388
JOURNAL OF HYDROMETEOROLOGY
VOLUME 9
Evaluation of Precipitation Products for Global Hydrological Prediction
NATHALIE VOISIN, ANDREW W. WOOD,
AND
DENNIS P. LETTENMAIER
Department of Civil and Environmental Engineering, University of Washington, Seattle, Washington
(Manuscript received 15 June 2007, in final form 7 September 2007)
ABSTRACT
Accurate precipitation data are critical for hydrologic prediction, yet outside the developed world in situ
networks are so sparse as to make alternative methods of precipitation estimation essential. Several such
alternative precipitation products that would be adequate to drive hydrologic prediction models at regional
and global scales are evaluated. As a benchmark, a gridded station-based dataset is used, which is compared
with the global 40-yr ECMWF Re-Analysis (ERA-40), and a satellite-based dataset [i.e., the Global Precipitation Climatology Project One-Degree Daily (GPCP 1DD)]. Each dataset, with a common set of other
meteorological forcings aside from precipitation, was used to force the Variable Infiltration Capacity (VIC)
macroscale hydrology model globally for the 1997–99 period for which the three datasets overlapped. The
three precipitation datasets and simulated hydrological variables (i.e., soil moisture, runoff, evapotranspiration, and snow water equivalent) are compared in terms of the implied water balances of the continents,
and for prediction of streamflow for nine large river basins. The evaluations are in general agreement with
previous but more local evaluations of precipitation products and water balances: the precipitation datasets
agree reasonably on the seasonality but less on monthly anomalies. Furthermore, the largest differences in
precipitation are in mountainous regions and regions where in situ networks are sparse (such as Africa).
Derived runoff is highly sensitive to differences in precipitation forcings. At a global level, all three
simulations result in water budgets that are within the range of other water balance climatologies. Although
uncertainties in the three datasets preclude an evaluation of which one has the lowest errors, overall
ERA-40 is preferred because of its agreement with the station-based dataset in locations where the station
density is high, its periodic availability, and its temporal resolution.
1. Introduction
Early applications of hydrologic modeling focused
mostly on small spatial scales (generally thousands of
kilometers squared and less). Even when applied over
larger areas, hydrologic models have usually been
implemented on a basin by basin (or what might be
termed “bottom up”) manner. The types of problems,
and spatial scales, to which hydrological models are
now applied, however, have expanded considerably as
questions of how streamflow and other hydrologic variables will respond to a number of change agents (climate and land cover/land use being the most obvious)
have come to the fore (Maurer 2007; Costa-Cabral et al.
2008; Tang et al. 2006). Furthermore, as the skill of
weather and climate forecast models and methods have
Corresponding author address: D. P. Lettenmaier, Department
of Civil and Environmental Engineering, University of Washington, P.O. Box 352700, Seattle, WA 98195.
E-mail: dennisl@u.washington.edu
DOI: 10.1175/2007JHM938.1
© 2008 American Meteorological Society
improved, a demand has evolved for linking such models, which have relatively large spatial scales, with hydrologic models. As a result, macroscale hydrology
models, designed for regional, continental, and even
global scales have evolved. Among these are the Variable Infiltration Capacity (VIC) model of Liang et al.
(1994), and the University of Waterloo hydrologic
model (WATflood; Snelgrove et al. 2005).
As the spatial scales of interest for the application of
hydrologic models have increased, so has the need to
explore alternative sources for the primary hydrologic
forcing variable (i.e., precipitation). Although gridded
station data [e.g., the continental U.S. dataset of Maurer et al. (2002) and the global dataset of Adam et al.
(2006)] remain the primary source of precipitation data
for large-scale hydrologic prediction, other sources
have a number of advantages, particularly in regions
where in situ observations are sparse. These alternative
sources include precipitation products derived from satellite remote sensing, and analysis and reanalysis fields
from global and regional numerical weather prediction
JUNE 2008
VOISIN ET AL.
models such as the National Centers for Environmental
Prediction–National Center for Atmospheric Research
(NCEP–NCAR) reanalysis (Kalnay et al. 1996), the 40yr European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-40; Uppala et al.
2005), and the North American Regional Reanalysis
(NARR; Mesinger et al. 2006).
The objective of this paper is to evaluate alternative
precipitation products suitable to drive hydrologic prediction models at large scales, particularly for parts of
the globe where in situ networks are sparse. We evaluate three alternative sources of hydrological model precipitation forcings: a gridded station dataset, a satellite
observation–based precipitation dataset and a numerical weather prediction model analysis precipitation
field. We also evaluate the differences in the simulated
water balances resulting from use of these three
datasets. The hydrologic evaluations are performed
over a range of accumulation times and at the spatial
resolution of hydrologic predictions as simulated by the
VIC macroscale hydrology model over large river basins.
Candidate gridded global datasets based on observations included Adam and Lettenmaier (2003), Adam et
al. (2006), Chen et al. (2002), Willmott and Matsuura
(2001), the Climatic Research Unit time series dataset
(CRU TS 2.1; Mitchell and Jones 2005), and the Global
Precipitation Climatology Project (GPCP) version 1
(Huffman et al. 1997). Most are monthly time series.
Adam et al. (2006) was chosen because of its daily time
step, spatial resolution, period of record, adjustment for
gauge undercatch bias and orography, and the availability of other forcing variables such as temperature
and wind required by the hydrologic model VIC.
Datasets from the second of these three classes include the Global Precipitation Climatology Project
One-Degree Daily (GPCP 1DD) dataset of Huffman et
al. (2001) for the period 1997–2006. These data are
based on multiple passive microwave (PMW) and infrared (IR) satellite observations. The monthly total
GPCP 1DD precipitation matches the GPCP version 2
monthly values (Adler et al. 2003). Other satellite precipitation datasets that we considered, which are based
on PMW as well as IR and satellite radar [in the case of
the Tropical Rainfall Measuring Mission (TRMM)], include the Climate Prediction Center Morphing technique (CMORPH; Joyce et al. 2004), the Precipitation
Estimation from Remotely Sensed Information using
Artificial Neural Networks (PERSIANN; Sorooshian
et al. 2000), the Climate Prediction Center Merged
Analysis of Precipitation (CMAP; Xie and Arkin 1997),
and the TRMM 3B42RT (Huffman et al. 2003). As
noted below, our choice of GPCP 1DD is based pri-
389
marily on its long (relative to other satellite precipitation datasets) period of record, its global coverage, and
its widespread use.
Among numerical weather prediction analysis fields,
we assess the ECMWF ERA-40 reanalysis (Uppala et
al. 2005), which, like GPCP 1DD, is global, but is available for the much longer period 1957–2002 and therefore provides complete overlap with the 1979–99 gridded station data of Adam et al. (2006). The ERA-40
reanalysis was chosen over the NCEP–NCAR reanalysis in view of the more numerous forecast products
available (e.g., the ensemble prediction system and
monthly and seasonal forecasts) and more recent developments (hence higher resolution among other considerations) in view of later applications.
ERA-40 and GPCP versions 1 and 2 have already
been compared at a basin scale or regionally with each
other (Serreze et al. 2005; Betts et al. 2003a; Troccoli
and Kahlberg 2004) or with other gauge-based datasets
(Hagemann et al. 2005; Betts et al. 2003b, 2005; Brock
et al. 1995). In this paper however, we compare the
three precipitation data products over the entire global
land area, giving a more comprehensive assessment of
their differences at large scales. We then use each of
these datasets (described in more detail in the following
section) to force the VIC hydrological model, and compare the simulated hydrological variables (e.g., soil
moisture, snow water equivalent, evapotranspiration,
and runoff) in terms of the predicted streamflow for
nine large river basins, and the implied water balances
of the continents. Although we recognize that inevitably results will vary regionally and that there are different sources of uncertainty inherent in each of the
datasets, our intention is to provide a basis for determination of the appropriateness of each of the datasets
for flood prediction in large river basins, especially
those with sparse in situ data.
The first section describes the three classes of precipitation datasets used in our analysis. The following
section summarizes the hydrology model and the experimental design. The last section presents the differences in the precipitation fields and the induced differences in the derived simulated hydrology variables for
nine large river basins and then for each continent.
2. Datasets
As noted above, the three precipitation datasets we
evaluate are 1) the satellite-based GPCP 1DD data, 2)
the ERA-40 reanalysis, and 3) the observation-based
gridded dataset of Adam et al. (2006). Each of these
datasets is described briefly below.
390
JOURNAL OF HYDROMETEOROLOGY
a. GPCP 1DD
The GPCP 1DD data were provided by the National
Aeronautics and Space Administration (NASA) Goddard Space Flight Center’s (GSFC) Laboratory for Atmospheres, which develops and computes the GPCP
1DD as a contribution to the Global Energy and Water
Experiment (GEWEX) Global Precipitation Climatology Project. The GPCP 1DD combines IR techniques
(e.g., the Threshold-Matched Precipitation Index; Huffman et al. 2001) in the 40°S–40°N latitude band and the
rescaled Television and Infrared Observation Satellite
(TIROS) Operational Vertical Sounder (TOVS) precipitation dataset of Susskind et al. (1997) outside the
40°S–40°N latitude band. The Goddard Profiling Algorithm (GPROF) version 6.0 (Kummerow et al. 1996;
Olson et al. 1999), which relates the observed passive
microwave brightness temperature to hydrometeor
profile characteristics and then to daily surface precipitation rate, is used to derive the fractional daily occurrence of precipitation. The monthly accumulations
come from the spatial interpolation of the monthly 2.5°
values of the GPCP version 2 (Adler et al. 2003), which
is based on satellite and gauge observations. GPCP
1DD is presently the only satellite-based precipitation
product available at a daily time scale that covers the
entire globe and has substantial temporal overlap with
the reanalysis and gridded station dataset of Adam et
al. (2006). In the applications reported here, GPCP
1DD was interpolated from 1° to 0.5° [i.e., the synergraphic mapping system (SYMAP) algorithm; Shepard
1984] and missing values (modest in number) were interpolated from nearby grid cells or when this was not
possible, persisted from the day before.
