iri – international research institute for climate and society ...blyon/references/p30.pdfmodeling...

Development of a 50-Year High-Resolution Global Dataset of Meteorological Forcingsfor Land Surface Modeling

JUSTIN SHEFFIELD

Department of Civil and Environmental Engineering, Princeton University, Princeton, New Jersey

GOPI GOTETI

Department of Earth System Science, University of California, Irvine, Irvine, California

ERIC F. WOOD

Department of Civil and Environmental Engineering, Princeton University, Princeton, New Jersey

(Manuscript received 28 March 2005, in final form 24 August 2005)

ABSTRACT

Understanding the variability of the terrestrial hydrologic cycle is central to determining the potential forextreme events and susceptibility to future change. In the absence of long-term, large-scale observations ofthe components of the hydrologic cycle, modeling can provide consistent fields of land surface fluxes andstates. This paper describes the creation of a global, 50-yr, 3-hourly, 1.0° dataset of meteorological forcingsthat can be used to drive models of land surface hydrology. The dataset is constructed by combining a suiteof global observation-based datasets with the National Centers for Environmental Prediction–NationalCenter for Atmospheric Research (NCEP–NCAR) reanalysis. Known biases in the reanalysis precipitationand near-surface meteorology have been shown to exert an erroneous effect on modeled land surface waterand energy budgets and are thus corrected using observation-based datasets of precipitation, air tempera-ture, and radiation. Corrections are also made to the rain day statistics of the reanalysis precipitation, whichhave been found to exhibit a spurious wavelike pattern in high-latitude wintertime. Wind-induced under-catch of solid precipitation is removed using the results from the World Meteorological Organization(WMO) Solid Precipitation Measurement Intercomparison. Precipitation is disaggregated in space to 1.0°by statistical downscaling using relationships developed with the Global Precipitation Climatology Project(GPCP) daily product. Disaggregation in time from daily to 3 hourly is accomplished similarly, using theTropical Rainfall Measuring Mission (TRMM) 3-hourly real-time dataset. Other meteorological variables(downward short- and longwave radiation, specific humidity, surface air pressure, and wind speed) aredownscaled in space while accounting for changes in elevation. The dataset is evaluated against the bias-corrected forcing dataset of the second Global Soil Wetness Project (GSWP2). The final product providesa long-term, globally consistent dataset of near-surface meteorological variables that can be used to drivemodels of the terrestrial hydrologic and ecological processes for the study of seasonal and interannualvariability and for the evaluation of coupled models and other land surface prediction schemes.

1. Introduction

The availability of large-scale, long-term datasets ofthe land surface water and energy budgets is essentialfor understanding the global environmental system andinteractions with human activity, especially in the face

of potential climatic change. However, consistent ob-servations of components of the land surface water andenergy budgets are routinely unavailable over largescales. While some terms of the surface water balanceare reasonably well observed at least over some parts ofthe globe (precipitation and runoff in particular), otherterms including evapotranspiration, soil moisture, andsurface water are virtually absent of direct observationsat large scales. Many of these variables are difficult tomeasure because of technical, monetary, and politicallimitations. In the case of soil moisture, which forms a

Corresponding author address: Dr. Justin Sheffield, Depart-ment of Civil and Environmental Engineering, Princeton Univer-sity, Princeton, NJ 08544.E-mail: [email protected]

3088 J O U R N A L O F C L I M A T E VOLUME 19

© 2006 American Meteorological Society

JCLI3790

key element for drought assessment and medium- andlong-range prediction, global (or even regional, withonly a few exceptions) in situ measurement networksare grossly inadequate for hydrologic prediction pur-poses, and land surface hydrology models have gener-ally evolved without the use of direct observations ofthis key state variable. In terms of surface energy fluxesand evaporation, these are inherently difficult to mea-sure and are thus essentially nonexistent over largescales. The use of remote sensing has provided greatpotential for the large-scale measurement of some vari-ables (notably albedo, radiative surface temperature,and soil moisture) but is restricted to indirect quantitiesand, in the case of soil moisture, to low-vegetated re-gions and the top few centimeters.

