northern Canada are less
than 15 mm.
13
Temporal Correlations (1979-1989) The figures at
left show the percentage of grid cells (y-axis)
in the Arctic terrestrial drainage for which the
squared correlation with observed precipitation
time series is less than the value indicated on
the x-axis. ERA-40 shows greatly improved
performance relative to NCEP-1, especially in
winter. However, performance is no better than
ERA-15. All of the reanalyses beat the
perfomance of the GPCP satellite-derived
precipitation estimates.
14
Squared Correlations, Observed vs. ERA-40
Precipitation There are large areas,
especially over Eurasia, where ERA-40 performs
quite well (squared correlations exceed 0.50).
But for many areas, perfomance appears to be
poor. But in data-sparse areas, the observed
gridded time series are of poor quality.
ERA-40 may well be perfoming better than is
indicated.
15
Squared Correlations Observed vs GPCP The
better performance of ERA-40 (and
ERA-15) relative to the GPCP satellite product is
abundantly clear. Satellite retrievals have a
hard time dealing with the heterogeneous
emissivity of land surfaces, and work best
over open ocean regions. Over land, reanalysis is
the way to go. The satellite retrievals are
considered to be improved after 1987 (the SSM/I
era).
16
For the central Arctic Ocean, spotty coverage is
provided by data from the Russian North Pole
program, which ended in 1991. The best we can
really do is compare monthly observations from
the drifting ice camps with reanalysis data at
the closest grid point. In winter, none of the
reanalyses seem to perform well. Agreement
inproves in summer, with squared correlations of
0.57 (ERA-40), 0.56 (ERA-15) and 0.24 (NCEP-1).
This may reflect the stronger precipitation
signals in summer.
17
Parting Comments Compared to middle latitudes,
the polar regions have received relatively little
attention by the NWP community. But the
recent change in this attitude, exemplified by
the Fairbanks workshop, efforts by the ERA-40
group and momentum towards an ASR, can have many
benefits. One of these is better surface fields.
Compared to NCEP-1, ERA-40 represents a
strong step forward. We are already making use
of the ERA-40 precipitation forecasts to compile
time series through de-biasing and blending with
observations. But ERA-40 seems to be little
better than ERA-15. We have a long way to go.