VOLUME 9
TOVS data began to be assimilated (Hernandez et al.
2004). Therefore, for this analysis we only considered
the post-1978 period of ERA-40 data. Coincidentally,
this period overlaps the entire 1979–99 period of the
Adam et al. (2006) gridded station data. The ERA-40
precipitation field was interpolated from its native N80
Gaussian grid to a 0.5° latitude–longitude grid over the
global land areas using the SYMAP algorithm (Shepard
1984).
c. Gridded station data
Adam et al.’s (2006) precipitation dataset (hereafter
A2006) is based on the global gridded station data of
Willmott and Matsuura (2001), corrected for gauge undercatch using the methods described in Adam and
Lettenmaier (2003) and for orographic effects in topographically complex regions using methods described in
Adam et al. (2006). The Willmott and Matsuura (2001)
monthly precipitation was disaggregated to daily using
available daily gauges statistics as explained in Adam
and Lettenmaier (2003). Its spatial resolution is 0.5°
and the dataset spans the period 1979–99. A2006 takes
into account the effects of both systematic instrumentrelated bias and bias related to station location in orographically complex regions. We note that there remain
questions as to the effects of the corrections, and some
evidence suggests (e.g., over North America, as explained later) that A2006 precipitation estimates may
be biased upward, at least in some regions. Nonetheless, we use A2006 as the baseline in our analyses, and
as a matter of convention, we show all results as differences relative to A2006.
d. Previous comparisons and evaluations
b. ERA-40
The ERA-40 reanalysis (Uppala et al. 2005) data,
obtained from NCAR, include air temperature, surface
wind, and 6-h cumulative precipitation 4 times daily for
the period from September 1957 to August 2002 at a
spatial resolution of about 1.125° latitude–longitude
(N80 Gaussian grid). The ERA-40 reanalysis assimilates data from numerous sources, including satellite
and other observations, mostly of so-called free atmosphere variables such as humidity, temperature, and
wind. Relatively few surface observations are assimilated, and in any event, precipitation observations are
not assimilated, so the ERA-40 precipitation fields can
be considered to be functionally independent of the
satellite and gridded station datasets. Clearly, observation technologies have improved over the 1958–2002
period; however the greatest change is attributable to
the “satellite divide,” which occurred in late 1978 when
ERA-40 and GPCP version 1 (Huffman et al. 1997)
precipitation fields have previously been compared
globally by Hagemann et al. (2005) and over the Amazon basin by Betts et al. (2003a), the Mississippi basin
by Betts et al. (2005), the Mackenzie basin by Betts et
al. (2003b), and over the pan-Arctic region by Serreze
et al. (2005). Hagemann et al. (2005) found that ERA40 precipitation was generally higher than GPCP version 1 in the intertropical convergence zone (ITCZ).
Troccoli and Kallberg (2004) derived correction factors
in an attempt to reduce the apparent high bias of ERA40 precipitation (relative to GPCP) in the 30°S–30°N
latitude band over the ocean only, by assuming that the
evaporation is error free and by closing the water budget based on this assumption. Despite the correction
over the oceans, they found that ERA-40 precipitation
was generally higher than GPCP within 10° of the equator and in particular near Africa. A similar analysis by
JUNE 2008
VOISIN ET AL.
Janowiak et al. (1998) showed that the NCEP–NCAR
reanalysis (Kalnay et al. 1996) had larger monthly precipitation over the ITCZ land areas than GPCP version
1. Betts et al. (2005) compared ERA-40 1979–2001
monthly precipitation fields with several gauge-based
precipitation datasets over land (not including A2006)
and concluded that ERA-40 had a low bias in annual
precipitation over the Amazon basin, but that the
source of the low annual bias was the rainy season,
whereas ERA-40 precipitation was biased upward in
the dry season. In the Mackenzie basin, Betts et al.
(2003b) generally reported higher monthly precipitation in ERA-40 than observations, although the differences decreased toward the end of the evaluation period (in 1997) when the density of gauges was lower.
Serreze et al. (2005) evaluated ERA-40 over the major
Arctic drainage basins and found that ERA-40 underestimated precipitation over northern Europe and Russia but overestimated precipitation over the Canadian
portion of the domain. ERA-40 precipitation was compared to A2006 in the pan-Arctic for the 1979–99 period by Su et al. (2006), who found that ERA-40 captured well the monthly variability, with annual precipitation slightly (6%) lower than A2006 for the Ob,
Yenisei, Lena, and Mackenzie basins.
It should be noted that some of these comparisons
are complicated by the fact that various GPCP versions
incorporate observations in different ways. For instance, the GPCP version 2 combination (Adler et al.
2003) includes high-latitude precipitation gauge data
that version 1 did not have. GPCP version 1 was indirectly evaluated by Nijssen et al. (2001b) who concluded that GPCP version 1 underestimated precipitation in orographically complex areas, especially in the
Columbia and Brahmaputra basins as deduced from the
comparison of simulated and observed river discharge.
GPCP version 2 was also evaluated by Adler et al.
(2003) over two 2.5° grid cells in Oklahoma through
comparisons with the Oklahoma Mesonet network.
Adler et al. showed a 1% upward bias relative to observations for 1998–2000 monthly precipitation. Because GPCP 1DD is scaled to match GPCP version 2
monthly values, the two datasets should be similar for
the same temporal and spatial aggregations.
391
downward radiation. VIC predicted hydrologic state
variables (soil moisture and snow water equivalent) and
moisture (evapotranspiration and runoff) and energy
(latent and sensible heat, reflected solar radiation, and
emitted longwave radiation) fluxes. In practice, only
precipitation, minimum and maximum daily temperatures, and surface wind are required to force VIC. The
other model forcing variables are indexed to daily temperature and the daily temperature range as described
by Maurer et al. (2002). The input fields for our three
VIC simulations were completed with Adam et al.
(2006) temperature (along with other fields derived
from the surface air temperature and daily temperature
range) and wind fields. Identical forcing fields—except
precipitation—and common model initial conditions
and parameters allowed us to isolate the effects of differences in precipitation fields on our hydrologic simulations.
a. Common temperature and wind fields
The station-based A2006 global precipitation dataset
at 0.5° latitude–longitude spatial resolution was used
directly to force VIC, along with daily temperature
maxima and minima, which were extended from Nijssen et al. (2001a), and wind data was taken from
NCEP–NCAR reanalysis using linear interpolation. To
complement the 1979–99 precipitation dataset, the
1979–93 Nijssen et al. (2001a) global daily temperature
dataset was extended to 1999 at 0.5° spatial resolution
(Adam et al. 2006). Note that the original data reported
in Nijssen et al. (2001a) were derived at 0.5° and then
aggregated to 2°; the aggregation was dropped. As in
Maurer et al. (2002), surface wind was taken from the
NCEP–NCAR reanalysis (Kalnay et al. 1996) and linearly interpolated to 0.5°. VIC was then run with identical temperature and wind fields, and with the three
precipitation datasets: A2006, ERA-40 precipitation interpolated to 0.5°, and the GPCP 1DD precipitation
field (also interpolated to 0.5°).
We considered using ERA-40 surface air temperature and other variables in our hydrologic simulation
with ERA-40 precipitation input field; however, we
opted not to do so in the interest of isolating the hydrologic implications of differences in precipitation
alone.
3. Experimental design
The three precipitation datasets described above
were formulated to provide forcings to the VIC model,
which was run over the global land areas at 0.5° latitude–longitude spatial resolution globally. The VIC
model was forced at the land surface by precipitation,
surface air temperature, surface wind, humidity, and
b. Common initialization and parameterization
For all three VIC simulations, a common model initialization strategy was used. Specifically, the model
was run using Adam et al.’s (2006) forcing dataset for
the 1979–96 period starting with initial soil moisture set
to field capacity on 1 January 1979. The snow and soil
392
JOURNAL OF HYDROMETEOROLOGY
VOLUME 9
FIG. 1. Location of the nine primary basins (dotted numbered). Different grays delineate continental
boundaries (North America, South America, Africa, Europe–Middle East, Russia, Asia, and Oceania).
Greenland is not included in this analysis.
moisture values on 1 January 1997 were then used to
initialize simulations for each of the three datasets for
the period 1997–99, which is the period of comparison
in this study. In this way, differences in the derived
hydrologic variables are entirely attributable to differences in the precipitation forcing data (quantity and
spatial distribution) during the period 1997–99, and not
to model initialization. For all three datasets, model
parameters were taken from Nijssen et al. (2001b), interpolated to the 0.5° grid. Other soil and vegetation
parameters were derived at 0.5° resolution, as described
in Su et al. (2005) and used in Su et al. (2005) and
Haddeland et al. (2006).