It has been suggested that an alternative to estimat-ing large-scale water cycle terms directly from observa-tions is to use land surface models (LSMs), in eitheroffline (forced with surface meteorological observa-tions) or coupled (with an atmospheric GCM) modes(e.g., Lau et al. 1994; Liang et al. 1994; Levis et al. 1996;Werth and Avissar 2002). LSMs close the water budgetby construct, so if the meteorological forcing data areaccurate and model biases are small, these constructedwater balance terms might be used in lieu of observa-tions and provide a consistent picture of the water andenergy budgets. Budget closure is not achievable fromobservations even at small scales. In fact, analyses ofwater and energy cycle variables estimated through ob-servations (in situ and/or remote sensing) will not pro-vide water cycle closure (Roads et al. 2003; Pan andWood 2004) because of sampling and retrieval errors.However, through research activities like the NorthAmerican Land Data Assimilation System (NLDAS;K. E. Mitchell et al. 2004) and Global LDAS (GLDAS;Rodell et al. 2004), the capability of land surface mod-els to produce meaningful estimates of land surface hy-drologic conditions over large areas has been demon-strated. Therefore, the contention is that observation-forced, offline simulations using state-of-the-art landsurface models provide the best estimate of global wa-ter cycle variables.

Nevertheless, while estimates of water cycle variablesobtained through land surface modeling are consistent,these estimates can be subject to large errors due toerrors in model inputs and meteorological forcings. Theimportance of accurate forcings for large-scale land sur-face modeling efforts has been demonstrated previ-ously (Berg et al. 2003; Fekete et al. 2004; Nijssen andLettenmaier 2004). Results from the NLDAS project(K. E. Mitchell et al. 2004) indicated that first-ordererrors in the land surface simulations were due to in-accurate specification of the forcings and especially in

precipitation (Robock et al. 2003; Pan et al. 2003).Other studies have shown the sensitivity of the landsurface to the atmospheric forcings and especially pre-cipitation (Berg et al. 2003; Fekete et al. 2004; Sheffieldet al. 2004). The conclusion is that accurate forcings arenecessary to provide accurate land surface simulationswhen compared to observations. The implication is thatthe use of sufficiently accurate forcings for land surfacemodeling in regions of sparse land surface observationswill provide a suitable surrogate.

The availability of near-surface meteorological ob-servations is not pervasive across all global areas andcertainly not at the spatial and temporal resolutionsthat are required by land surface hydrologic models formost hydrologic applications. Coupled with the lack oftemporal extent and consistency in the majority of ob-servations, the development of forcing datasets usingobservations alone is unsatisfactory. With the increas-ing availability of remote sensing products, the prospectfor the future is more promising, although this does nothelp in the development of long-term retrospectivedatasets that are required for extracting informationabout climate variability. In the global context, the useof atmospheric reanalysis products may be the only al-ternative for providing near-surface meteorologicalforcings at high temporal resolution. In contrast to thelack of terrestrial observations, the relative wealth ofobservations of the atmosphere and sea surface has al-lowed the emergence of a number of global, long-term,reanalysis datasets, such as the National Centers forEnvironmental Prediction–National Center for Atmo-spheric Research (NCEP–NCAR; Kalnay et al. 1996;Kistler et al. 2001), the 40- and 15-yr European Centrefor Medium-Range Weather Forecasts [(ECMWF)ERA-40 and ERA-15; Gibson et al. 1997], the NCEP–Department of Energy (DOE; Kanamitsu et al. 2002),and the National Aeronautics and Space Administra-tion Data Assimilation Office (NASA DAO; Schubertet al. 1993) reanalyses. These products are constructedusing “frozen” versions of numerical weather predic-tion and assimilation systems that ingest a variety ofatmospheric and sea surface observations to providelong-term, continuous fields in time and space of atmo-spheric (and land surface) variables. Although thesemodel-derived fields may not be perfect, they are self-consistent and are used by many to force models of theland surface water and energy balances.