4. Results
We present results for annual precipitation and derived hydrologic variables for nine large river basins,
and then for each continent. We also compare ERA-40
and GPCP 1DD precipitation at a daily time scale for
the 1997–2002 period (longer overlap). The A2006 data
are not included in these last comparisons because they
end in 1999 and in any event the monthly observations
are disaggregated to a daily time step using a statistical
scheme based on observed daily precipitation. Hereafter, references to A2006, GPCP 1DD, and ERA-40
signify either the raw precipitation data or the hydrologic variables simulated by VIC when forced with
those datasets. For example, ERA-40 runoff refers to
the VIC-simulated runoff when forced with ERA-40
data, and not the ERA-40 runoff as simulated by the
ECMWF land surface model embedded in the ERA-40
reanalysis. Comparisons, unless otherwise specified, are
relative to A2006. Although A2006 precipitation is
based on observations and is expected therefore to be
closer to the truth in regions with high precipitation
gauge density, uncertainties as explained below arise in
certain regions and the choice of A2006 as reference is
mostly a matter of convention, and this convention
should not necessarily be interpreted as meaning that it
is generally preferred.
a. Primary basins
Nine primary basins were selected from among the
26 simulated by Nijssen et al. (2001b) based on the
presence of minimal anthropogenic effects (primarily
reservoir storage and diversion), availability of observed river discharge, and our desire to represent a
range of hydroclimatic conditions. Observed discharge
data come from the Global Runoff Data Center
(GRDC). Because there is no specific calibration of the
VIC model in this analysis (e.g., to minimize differences
between simulated and observed discharge when a
given precipitation dataset is used to force the VIC
model), the comparison with observed discharges is not
intended to help decide which dataset is closer to reality, but rather is a means of evaluating the sensitivity of
predicted runoff to differences in the forcing data. The
basins we selected are the Amazon, Congo, Danube,
Mackenzie, Mekong, Mississippi, Senegal, Yellow, and
Yenisei (Fig. 1).
Figure 2 and Table 1 show that in the two Arctic
basins (i.e., the Yenisei and Mackenzie), ERA-40 annual precipitation is very close to A2006, whereas
GPCP 1DD precipitation is much lower. This result is
similar to the findings of Su et al. (2006). Despite the
slight precipitation underestimation relative to A2006,
ERA-40 slightly overestimates the runoff (Table 1) because of different precipitation spatial distributions
(other meteorological forcings are identical). GPCP
1DD underestimates all derived hydrologic variables.
Serreze et al. (2005) found that ERA-40 overestimates
JUNE 2008
VOISIN ET AL.
393
FIG. 2. The 1997–99 annual precipitation P (left side of each panel and left axis), simulated
runoff R (center of each panel and left axis), and simulated discharge D (right side of each
panel and right axis) for GPCP 1DD (G), ERA-40 (E), and A2006 (A) for the primary basins
as defined in Nijssen et al. (2001b). For instance, the annual precipitation for GPCP 1DD is
coded PG. Here DO is the GRDC observed discharge for the available period at the stations.
precipitation in Arctic North America and underestimates it in Russia and northern Europe relative to
GPCP version 1. However GPCP version 2 (to which
GPCP 1DD is scaled) includes more gauge stations
than in version 1 (Adler et al. 2003), which might explain the difference in our results with theirs.
In the tropical basins (i.e., the Amazon, Congo,
Mekong, and part of the Senegal), GPCP 1DD and
ERA-40 both have lower precipitation than A2006 except in the Mekong, where ERA-40 is very close to
A2006 and GPCP 1DD is slightly higher than A2006
(Fig. 2; Table 1). ERA-40 is closer to A2006 in the
Congo and Mekong (⫺13.4% and ⫺0.4%, respectively)
but differs more in the Senegal (⫺51.6%) and the Amazon (⫺26.5%). Both ERA-40 and GPCP 1DD underestimate runoff relative to A2006 in all tropical basins.
As a result, ERA-40 and GPCP 1DD evapotranspiration estimates are usually lower than A2006 except in
the Mekong, and in the Congo for ERA-40 (Table 1).
Both ERA-40 and GPCP 1DD underestimate A2006
annual discharge (Fig. 2; Table 2) although ERA-40 is
closer to the GRDC (observed) annual discharge in the
Mekong, Senegal, and Congo. Betts et al. (2005) evaluated the ERA-40 water and energy budgets over the
Amazon basin using observations from Dai et al. (2004)
and Marengo (2004, 2005) over a longer period of time.
They concluded that “ERA-40 precipitation overestimates observations during the rainy season, underestimates during dry season and has a very low annual
bias.” Our results show the same tendency in seasonal
bias (not shown) with a low annual bias of ⫺2.3% in
precipitation for ERA-40 relative to GPCP 1DD. Rela-
394
JOURNAL OF HYDROMETEOROLOGY
VOLUME 9
TABLE 1. Annual precipitation (mm), simulated runoff (mm), evapotranspiration (mm), soil moisture (mm), and SWE (mm) and
corresponding relative differences in percent relative to A2006 (bold), and relative to GPCP 1DD (italic) for the nine primary basins.
Precipitation
Runoff
Evapotranspiration
Soil moisture
SWE
ERA-40
GPCP 1DD
ADAM
ERA-40
ERA-40
GPCP 1DD
ERA-40
GPCP 1DD
ADAM
ERA-40
ERA-40
GPCP 1DD
ERA-40
GPCP 1DD
ADAM
ERA-40
ERA-40
GPCP 1DD
ERA-40
GPCP 1DD
ADAM
ERA-40
ERA-40
GPCP 1DD
ERA-40
GPCP 1DD
ADAM
ERA-40
ERA-40
GPCP 1DD
Mississippi
Mackenzie
Amazon
Danube
Yenisei
Congo
Senegal
Yellow
Mekong
642
708
949
⫺9.3
⫺32.3
⫺25.3
192
200
406
⫺4.2
⫺52.7
⫺50.6
511
559
569
⫺8.6
⫺10.2
⫺1.7
728
729
789
⫺0.3
⫺7.
⫺7.5
85
52
93
65.1
⫺8.3
⫺44.4
404
291
408
39.0
⫺1.1
⫺28.8
147
61
140
142.2
4.6
⫺56.8
245
237
255
3.5
⫺3.9
⫺7.1
219
206
220
6.4
⫺0.5
⫺6.5
376
170
356
121.0
5.7
⫺52.2
1749
1791
2379
⫺2.3
⫺26.5
⫺24.7
581
630
1166
⫺7.8
⫺50.1
⫺46.0
1178
1167
1209
⫺1.0
⫺2.5
⫺3.5
494
494
529
⫺0.1
⫺6.7
⫺6.6
0
0
0
632
738
891
⫺14.5
⫺29.1
⫺17.1
180
238
366
⫺24.3
⫺50.8
⫺35.1
481
529
527
⫺9.1
⫺8.8
0.3
744
771
803
⫺3.5
⫺7.3
⫺3.9
130
135
200
⫺3.5
⫺35.0
⫺32.6
541
402
544
34.5
⫺0.7
⫺26.2
235
131
232
79.5
1.2
⫺43.6
299
286
314
4.5
⫺4.7
⫺8.8
241
227
250
6.1
⫺3.4
⫺8.9
863
514
637
68.0
35.5
⫺19.3
1462
1165
1689
25.5
⫺13.4
⫺31.0
402
201
592
99.8
⫺32.2
⫺66.1
1087
1004
1081
8.3
0.6
⫺7.1
1242
1159
1315
7.2
⫺5.6
⫺11.9
0
0
0
252
400
521
⫺36.9
⫺51.6
⫺23.3
34
107
192
⫺67.8
⫺82.1
⫺44.3
258
294
323
⫺12.2
⫺20.1
⫺9.0
507
565
595
⫺10.4
⫺14.9
⫺5.1
0
0
0
409
281
404
45.5
1.3
⫺30.4
126
40
100
212.8
26.7
⫺59.6
296
266
309
11.0
⫺4.4
⫺13.9
449
414
447
8.5
0.4
⫺7.4
21
1
0
1406
1469
1412
⫺4.3
⫺0.4
4.1
458
607
636
⫺24.6
⫺27.9
⫺4.5
944
857
778
10.2
21.4
10.2
455
453
454
0.4
0.2
⫺0.3
31
1
1
tively to A2006, ERA-40 precipitation is very close in
the dry season and has a bias in the rainy season (annual ⫺26.5%).
In the midlatitude rainy basins (i.e., the Danube, Yellow, and Mississippi), both ERA-40 and GPCP 1DD
underestimate precipitation relative to A2006, except
in the Yellow River basin where ERA-40 annual precipitation is slightly higher than A2006 (Fig. 2; Table 1).
The Yellow River basin has a semiarid cold climate,
and the relative apparent biases for ERA-40 are somewhat similar to the Arctic basins, with ERA-40 runoff
and snow water equivalent (SWE) both higher than
A2006 and evapotranspiration lower. Results for GPCP
1DD and the other basins show lower evapotranspiration, runoff, soil moisture, SWE, and annual discharge
relative to A2006 (Table 1). While the ERA-40 annual
TABLE 2. Relative differences in percent of annual discharge: relative to A2006 (bold), GPCP 1DD (italic), and relative to GRDC
observations (italic and bold). The GRDC monthly discharges have been computed based on the entire available record in order to
have some overlap with the 1997–99 period.
Obidos, Amazon
Kinshasa, Congo
Bratislava, Danube
Norman Wells, Mackenzie
Pakse, Mekong
Vicksburg, Mississippi
Bakel, Senegal
Huayuankou, Yellow
Igarka, Yenisei
ERA-40
GPCP 1DD
ERA-40
ERA-40
GPCP 1DD
A2006
⫺49.3
⫺35.0
⫺67.0
4.6
⫺23.7
⫺5.0
⫺81.3
50.2
⫺9.4
⫺44.0
⫺65.8
⫺39.9
⫺55.9
⫺29.7
⫺53.8
⫺39.9
⫺60.0
⫺47.4
⫺9.4
89.9
⫺45.1
137.2
8.6
105.8
⫺69.0
275.7
72.3
⫺47.8
11.7
⫺63.6
⫺8.3
⫺11.2
⫺4.5
26.2
366.4
⫺23.5
⫺42.4
⫺41.2
⫺33.6
⫺61.3
⫺18.2
⫺53.6
306.6
24.1
⫺55.6
2.9
72.0
10.5
⫺12.2
16.4
0.5
576.6
210.6
⫺15.6
JUNE 2008
VOISIN ET AL.
395
FIG. 3. The 1997–99 annual precipitation in mm day⫺1 for (a) ERA-40, (b) GPCP 1DD,
and (c) A2006.
discharge is very close to A2006 (and GRDC observations) in the Mississippi, GPCP 1DD is closer in the
Danube (Table 2; Fig. 2). Even though GPCP 1DD
precipitation is closer to A2006 than to ERA-40, the
GPCP 1DD SWE underestimation is either close (e.g.,
Danube) or much larger (e.g., Mississippi) than for
ERA-40 due to different spatial distributions.
b. Continental-scale evaluation
Figure 3 shows the 1997–99 annual daily average precipitation for ERA-40, GPCP 1DD, and A2006 over
the global land areas. Figure 4 shows the difference
between the 1997–99 annual daily precipitation averages of GPCP 1DD and A2006, ERA-40 and A2006,
and ERA-40 and GPCP 1DD. The delineation in Fig. 4
shows the area where the orographic correction of
A2006 was applied. Figure 5 shows the monthly cross
correlation for the longest overlap period between
A2006 and GPCP 1DD (1997–99), A2006 and ERA-40
(1979–99), and ERA-40 and GPCP 1DD (1997–August
2002). Each month value has first been standardized
(by subtracting the mean and dividing by the standard
deviation) with the corresponding month time series
(i.e., 1979–99 for A2006, 1997–2005 for GPCP 1DD,
396
JOURNAL OF HYDROMETEOROLOGY
VOLUME 9
FIG. 4. The 1997–99 annual precipitation relative error (%) of (a) GPCP 1DD with respect
to A2006, (b) ERA-40 with respect to A2006, and (c) ERA-40 with respect to GPCP 1DD.
The black has not been processed. The delineated contours indicate the area adjusted for
orography in A2006.
and 1979-August 2002 for ERA-40). The cross correlation is then applied to the standardized fields over the
overlap period. This cross correlation indicates the
pairwise agreement of the datasets with respect to
monthly anomalies, with the seasonal cycle removed.
Figure 6 shows the standard monthly cross correlation;
the cross correlation is applied to the 1979–99, 1997–
2001, and 1997–99 monthly time series of A2006 and
ERA-40, ERA-40 and GPCP 1DD, and A2006 and
GPCP 1DD, respectively, with no prior standardiza-
tion. This standard cross correlation provides a measure of the agreement in seasonality between the pairs
of datasets because the seasonal cycle mostly influences
the cross correlations rather than monthly anomalies.
For clarity, the terms “monthly anomaly” and “monthly
anomaly correlation” are used for the first cross correlation, and “seasonality” and “seasonality correlation”
for the second cross correlation. For purposes of these
comparisons, Eurasia is split between Asia, Russia, and
Europe–Middle East. Results are also segregated for
JUNE 2008
VOISIN ET AL.
397
FIG. 5. Cross correlation of monthly precipitation anomaly between (a) A2006 and GPCP
1DD for the 1997–99 period, (b) A2006 and ERA-40 for the 1979–99 period, and (c) ERA-40
and GPCP 1DD for the 1997–August 2002 period. Monthly values have first been standardized with the corresponding monthly 1979–99, 1979–August 2002, and 1997–2005 monthly
time series, before the computation of the cross correlation for the overlap period.
areas of complex terrain (delineation shown in Fig. 4)
and the tropical band, as well as for the entire global
land area.
1) NORTH AMERICA
Figures 3 and 4 and Table 3 show that over North
America, both GPCP 1DD and ERA-40 have lower
annual precipitation, (by 34.2% and 17.9%, respectively), relative to A2006. These results at the continental scale for North America are consistent with the results for the Mississippi basin shown in section 2a. The
GPCP 1DD and ERA-40 runoff is smaller than A2006
by 26.8% and 52.1%, respectively, over North America,
whereas evapotranspiration is less by 4.1% and 5.8%,
respectively. Differences in SWE are due to the differences in precipitation and to its spatial distribution (Fig.
4), and especially to the orographic correction in
A2006. ERA-40 SWE is only slightly lower than A2006
(by 2.1%) whereas GPCP 1DD (which is known to
substantially underestimate precipitation over mountainous areas) produces 23.5% less SWE. These differences have considerable geographic variability; ERA40 tends to have less SWE than A2006 in the Colorado,
398
JOURNAL OF HYDROMETEOROLOGY
VOLUME 9
FIG. 6. Cross correlation of monthly precipitation seasonality between (a) A2006 and GPCP
1DD for the 1997–99 period, (b) A2006 and ERA-40 for the 1979–99 period, and (c) ERA-40
and GPCP 1DD for the 1997–August 2002 period.
Mississippi, and Columbia basins and more in the
Yukon, Saint Lawrence, and Mackenzie basins (see the
appendix).
On the other hand, GPCP 1DD produces less SWE
than A2006 virtually everywhere. There is evidence of
a contrast in the differences in precipitation (Fig. 4) in
the vicinity of the U.S.–Canadian border, which may
have to do with the gauges used for the rescaling and
the gauge catch deficiency adjustment of the Adam and
Lettenmaier (2003) precipitation dataset. ERA-40 and
GPCP 1DD tend to be closer to each other over the
U.S., with both being substantially drier than to A2006.
Over Canada though, A2006 and GPCP 1DD tend to
be closer to each other, and both are drier than ERA40. As explained in Adam and Lettenmaier (2003), precipitation over Canada was handled in a slightly different way than elsewhere globally due to separate reports
of liquid and solid precipitation, and this difference applies to A2006 as well. An equivalent procedure was
adopted in GPCP version 1 (and 2) using Global Precipitation Climatology Center (GPCC) precipitation
with stations reporting solid and liquid values over
Canada. The use of this equivalent procedure in GPCP
versions performed specifically over Canada explains
the spatially homogeneous difference in magnitude between GPCP 1DD and A2006, while comparisons with
JUNE 2008
399
VOISIN ET AL.
TABLE 3. Annual average water balance components over the continents for ERA-40, GPCP 1DD, and A2006 (mm). The quantity
R/P is the ratio of runoff to precipitation. Correlation (seasonality) is the 1997–99 monthly correlation expressing the seasonality in
precipitation between ERA-40 and A2006, GPCP 1DD and A2006, and ERA-40 and GPCP 1DD (bold). Correlation (month) is the
monthly correlation expressing the monthly precipitation anomalies. Entries in italic are unitless.
North
South
America America Africa Oceania Russia
Precipitation
ERA-40
GPCP 1DD
A2006
Runoff
ERA-40
GPCP 1DD
A2006
Evapotranspiration ERA-40
GPCP 1DD
A2006
Soil moisture
ERA-40
GPCP 1DD
A2006
SWE
ERA-40
GPCP 1DD
A2006
R/P
ERA-40
GPCP 1DD
A2006
Correlation
ERA-40
(seasonality)
GPCP 1DD
ERA-40
Correlation
ERA-40
(month)
GPCP 1DD
ERA-40
601
482
733
277
181
380
340
334
355
380
370
403
122
96
125
0.46
0.38
0.52
0.55
0.61
0.61
0.45
0.42
0.51
1579
1314
1704
644
450
803
944
872
899
579
573
605
1
1
2
0.41
0.34
0.47
0.53
0.74
0.61
0.31
0.49
0.35
602
1268
549
1010
777
1079
167
654
123
367
296
463
453
621
439
634
476
604
679
402
676
428
727
428
0
1
0
1
0
2
0.28
0.51
0.22
0.36
0.38
0.43
0.55
0.57
0.69
0.68
0.61
0.65
0.23
0.49
0.30
0.50
0.26
0.53
ERA-40 tend to show some inhomogeneities in the vicinity of the U.S–Canadian border (Fig. 4). Despite this
locally specific procedure, monthly correlations are
more spatially consistent between ERA-40 and GPCP
1DD over the entire North American continent, than
with A2006 (Figs. 5 and 6).