The power of reanalyses is their consistent and co-herent framework for ingesting in situ and remote sens-ing data into a time- and space-discretized representa-tion of the global land, oceans, and atmosphere, in away that is essentially impossible to achieve directlyfrom observations. Reanalysis has been suggested as an

1 JULY 2006 S H E F F I E L D E T A L . 3089

alternate approach to the problem of estimating thesurface water balance, yet the reanalysis land surfaceproducts have many problems, including (i) the datathat are assimilated are primarily atmospheric profilesof moisture, temperature, and other variables, and few(if any) land surface data are assimilated, resulting inthe fact that they better represent variables, like atmo-spheric moisture and large-scale circulation than landsurface variables like soil moisture and snow–watercontent; (ii) the land surface is forced by precipitationthat is essentially a model output product so that errorsin the model representation of precipitation (Janowiaket al. 1998; Trenberth and Guillemot 1998; Serreze andHurst 2000), which can be quite large, are translatedinto errors in land surface fields like evapotranspira-tion, runoff, and soil moisture (e.g., Lenters et al. 2000;Maurer et al. 2001); and (iii) the effects of “nudging”the land surface to avoid drift have the effect of creat-ing unrealistic soil moisture and of biasing (by largeamounts in many cases) water budget flux terms (Bettset al. 1998, 2003a,b; Maurer et al. 2001; Roads et al.2002a,b).

The effect that these biases have on land surface pro-cesses has to be addressed for these products to be ofuse as forcings in modeling studies. The results of Berget al. (2003), who tested bias correction of the ECWMFreanalysis over North America, suggest that modelersusing reanalysis products for forcing LSMs should con-sider a bias reduction strategy for their input forcings.Also, Sheffield et al. (2004) showed that systematic bi-ases in reanalyses filter down into the modeled landsurface fluxes and states. Ngo-Duc et al. (2005) foundthat precipitation biases in the NCEP–NCAR reanaly-sis were responsible for significant errors in modeledstreamflow for continental-scale basins. Nevertheless,the results of such studies have shown that there is greatpotential for using hybrid datasets that combine re-analysis with observation-based datasets to remove bi-ases. This approach retains the consistency and conti-nuity of the reanalysis but constrains it to the best avail-able observation datasets, which are generally availableat coarser resolutions and reduced spatial and temporalextents.

This paper describes the development of a long-term,global dataset of near-surface meteorology that can beused to force models of the land surface water andenergy budgets. Reanalysis products are combined witha suite of observation-based global datasets that areused to correct for biases in the monthly mean valuesand intramonthly statistics of the reanalysis and fordownscaling in time and space to scales relevant forhydrologic applications. The dataset has global cover-age over the extrapolar land surface (i.e., excluding

Antarctica) at a 1.0° spatial resolution and a 3-hourlytime step for 1948–2000.

Previously, a number of studies have developedlarge-scale, long-term datasets of a similar nature.However, these have been limited to smaller domains(e.g., Maurer et al. 2002), and/or shorter time periods[e.g., Levis et al. 1996; Nijssen et al. 2001b; Interna-tional Satellite Land Surface Climatology Project(ISLSCP) I, Meeson et al. 1995; ISLSCP II, Hall et al.2005; second Global Soil Wetness Project (GSWP2),Dirmeyer et al. 2005], or have been implemented glo-bally, but at coarser spatial and temporal resolutions(e.g., Levis et al. 1996; Nijssen et al. 2001b; Ngo-Duc etal. 2005). This dataset represents an improvement overthese products in terms of higher spatial and temporalresolution and global coverage and through the imple-mentation of a number of enhancements in addition tocorrecting monthly biases and accounting for topo-graphic effects. These enhancements include (i) adjust-ments to precipitation for gauge undercatch; (ii) tem-poral and spatial disaggregation of precipitation anddownward solar radiation, accounting for observed sub-grid and diurnal variability statistics; (iii) adjustment torain day frequencies to match observed statistics; and(iv) trend correction and probability-weighted scalingfor biases in downward short- and longwave (SW andLW, respectively) radiation.

2. Datasets

The forcing dataset is based on the NCEP–NCARreanalysis, which includes near-surface meteorologicalvariables from 1948 to the present. This time periodprovides the length of data necessary to infer the vari-ability of the land surface water and energy budgets atup to multidecadal time scales. Alternative sources ofreanalysis data are available (including the NCEP–DOE and ERA-40 products) that have been shown tobe more accurate, in general, than the NCEP–NCARreanalysis. However, the NCEP–NCAR reanalysis of-fers the benefits of a long time period and ongoingproduction that may offset any potential deficienciesthat the bias correction methodology cannot address.Even if the ERA-40 or NCEP–DOE reanalysis hasbeen used, the comparisons would reveal any biasesthat exist in these products and the correction methodscould easily be applied.