2) SOUTH AMERICA
Averaged over South America, Table 3 shows that
ERA-40 precipitation is overall closer to A2006 (7.3%)
than GPCP 1DD (22.9%). Figure 4 shows that the
ERA-40 differences from A2006 are larger locally than
are GPCP 1DD. The largest ERA-40 precipitation differences from A2006 are localized in the northern part
of the continent, in the Sao Francisco, Amazon, and
Uruguay basins, and along the Andes (see also the appendix). As for North America, runoff is the most sensitive variable with ⫺19.9% and ⫺43.9% differences
for ERA-40 and GPCP 1DD, respectively, relative to
A2006 whereas evapotranspiration is higher by 5% for
ERA-40 and lower by 3% for GPCP 1DD relative to
A2006 (Table 3). SWE is much smaller for both ERA40 (50.8%) and GPCP 1DD (52.6%) relative to A2006,
a result that probably has to do with the much smaller
543
453
513
264
184
232
271
274
280
424
416
426
85
60
66
0.49
0.41
0.45
0.69
0.75
0.65
0.60
0.59
0.58
Asia
726
577
726
309
211
356
417
376
372
531
507
530
15
7
8
0.42
0.37
0.49
0.66
0.73
0.67
0.44
0.49
0.50
Europe and Sloped Tropical
Global
Middle East area
band
land areas
399
540
680
150
222
345
281
333
339
619
651
674
21
20
26
0.38
0.41
0.51
0.63
0.71
0.64
0.50
0.53
0.49
832
603
884
420
235
475
422
392
409
483
465
501
83
66
82
0.51
0.39
0.54
0.59
0.69
0.62
0.43
0.47
0.46
1206
1005
1293
482
329
577
736
683
709
608
604
643
0
0
0
0.40
0.33
0.45
0.57
0.72
0.64
0.26
0.38
0.31
738
631
817
312
221
379
435
424
436
509
505
532
47
35
43
0.42
0.35
0.46
0.60
0.70
0.63
0.43
0.47
0.46
areas affected by snow in South America relative to
North America when performing the orographic correction. The seasonality correlation (Table 3) relative
to A2006 is higher for GPCP 1DD (0.74) than for
ERA-40 (0.53). The monthly correlations (Fig. 5) show
that ERA-40, with correlations lower than 0.35, has
very different characteristics in the northern part of the
continent than the other two datasets. In the Amazon
basin, for instance, ERA-40 captures neither the seasonality nor the monthly anomalies of the other two
datasets. In general, GPCP 1DD and A2006 agree
much more with each other on the basis of monthly
anomalies and seasonality than either do with ERA-40,
especially in the northern part of the continent and
some high-altitude regions (Figs. 5 and 6).
3) AFRICA
In Africa, both ERA-40 and GPCP 1DD produce
lower precipitation than A2006, by 22.5% and 29.3%,
respectively (Table 3). ERA-40 in general agrees more
with GPCP 1DD than with A2006 (see Fig. 4). In areas
of very low precipitation, ERA-40 tends to produce less
precipitation than GPCP 1DD [e.g., in the Senegal and
Lake Chad basins (see the appendix)]. ERA-40 evapo-
400
JOURNAL OF HYDROMETEOROLOGY
transpiration is 4.7% lower and the runoff is 43.1%
lower (Table 3) than A2006. ERA-40’s lower precipitation relative to A2006 for very dry areas is confirmed
by the highest negative relative differences in dry basins
like the Senegal, Niger, and Lake Chad (see the appendix). GPCP 1DD precipitation is 29.3% lower than
A2006 (the greatest difference of all the continents),
leading to the largest relative difference in runoff
(⫺58%) and evapotranspiration (⫺7.8%) among all
continents as well. Figure 5 shows that ERA-40 tends
not to show the monthly anomalies in high precipitation
areas (central Africa) and drier northern Africa that
are present in A2006, while correlations are above 0.5
between ERA-40 and A2006 in dry southern Africa.
GPCP 1DD correlates more highly with A2006, although it presents some similar spatial characteristics to
ERA-40.
4) OCEANIA
Spatial results can be divided into two subregions
within Oceania: Australia–New Zealand and New
Guinea–Indonesia. ERA-40 is much wetter than A2006
and GPCP 1DD in New Guinea and Indonesia and
drier elsewhere (Fig. 4). Over all Oceania, ERA-40 has
17.6% higher precipitation than A2006 (Table 3), and
consequently, higher runoff (42.7%) and evapotranspiration (6.6%). ERA-40 soil moisture is less than A2006
(⫺6%), but this result mostly reflects the different characteristics of the two subregions: the increase in precipitation is in hot and wet areas where evapotranspiration is limited by the amount of water and not by the
energy necessary to evaporate the water (Fig. 4). This
implies that there is no or little change in soil moisture
over the wet areas and the decrease in soil moisture is
mostly traceable to Australia (dry areas with lower precipitation). GPCP 1DD precipitation is close to A2006
(⫺6.6%), but its runoff is 21% lower and evapotranspiration is higher by 4.9% due to differences in precipitation patterns (Table 3). ERA-40, GPCP 1DD, and
A2006 are well correlated (above 0.5) for monthly
anomalies over Australia, but correlations are low (between 0.05 and 0.5) over New Guinea and Indonesia,
like most other areas in the tropical band (Fig. 5).
5) RUSSIAN
PART OF
EURASIA
Over most of the Russian part of Eurasia, ERA-40
has lower precipitation than A2006 but much higher
precipitation in a few locations (Fig. 4), resulting in an
overall 5.8% higher precipitation and 14.4% higher
runoff than A2006 (Table 3). GPCP 1DD has lower
precipitation than A2006 (⫺11.8%), resulting in 20.7%
lower runoff. Both ERA-40 and GPCP 1DD evapotranspiration and soil moisture are close to A2006. As
VOLUME 9
found in other regions, GPCP 1DD has less SWE
(9.2%) relative to A2006, mostly because of lower precipitation in mountainous areas. ERA-40 has more
SWE (28%). ERA-40, GPCP 1DD, and A2006 reasonably agree on the monthly anomalies (Fig. 5).
6) ASIAN
PART OF
EURASIA
Figure 4 shows that ERA-40 differences from the
other two datasets are variable over Asia with higher
precipitation relative to A2006 in the Himalayas but a
tendency toward lower precipitation elsewhere, as for
Russia. Locally, ERA-40 precipitation is higher in the
Brahmaputra, Ganges, and Indus River basins (see the
appendix), lower in the southern and western basins of
Asia, and close to A2006 in south eastern basins like
the Mekong (Table 1 and see the Mekong River basin
discussion above). Overall ERA-40 precipitation for
Asia is very close to A2006, while GPCP 1DD precipitation is 20.4% lower than A2006 (Table 3). ERA-40
runoff is 13.2% lower than A2006 and evapotranspiration is 12.4% higher. ERA-40 SWE is much higher than
A2006 (96.9%). GPCP 1DD runoff is 40.6% lower than
A2006, while evapotranspiration and soil moisture are
close to A2006. Similar to other regions, GPCP 1DD
SWE is lower than A2006. Although GPCP 1DD and
ERA-40 do not agree well on the amount of precipitation over the Himalayas (Fig. 4), their agreement there
is highest in terms of monthly anomaly correlations
(Fig. 5). On the other hand, ERA-40 and A2006
monthly anomalies agree the least in the Himalaya region. However, aside from the Himalayan area, the
agreement among the datasets in this region in terms of
the seasonality of precipitation is among the highest of
all continents (Table 3).
7) EUROPE
AND THE
MIDDLE EAST
In Europe and the Middle East, ERA-40 precipitation tends to be consistently lower than A2006 with an
overall ⫺41.3% difference relative to A2006, while
GPCP 1DD is 20.6% lower than A2006 (Table 3), but
in a spatially inhomogeneous manner (Fig. 4). This is
the largest difference between ERA-40 and A2006 over
all zones, and is attributable to the substantial area over
which the A2006’s orographic adjustment is applicable.
ERA-40 runoff is consistently lower than the A2006
runoff (⫺56.1%), while GPCP 1DD runoff is 35.4%
lower on average (Table 3). GPCP 1DD evapotranspiration and soil moisture are very close to A2006
whereas ERA-40 evapotranspiration is 17.3% lower.
Similar to all other continents GPCP 1DD has lower
SWE relatively to A2006. Despite locally varying results, ERA-40 also has lower SWE (⫺18.4%). Monthly
anomaly correlations are fairly consistent among ERA-
JUNE 2008
VOISIN ET AL.
40, GPCP 1DD, and A2006 (Fig. 5; Table 3). As shown
in Fig. 6, the monthly seasonality agrees well among the
three datasets.