The reanalysis data are combined with a suite ofglobal, observation-based datasets of precipitation,temperature, and radiation. Table 1 summarizes thecontributing datasets that are used in the developmentof the forcing dataset and these are described in moredetail in the following sections. These observation-based datasets are generally available at coarser tem-


poral resolutions (e.g., monthly). The reanalysis is es-sentially used to downscale these observation datasetsto the subdaily temporal scale necessary for land sur-face modeling. In contrast, the observation-baseddatasets are generally available at higher spatial reso-lutions and are used to downscale the reanalysis inspace. Thus, a hybrid forcing dataset is formed by usingthe submonthly variability in the reanalysis with biascorrections made on a monthly scale.

a. NCEP–NCAR reanalysis

The NCEP–NCAR reanalysis (hereafter referred toas the NCEP reanalysis) is a retrospective global analy-sis of atmospheric and surface fields extending from1948 to the near present (Kalnay et al. 1996; Kistler etal. 2001). Available observations are assimilated into aglobal atmospheric spectral model implemented at ahorizontal resolution of T62 (approximately 2.0°) andwith 28 sigma vertical levels. The reanalysis is createdusing a frozen version of the data assimilation system,although assimilated observations are subject to chang-ing observing systems. Consistent gridded output fieldsare generated continuously in time and are classifiedaccording to how they are determined and on their re-liability. Class “A” variables are strongly influenced byassimilated observations and are therefore regarded asbeing the most reliable fields (e.g., upper air tempera-tures and geopotential height). Less reliable are class“B” variables (moisture, divergent wind, and surfaceparameters), which are influenced by observations andthe model. Class “C” variables (surface fluxes and heat-ing rates) are completely determined by the model andas such, are the least reliable. Precipitation is classifiedas a class C variable.

b. Climatic Research Unit monthly climate variables

The Climatic Research Unit (CRU) product is a 0.5°gridded dataset of monthly terrestrial surface climatevariables for the period of 1901–98 (New et al. 1999,2000) and updated to 2000 by T. D. Mitchell et al.

(2004, manuscript submitted to J. Climate, hereafterMCJHN). The spatial coverage extends over all landareas including oceanic islands but excluding Antarc-tica. Fields of monthly climate anomalies, relative to a1961–90 climatology, were interpolated using thin-platesplines from surface climate data. The anomaly gridswere then combined with the 1961–90 climatology, re-sulting in grids of monthly climate over the full period.Primary variables (precipitation, mean temperature,and diurnal temperature range) are interpolated di-rectly from station observations. The secondary vari-ables (including rain day frequency and cloud cover)are interpolated from merged datasets comprising sta-tion observations and, in regions without station data,from synthetic data estimated using predictive relation-ships with the primary variables.

c. Global Precipitation Climatology Project dailyprecipitation

The Global Precipitation Climatology Project (GPCP)daily, 1997–present, 1.0° precipitation product (Huff-man et al. 2001) is based on a combination of estimatesfrom a merged satellite IR dataset over 40°N–40°S anda rescaling of the Susskind et al. (1997) Television In-frared Observation Satellite (TIROS) Operational Ver-tical Sounder (TOVS) estimates at higher latitudes.Both contributing estimates are scaled to match theGPCP version 2 monthly satellite–gauge dataset totals(Huffman et al. 1997). Rain day frequencies of the IR-based estimate are adjusted to match data from theSpecial Sensor Microwave Imager (SSM/I) retrieval.The TOVS-based rain day frequencies are adjusted tothe IR-based estimate at 40°N and 40°S separately.

d. Tropical Rainfall Measuring Mission 3-hourlyprecipitation

The Tropical Rainfall Measuring Mission (TRMM) isa joint mission between NASA and the Japan Aero-space Exploration Agency (JAXA). The TRMM satel-lite was launched in November of 1997 and covers the

TABLE 1. Summary of datasets used in the construction of the forcing dataset. The temporal resolutions given here are those usedin this study but original data may be available at finer temporal resolutions. Variables are precipitation (P), surface air temperature(T ), downward shortwave radiation (SW), downward longwave radiation (LW), surface air pressure (Ps), specific humidity (q), windspeed (w), and cloud cover (Cld).

Dataset Variables Temporal coverage Spatial coverage Source

NCEP–NCAR reanalysis P, T, SW, LW, q, Ps, w 1948–present, 6 hourly Global, �2.0° � 2.0° Kalnay et al. (1996)CRU TS2.0 P, T, Cld 1901–2000, monthly Global land excluding

Antarctica, 0.5° � 0.5°MCJHN

GPCP P 1997–present, daily Global, 1.0° � 1.0° Huffman et al. (2001)TRMM P Feb 2002–present,

3 hourly50°S–50°N, 0.25° � 0.25° Huffman et al. (2003)

NASA Langley SRB LW, SW 1983–95, monthly Global, 1.0° lat �1.0°–120° lon

Stackhouse et al. (2004)


Tropics between approximately 40°S and 40°N latitude.A number of experimental, real-time datasets based onthe TRMM products and other satellite sources are cur-rently available (Huffman et al. 2003), including the3B42RT product, which is a merger of the 3B40RT and3B41RT products. The 3B40RT product is a merger ofall available SSM/I and TRMM Microwave Imager(TMI) precipitation estimates. The SSM/I data are cali-brated to the TMI using separate global land and oceanmatched histograms. The 3B41RT product consists ofprecipitation estimates from geostationary IR observa-tions using spatially and temporally varying calibrationsby the 3B40RT product.

e. NASA Langley monthly surface radiation budget

The NASA Langley Research Center product isavailable from 1983–1995 (Gupta et al. 1999) with anextension to 2001 being planned (Stackhouse et al.2004). The primary data sources are satellite data fromthe International Satellite Cloud Climatology Project(ISCCP) C1 product (Rossow and Schiffer 1991) andfrom the Earth Radiation Budget Experiment (ERBE;Barkstrom et al. 1989). The C1 data provide cloud pa-rameters derived from a network of geostationary sat-ellites and from NOAA’s polar orbiters, along withtemperature and humidity profiles from TOVS, on a2.5° equal-area global grid and a 3-hourly time resolu-tion. Monthly average clear-sky planetary albedos usedfor deriving surface albedos over snow/ice-free land ar-eas were obtained from ERBE data. Two versions areavailable for short- and longwave radiation. First, thesurface radiation budget (SRB)-SW and SRB-LWproducts are derived using the algorithms of Pinker andLaszlo (1992) and Fu et al. (1997), respectively. Second,the SRB-QCSW [quality check (QC)] and SRB-QCLWproducts are derived using the algorithms of Darnell etal. (1992) and Gupta et al. (1992), respectively. Com-parison of these products with surface measurementshas indicated that no one product is superior globally.For example, the SW product underestimates short-wave radiation over the higher elevations of the Ti-betan Plateau and western China whereas the QCSWproduct does not, although comparisons undertaken forthe GSWP2 over North America showed that the SWproduct performed better (GSWP2 forcing data avail-able online at http://www.jamstec.go.jp/frcgc/research/p2/masuda/gswp/b1alpha.html). Given these prelimi-nary analyses, the SRB-QCSW and SRB-LW productsare used in this study.

3. Development of the forcing dataset

The development of the forcing dataset has pro-gressed through a number of stages in terms of the

spatial and temporal resolution and the sophisticationof the correction methods. This has resulted in a num-ber of intermediate products at coarser spatial and tem-poral resolutions. To perform calculations of the landsurface water and energy cycles, land surface models ingeneral require subdaily time series of the followingnear-surface atmospheric variables: precipitation, airtemperature, downward short- and longwave radiation,surface pressure, specific humidity, and wind speed. Ini-tially, the reanalysis variables were bilinearly interpo-lated from their native resolution of approximately 1.9°latitude � 1.875° longitude to a 2.0° regular grid with con-sideration for changes in elevation (see section 3b).This grid is commensurate with the observation-baseddatasets. Next, corrections are made to the daily pre-cipitation statistics. All variables are then downscaledin space to 1.0° resolution (again with corrections forchanges in elevation) and downscaled in time to a 3-hourlytime step. Finally, biases at the monthly scale are re-moved. The following sections describe in detail thevarious stages in the development of the forcing dataset.