8) COMPLEX
TERRAIN
We separately analyzed all areas of the globe with
substantial topographic complexity at large scales,
which, following A2006, was defined as 0.5° latitude–
longitude grid cells having a slope larger than 6 m (1000
m)⫺1 (Fig. 4 of A2006, delineation in Fig. 4). In those
areas, ERA-40 is close to A2006 with a ⫺5.9% difference relative to A2006, while GPCP 1DD is 31.7%
lower on average (which reflects the absence of an orographic correction in GPCP 1DD). Both ERA-40 and
GPCP 1DD evapotranspiration and soil moisture are
close to A2006 but ERA-40 runoff is 11.4% lower and
GPCP 1DD is 50.4% lower (Table 3). As in previous
results, ERA-40 SWE is close to A2006 despite local
differences and GPCP 1DD SWE is much lower
(⫺20.2%). Interestingly, even though GPCP 1DD does
not agree well with A2006 in annual quantities, it does
agree the best in terms of monthly seasonality and
monthly anomalies (Table 3). ERA-40 is closer to
GPCP 1DD than A2006 by those two measures.
9) TROPICAL
BAND
We define the tropical band as the global land area
between 25°N and 25°S. It corresponds to the average
ITCZ location over land areas and includes the seasonal zonal shift. In the tropical band, ERA-40 has
higher precipitation than GPCP 1DD (20.3%; Table 3),
which is in agreement with Hagemann et al. (2005).
However both ERA-40 and GPCP 1DD have lower
precipitation relative to A2006 (6.6% and 22.3%, respectively). As elsewhere, runoff is the most sensitive of
the simulated hydrology variables with ⫺16.4% and
⫺43% differences for ERA-40 and GPCP 1DD, respectively, relative to A2006 although local differences
vary (Congo, Niger, Indonesia, Amazon, and Orinoco,
see the appendix). GPCP 1DD and A2006 have the
closest agreement in terms of monthly and seasonality
anomalies (Table 3).
10) GLOBAL
LAND AREAS
Globally, ERA-40 has 9.6% lower precipitation than
A2006, resulting in 17.7% lower runoff, similar evapotranspiration, and 4.3% lower soil moisture (Table 3).
ERA-40 SWE is 8.6% higher than A2006, mostly because of higher SWE in Russia and Asia as noted earlier. GPCP 1DD has 22.8% lower precipitation than
A2006, with 41.8% lower runoff, 2.9% lower evapo-
401
transpiration, and 5% lower soil moisture. SWE is 19%
lower than A2006. Similar to most of the regional differences, the global monthly correlation is the highest
(0.47) between GPCP 1DD and A2006, although the
monthly correlation between ERA-40 and GPCP 1DD
is very close (0.46). The monthly seasonality correlation
is also the highest between GPCP 1DD and A2006
(0.70).
Table 4 (adapted and extended from Maurer et al.
2000) shows the global water balance for the three
datasets, in comparison with other published climatologies. ERA-40 (using all land surface budget components from ERA-40, as contrasted with VIC predictions
driven with ERA-40 precipitation), and GPCC precipitation and GRDC observed runoff estimates for the
corresponding global land area (GRDC 2004, unpublished manuscript) have been added to complete the
range of climatologies. Some uncertainty is inherent to
the different processing of those climatologies. For example, the difference in precipitation between raw
ERA-40 (1997–99) and ERA-40 used to force VIC during 1997–99 is due to the different processing steps (e.g.,
including only ERA-40’s 2.5° land cells as opposed to
interpolation to the 0.5° land grid using a few ocean
cells). Taken in comparison with all of the Table 4 estimates, A2006 precipitation is at the high end of the
range of the climatologies, whereas GPCP 1DD has the
lowest estimate and ERA-40 precipitation interpolated
to the VIC grid is near the middle of the range. VICsimulated evapotranspiration estimates are all lower
than the average of the alternate climatologies and runoff is larger. However, the highest evapotranspiration
estimates in the climatology range are due to very low
runoff estimates in comparison to observed GRDC
runoff. When evapotranspiration estimates are derived
by subtracting the GRDC runoff from the available
precipitation climatologies [ERA-40, GPCP version 1,
Lvovitch (1973), Baumgartner and Reichel (1975), and
GPCC)], the simulated VIC evapotranspiration estimates using the three different precipitation datasets
(ERA-40, GPCP 1DD, and A2006) are then higher
than the average climatology evapotranspiration estimate (404 mm; Table 4). Despite the differences in precipitation between the different datasets (i.e., A2006,
ERA-40, and GPCP 1DD), all three runoff ratios are
within the range of the other estimates, although somewhat higher for A2006 and ERA-40 (Table 4).
c. ERA-40 and GPCP 1DD precipitation
intermittency
Both ERA-40 and GPCP 1DD datasets are available
at a daily time steps. The 1997–August 2002 daily in-
402
JOURNAL OF HYDROMETEOROLOGY
VOLUME 9
TABLE 4. Comparison of water balance (mm yr⫺1) over global land areas [excluding Greenland and Antarctica; adapted from
Maurer et al. (2000)].
Model (M) or climatology (C)
P
ET
R
R/P
VIC with A2006 input (M 1997–99)
VICa with ERA-40 input (M 1997–99)
VICa with GPCP 1DD input (M 1997–99)
ERA-40 (M 1997–99)
VICc with Nijssen et al. (2001a) input (M 1979–93)
ERA-40 (M 1979–2001)
Lvovitch (1973) (C)
Baumgartner and Reichel (1975) (C)
Korzun (1978) (C)
Oki et al. (1995) (C)
Oki (1997) (C)
GRDC runoff with GPCC precipitation (C)
Means (of climatologies, excluding 1997–99)
817
738
631
698b
727d
719
834
746
727d
727d
727d
895
784
436
435
423
412
483
458
540
480
424e
562e
483e
515f
493
379
312
221
348
244
347
294
266
303
165
244
380
280
0.46
0.42
0.35
0.50
0.34
0.48
0.35
0.36
0.42
0.23
0.34
0.42
0.37
a
P ⫺ RGRDC
318
347
339
454
366
515
404
a
VIC calibration using Nijssen et al. (2001a) input, based on GPCP version 1.
Different precipitation than VIC with ERA-40 input (1997–99) because of the different spatial resolution (ERA-40 at 2.5°). The
interpolation to 0.5° included a few ocean cells.
c
No VIC calibration.
d
Precipitation climatology from GPCP version 1 (gauge-only product).
e
Based on cited runoff and GPCP precipitation.
f
Based on GRDC runoff and GPCC precipitation (as in GRDC 2004, unpublished manuscript).
b
termittency and root-mean-square differences (RMSDs)
were compared between the two datasets on a grid cell
by grid cell basis. Agreements of rain occurrence (either both datasets record a zero or both datasets record
a nonzero rain amount) vary from 59% in North
America and Russia to 74% in Africa (Table 5). Those
values increase when the intermittency is computed as
an average over river basins (not shown). Figure 7
shows the global intermittency agreement and RMSD.
GPCP 1DD and ERA-40 daily intermittency agree best
in the tropical band and least in the Arctic regions. The
intermittency agreement results are lower than the
0.85–0.94 values, depending on the density of the gauge
network, obtained by Adler et al. (2003) who compared
monthly GPCP version 2 with two 2.5° cells within
Oklahoma Mesonet region (Brock et al. 1995) because
our analysis was made at a higher temporal resolution
(daily), and on a much larger domain.
TABLE 5. Daily RMSD and intermittency agreement between
1997–August 2002 ERA-40 and GPCP 1DD.
North America
South America
Africa
Oceania
Russia
Asia
Europe–Middle East
Daily RMSD
(mm day⫺1)
Intermittency
agreement
0.033
0.063
0.025
0.032
0.034
0.054
0.017
0.59
0.69
0.74
0.72
0.59
0.61
0.67
Daily RMSDs relative to GPCP 1DD (Fig. 7a) are
greatest in the equatorial regions, various parts of Russia, northern Quebec and Ontario, western North
America, and India. RMSDs are smallest in Europe
and the Middle East (Table 5).
5. Conclusions
Global hydrologic simulations using the hydrological
model VIC and three different meteorological forcing
datasets (A2006, ERA-40, and GPCP 1DD) were
evaluated for the 1997–99 period during which all
datasets were available. The results of our comparisons
are generally consistent with previous, more local, comparisons of the three datasets, but the global comparisons in this paper allow a larger spatial perspective of
the differences among the datasets and their implications for the water balance at large scales. In particular,
GPCP 1DD tends to have less precipitation than both
A2006 and ERA-40 nearly everywhere and especially
in mountainous areas. ERA-40 precipitation is generally intermediate between A2006 and GPCP 1DD.
Simulated evapotranspiration tends to be toward the
high end of the range of climatologies, when derived by
difference using GRDC runoff observations. Globally
and in every continent, simulated runoff is much more
sensitive to precipitation differences than is evapotranspiration. At the global scale, simulated runoff to precipitation ratios were in the range of previous climatology estimates.
JUNE 2008
VOISIN ET AL.
403
FIG. 7. (a) RMSD (mm day⫺1) of ERA-40 with respect to GPCP 1DD and (b) daily
precipitation intermittency agreement between ERA-40 and GPCP 1DD for the 1997–August
2002 period.
Precipitation seasonality generally agreed well among
the three precipitation datasets. However, monthly precipitation anomalies were only moderately correlated
globally, due mostly to low correlations among the three
datasets in South America, Africa, and the tropical band.
The Russian part of Eurasia has on average the highest
correlations among the three precipitation datasets in
terms of monthly anomalies. The daily agreement of
intermittency of precipitation between GPCP 1DD and
ERA-40 exceeded 50% over most of the globe.
Because all three precipitation datasets are subject to
errors, albeit of different types, no conclusion could be
drawn as to which precipitation dataset is closest to
truth in general. For example, the lowest agreement
among the precipitation datasets is over Africa, which
has the lowest observation network density. River discharge does not help much in resolving the discrepancies because its simulation is dependent on calibration
to river discharge observations.
Nonetheless, there is some overall preference for
ERA-40 among the three datasets because of its long
climatology (cf. GPCP 1DD), daily availability, and
good agreement with A2006 over at least some parts of
the globe with high station densities, despite an overall
tendency for underestimation relative to A2006 except
in the Arctic and in Asia, and a significant apparent
precipitation overestimation in the Himalaya. Those
differences could be corrected via a bias correction or
adjustment of monthly values with respect to A2006.
However, such an approach would need to first address
issues such as the discontinuity in A2006 precipitation
at the Canadian–U.S. border and questions as to the
magnitude of the orographic adjustment in some locations. Bias correction of the satellite-based datasets is
problematic because of the relatively short satellite
records and the lack of overlap of the climatologies,
although these issues may be resolvable with time. Our
preference therefore is for ERA-40 precipitation for
use in global hydrological applications, recognizing the
nature of apparent biases over some portions of the
global land areas.
Acknowledgments. The authors wish to thank Jenny
Adam for her advice concerning the A2006 dataset.
541
402
544
235
131
232
299
286
314
241
227
250
863
514
637
Yenisei
489
331
469
208
85
185
269
257
278
219
207
223
706
366
566
Lena
766
906
1006
117
178
258
659
720
724
634
677
705
0
0
0
490
588
581
110
143
208
365
422
356
558
606
594
3
8
129
846
898
1106
156
164
340
709
741
751
675
684
724
0
0
0
Asia
902
338
724
528
65
465
388
314
288
608
539
590
399
74
198
2669
1621
2359
1497
626
1276
1174
1005
1080
444
411
442
0
0
0
1701
1877
1983
777
996
1093
963
915
921
963
978
987
0
0
0
Uruguay
2661
723
1305
1962
287
841
633
467
466
839
593
674
1512
280
334
796
999
1107
241
435
576
561
566
531
426
452
468
0
0
0
2019
1376
2291
1058
589
1547
962
795
741
639
591
629
6
0
2
519
945
978
101
432
491
429
517
491
403
442
442
0
0
0
1055
869
1056
470
358
525
578
513
531
435
402
421
58
12
13
North America
409
281
404
126
40
100
296
266
309
449
414
447
21
1
0
1406
1469
1412
458
607
636
944
857
778
455
453
454
31
1
1
Huang
He Chang Mekong
1337
1248
1468
508
523
712
818
716
743
645
637
656
0
0
0
Xi
1749
1791
2379
581
630
1166
1178
1167
1209
494
494
529
0
0
0
642
708
949
192
200
406
511
559
569
728
729
789
642
708
949
289
256
548
82
37
200
242
260
361
416
394
458
289
256
548
646
475
985
372
220
631
390
389
433
558
527
615
646
475
985
364
313
571
28
18
147
361
323
432
305
300
355
364
313
571
425
310
503
156
94
208
254
231
253
209
199
223
425
310
503
824
781
975
335
253
427
541
572
579
665
668
705
824
781
975
404
291
408
147
61
140
245
237
255
219
206
220
404
291
408
Rio
Saint
Amazon Mississippi Colorado Columbia Grande Yukon Lawrence Mackenzie
1341
770
1188
703
309
663
613
485
517
756
673
743
943
327
394
Indus Bramahputra Ganges Godavari Irrawaddy Krishna
South America
576
484
490
144
88
100
416
396
386
445
430
440
244
130
85
Amur
La
Sao
Parana Plata Colorado Francisco Orinoco
564
507
514
198
152
158
356
358
354
730
719
724
812
618
557
Ob
ERA-40
1215
GPCP 1DD 1150
ADAM
1355
Runoff
ERA-40
319
GPCP 1DD 299
ADAM
466
Evapotranspiration ERA-40
928
GPCP 1DD 883
ADAM
916
Soil moisture
ERA-40
738
GPCP 1DD 741
ADAM
780
SWE
ERA-40
0
GPCP 1DD
0
ADAM
0
Precipitation
(b)
ERA-40
GPCP 1DD
ADAM
Runoff
ERA-40
GPCP 1DD
ADAM
Evapotranspiration ERA-40
GPCP 1DD
ADAM
Soil moisture
ERA-40
GPCP 1DD
ADAM
SWE
ERA-40
GPCP 1DD
ADAM
Precipitation
(a)
Russia
1354
855
1350
628
254
732
706
604
624
596
531
571
339
52
54
Salween
Annual 1997–99 precipitation, simulated runoff, evapotranspiration, soil moisture, and SWE for several basins, in mm, in (a) the Russian and Asian parts of Eurasia, (b) North
and South America, (c) Africa, and (d) Europe and Middle East.
Detailed Analysis for Large Basins with Each Continent
APPENDIX
404
JOURNAL OF HYDROMETEOROLOGY
VOLUME 9
(d)
SWE
Soil moisture
Evapotranspiration
Runoff
Precipitation
SWE
Soil moisture
Evapotranspiration
Runoff
Precipitation
(c)
ERA-40
GPCP 1DD
ADAM
ERA-40
GPCP 1DD
ADAM
ERA-40
GPCP 1DD
ADAM
ERA-40
GPCP 1DD
ADAM
ERA-40
GPCP 1DD
ADAM
ERA-40
GPCP 1DD
ADAM
ERA-40
GPCP 1DD
ADAM
ERA-40
GPCP 1DD
ADAM
ERA-40
GPCP 1DD
ADAM
ERA-40
GPCP 1DD
ADAM
632
738
891
180
238
366
481
529
527
744
771
803
130
135
200
Danube
1462
1165
1689
402
201
592
1087
1004
1081
1242
1159
1315
0
0
0
Congo
393
708
795
69
220
293
381
498
510
668
760
779
25
30
42
Elbe
252
400
521
34
107
192
258
294
323
507
565
595
0
0
0
Senegal
499
693
800
100
209
297
433
496
509
590
647
674
40
32
52
Oder
555
512
693
141
101
224
425
416
460
736
736
781
0
0
0
Nile
711
938
1099
258
378
541
493
577
572
763
811
844
86
133
136
829
740
1032
109
99
313
735
659
717
1039
1031
1126
0
0
0
Malawi
333
378
393
39
54
66
306
332
335
470
496
505
0
0
0
Oranje
578
913
1103
179
383
547
449
537
549
708
791
817
231
272
375
Rhine
439
657
564
111
232
158
342
407
398
729
794
769
354
451
356
Don
Europe–Middle East
Rhone
473
589
740
173
210
326
329
380
410
649
690
710
0
0
0
Niger
Africa
APPENDIX. (Continued)
573
720
671
162
251
210
424
462
460
728
768
764
290
293
242
Dnieper
601
546
648
50
47
95
570
516
569
526
530
554
0
0
0
Limpopo
262
311
483
123
113
305
173
223
192
589
595
609
35
27
114
Euphrates
Tigris
742
732
1016
107
122
302
674
637
713
777
782
862
0
0
0
Zambezi
617
668
754
179
214
285
447
463
471
655
672
698
122
78
107
Wisla
236
318
451
37
83
170
216
235
275
458
480
501
0
0
0
Lake
Chad
559
510
663
138
80
164
442
441
490
551
563
610
0
0
0
Webi–
Shabelle
JUNE 2008
VOISIN ET AL.
405
406
JOURNAL OF HYDROMETEOROLOGY
REFERENCES
Adam, J. C., and D. P. Lettenmaier, 2003: Adjustment of global
gridded precipitation for systematic bias. J. Geophys. Res.,
108, 4257, doi:10.1029/2002JD002499.
——, E. A. Clark, D. P. Lettenmaier, and E. F. Wood, 2006: Correction of global precipitation products for orographic effects. J. Climate, 19, 15–38.
Adler, R. F., and Coauthors, 2003: The version-2 Global Precipitation Climatology Project (GPCP) monthly precipitation
analysis (1979–present). J. Hydrometeor., 4, 1147–1167.
Baumgartner, A., and E. Reichel, 1975: The World Water Balance:
Mean Annual Global, Continental and Maritime Precipitation, Evaporation and Runoff. Elsevier Scientific, 179 pp.
Betts, A. K., J. H. Ball, M. Bosilovich, P. Viterbo, Y.-C. Zhang,
and W. B. Rossow, 2003a: Intercomparison of water and energy budgets for five Mississippi subbasins between ECMWF
reanalysis (ERA-40) and NASA Data Assimilation Office
fvGCM for 1990-1999. J. Geophys. Res., 108, 8618,
doi:10.1029/2002JD003127.
——, ——, and P. Viterbo, 2003b: Evaluation of the ERA-40
surface water budget and surface temperature for the Mackenzie River basin. J. Hydrometeor., 4, 1194–1211.
——, ——, ——, A. Dai, and J. Marengo, 2005: Hydrometeorology of the Amazon in ERA-40. J. Hydrometeor., 6, 764–774.
Brock, F. V., K. C. Crawford, R. L. Elliott, G. W. Cuperus, S. J.
Stadler, H. L. Johnson, and M. D. Eilts, 1995: The Oklahoma
Mesonet—A technical overview. J. Atmos. Oceanic Technol.,
12, 5–19.
Chen, M. P., P. Xie, J. E. Janowiak, and P. A. Arkin, 2002: Global
land precipitation: A 50-yr monthly analysis based on gauge
observations. J. Hydrometeor., 3, 249–266.
Costa-Cabral, M. C., J. E. Richey, G. Goteti, D. P. Lettenmaier,
C. Feldkoetter, and A. Snidvongs, 2008: Landscape structure
and use, climate, and water movement in the Mekong River
basin. Hydrol. Processes, in press, doi:10.1002/hyp.6740.
Dai, A., K. E. Trenberth, and T. Qian, 2004: A global dataset of
Palmer severity index for 1870–2002: Relationship with soil
moisture and effects of surface warming. J. Hydrometeor., 5,
1117–1130.
Haddeland, I., T. Skaugen, and D. P. Lettenmaier, 2006: Anthropogenic impacts on continental surface water fluxes. Geophys. Res. Lett., 33, L08406, doi:10.1029/2006GL026047.
Hagemann, S., K. Arpe, and L. Bengtsson, 2005: Validation of the
hydrological cycle of ERA-40. ERA-40 Project Rep. Series
24, European Centre for Medium-Range Weather Forecasts,
Reading, United Kingdom, 46 pp.
Hernandez, A., G. Kelly, and S. Uppala, 2004: The TOVS/
ATOVS observing system in ERA-40. ERA-40 Project Rep.
Series 16, European Centre for Medium-Range Weather
Forecasts, Reading, United Kingdom, 49 pp.
Huffman, G. J., and Coauthors, 1997: The Global Precipitation
Climatology Project (GPCP) combined precipitation dataset.
Bull. Amer. Meteor. Soc., 78, 5–20.
——, R. F. Adler, M. Morrissey, D. T. Bolvin, S. Curtis, R. Joyce,
B. McGavock, and J. Susskind, 2001: Global precipitation at
one-degree daily resolution from multisatellite observations.
J. Hydrometeor., 2, 36–50.
——, ——, E. F. Stocker, D. T. Bolvin, and E. J. Nelkin, 2003:
Analysis of TRMM 3-hourly multisatellite precipitation estimates computed in both real and post-real time. Preprints,
12th Conf. on Satellite Meteorology and Oceanography, Long
Beach, CA, Amer. Meteor. Soc., P4.11.
VOLUME 9
Janowiak, J. E., A. Gruber, C. R. Kondragunta, R. E. Livezey,
and G. J. Huffman, 1998: A Comparison of the NCEP/
NCAR reanalysis precipitation and the GPCP rain gauge–
satellite combined dataset with observational error considerations. J. Climate, 11, 2960–2979.
Joyce, R. J., J. E. Janowiak, P. A. Arkin, and P. Xie, 2004:
CMORPH: A method that produces global precipitation estimates from passive microwave and infrared data at high
spatial and temporal resolution. J. Hydrometeor., 5, 487–503.
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437–471.
Kummerow, C., W. S. Olson, and L. Giglio, 1996: A simplified
scheme for obtaining precipitation and vertical hydrometeor
profiles from passive microwave sensors. IEEE Trans.
Geosci. Remote Sens., 34, 1213–1232.
Korzun, V. I., 1978: World Water Balance and Water Resources of
the World. Studies and Reports in Hydrology, Vol. 25,
UNESCO, 663 pp.
Liang, X., D. P. Lettenmaier, E. F. Wood, and S. J. Burges, 1994:
A simple hydrologically based model of land surface water
and energy fluxes for general circulation models. J. Geophys.
Res., 99, 14 415–14 428.
Lvovitch, M. I., 1973: The global water balance. Eos, Trans. Amer.
Geophys. Union, 54, 28–42.
Marengo, J. A., 2004: Interdecadal variability and trends of rainfall across the Amazon basin. Theor. Appl. Climatol., 78, 79–
96.
——, 2005: Characteristics and spatio-temporal variability of the
Amazon River basin water budget. Climate Dyn., 24, 11–22.
Maurer, E. P., 2007: Uncertainty in hydrologic impacts of climate
change in the Sierra Nevada, California, under two emissions
scenarios. Climatic Change, 82 (3–4), 309–325.
——, B. Nijssen, and D. P. Lettenmaier, 2000: Use of reanalysis
land surface water budget variables in hydrologic studies.
GEWEX News, No. 4, International GEWEX Project Office,
Silver Spring, MD, 6–8.
——, A. W. Wood, J. C. Adam, and D. P. Lettenmaier, 2002: A
long-term hydrologically based dataset of land surface fluxes
and states for the conterminous United States. J. Climate, 15,
3237–3251.
Mesinger, F., and Coauthors, 2006: North American regional reanalysis. Bull. Amer. Meteor. Soc., 87, 343–360.
Mitchell, T. D., and P. D. Jones, 2005: An improved method of
constructing a database of monthly climate observations and
associated high-resolution grids. Int. J. Climatol., 25, 693–712.
Nijssen, B., R. Schnur, and D. P. Lettenmaier, 2001a: Global retrospective estimation of soil moisture using the Variable Infiltration Capacity land surface model, 1980–93. J. Climate,
14, 1790–1808.
——, G. M. O’Donnell, D. P. Lettenmaier, D. Lohmann, and
E. F. Wood, 2001b: Predicting the discharge of global rivers.
J. Climate, 14, 3307–3323.
Oki, T., 1997: Validating the runoff from LSP-SVAT models using
a global river routing network by one degree mesh. Preprints,
13th Conf. on Hydrology, Long Beach, CA, Amer. Meteor.
Soc., 319–322.
——, K. Musiake, H. Matsuyama, and K. Masuda, 1995: Global
atmospheric water balance and runoff from large river basins.
Hydrol. Processes, 9, 655–678.
Olson, W. S., C. D. Kummerow, Y. Yong, and W.-T. Tao, 1999:
Atmospheric latent heating distributions in the tropics from
satellite passive microwave radiometer measurements. J.
Appl. Meteor., 38, 633–644.
JUNE 2008
VOISIN ET AL.
Shepard, D. S., 1984: Computer mapping: The SYMAP interpolation algorithm. Spatial Statistics and Models, G. L. Gaile
and C. Willmott, Eds., Springer, 133–145.
Serreze, M. C., A. P. Barrett, and F. Lo, 2005: Northern highlatitude precipitation as depicted by atmospheric reanalysis
and satellite retrievals. Mon. Wea. Rev., 133, 3407–3430.
Snelgrove, K. R., E. D. Soulis, F. R. Seglenieks, and N. Kouwen,
2005: The application of hydrological models in MAGS: Lessons learned for PUB. Prediction in Ungauged Basins: Approaches for Canada’s Cold Regions, C. Spence, J. W.
Pomeroy, and A. Pietroniro, Eds., Canadian Water Resources Association, 139–164.
Sorooshian, S., K. Hsu, X. Gao, H. V. Gupta, B. Imam, and D.
Braithwaite, 2000: Evaluation of PERSIANN system satellite-based estimates of tropical rainfall. Bull. Amer. Meteor.
Soc., 81, 2035–2046.
Su, F., J. C. Adam, L. C. Bowling, and D. P. Lettenmaier, 2005:
Streamflow simulations of the terrestrial Arctic domain. J.
Geophys. Res., 110, D08112, doi:10.1029/2004JD005518.
——, ——, K. E. Trenberth, and D. P. Lettenmaier, 2006: Evaluation of surface water fluxes of the pan-Arctic land region
with a land surface model and ERA-40 reanalysis. J. Geophys. Res., 111, D05110, doi:10.1029/2005JD006387.
407
Susskind, J., P. Piraino, L. Rokke, L. Iredell, and A. Mehta, 1997:
Characteristics of the TOVS Pathfinder A dataset. Bull.
Amer. Meteor. Soc., 78, 1449–1472.
Tang, Q., T. Oki, and S. Kanae, 2006: A distributed biosphere
hydrological model (DBHM) for large river basin. Ann. J.
Hydraul. Eng. JSCE, 50, 37–42.
Troccoli, A., and P. Kallberg, 2004: Precipitation correction in the
ERA-40 reanalysis. ERA-40 Project Rep. Series, Vol. 13,
European Centre for Medium-Range Weather Forecasts,
Reading, United Kingdom, 10 pp.
Uppala, S. M., and Coauthors, 2005: The ERA-40 re-analysis.
Quart. J. Roy. Meteor. Soc., 131, 2961–3012.
Willmott, C. J., and K. Matsuura, cited 2001: Terrestrial air temperature and precipitation: Monthly and annual time series
(1950–1999) (version 1.02). Center for Climate Research,
University of Delaware, Newark, DE. [Available online at
http://climate.geog.udel.edu/˜climate/html_pages/archive.
html.]
Xie, P., and P. A. Arkin, 1997: Global precipitation: A 17-year
monthly analysis based on gauge observations, satellite estimates, and numerical model outputs. Bull. Amer. Meteor.
Soc., 78, 2539–2558.