a. Correction of the reanalysis rain day anomaly

A high-latitude anomaly in the rain day statistics ex-ists in the NCEP reanalysis in the winter months of theNorthern Hemisphere (Cullather et al. 2000; Serrezeand Hurst 2000; Sheffield et al. 2004). The anomalyresults from the use of a simplified approximation formoisture divergence in the atmospheric forecast modelused in the reanalysis. This results in a spurious wave-like pattern in the monthly rain day statistics that ismost noticeable in the Northern Hemisphere winter athigh latitudes (Fig. 1). The anomaly filters down intoland surface states when the precipitation is used toforce a land surface model (Sheffield et al. 2004). Thisstudy also showed the sensitivity of the land surface tothe monthly rain day statistics. Using various estimatesof rain day statistics but the same monthly totals toforce a land surface model resulted in large differencesin estimated water balance components (up to 9% errorin global average evaporation and 17% in runoff, withhigher values at the continental and regional scale).The conclusion is that it is vital to use the best estimatesof not only monthly total precipitation but also monthlyrain day statistics to achieve accurate simulations of theland surface water budget. A correction to the rain daystatistics is described in detail in Sheffield et al. (2004)and a brief description is given here. The correctioninvolves resampling the daily precipitation data tomatch the statistics of observation-based terrestrialdaily precipitation datasets [CRU, GPCP, and a 15-yrgauge-based dataset developed by Nijssen et al.(2001b)]. To ensure consistency in the related meteo-


rological variables, these are also resampled for thesame days for which the precipitation was resampled.Figure 1 shows the correction of the NCEP precipita-tion using the CRU dataset. In addition to correctingthe high-latitude rain day anomaly, differences that oc-cur elsewhere are also removed. For example, in theTropics, the high number of rain days in the NCEPdataset is reduced to the levels found in the CRUdataset. One side effect of this correction method is thatspatial consistency at the daily time scale is not main-tained because the correction is carried out indepen-dently on each grid cell. Sheffield et al. (2004) foundthat the effect of this on the large-scale terrestrial waterbalance was small compared to that resulting from thecorrection of the precipitation frequencies.

b. Spatial downscaling

1) PRECIPITATION

Daily precipitation (corrected for monthly rain daybiases) was downscaled from 2.0° to 1.0° resolution us-

ing a probabilistic approach based on relationships be-tween precipitation intensity and grid cell fractionalprecipitation coverage. Precipitation varies consider-ably in space, especially at daily time scales, and it hasbeen recognized that the land surface is sensitive to thisvariability (Johnson et al. 1993; Eltahir and Bras 1993)and that the effects on the atmosphere through en-hanced feedback can be significant (Hahmann 2003). Ingeneral, low-intensity, large-area precipitation will tendto increase evaporation and infiltration compared tohigh-intensity, localized precipitation that will result inincreased runoff production through infiltration excess.

For downscaling the precipitation data to 1.0° reso-lution, it is of interest to know the fractional wettedarea within the 2.0° grid cell and the distribution ofprecipitation intensities among the 1.0° grid cellswithin. Fractional area is seasonally and geographicallyvariable (Gong et al. 1994) and depends, among otherfactors, on storm type, grid resolution, and temporalscale (Eltahir and Bras 1993). Figure 2 shows an ex-ample of the scaling behavior of precipitation fractional

FIG. 1. Average January precipitation statistics for the NCEP and corrected datasets: (a) number of precipitation days and (b) totalprecipitation (mm day�1) from the NCEP dataset, showing the spurious wavelike pattern in Northern Hemisphere high latitudes; (c),(d) same as in (a), (b), but as corrected by Sheffield et al. (2004) using data from the CRU TS2.0 global 1901–2000 climate dataset ofMCJHN.


Fig 1 live 4/C

area for the 0.25° TRMM and 1.0° GPCP datasets. Val-ues for a range of spatial resolutions are shown suchthat the scale is relative to the dataset resolution. Frac-tional area of the TRMM data drops off rapidly withincreasing scale but tends toward a threshold value (atlarger scales) that appears to be seasonal. At the scaleof the forcing dataset (1.0°), the fractional coverage ofthe TRMM data is on average much less than 1 (fullcoverage). This implies that downscaling by simply ap-plying the 2.0° grid cell average precipitation to the four1.0° cells may be inappropriate in terms of representingthe spatial variability of wet and dry areas, with subse-quent effects on the land surface hydrology. The GPCPdata show similar relative scaling behavior (windowsizes larger than 4.0° had a limited number of land cellsand so were not included) that may indicate some formof self-similarity. Because of this and their multiyearglobal coverage, the GPCP data were considered suit-able for downscaling the NCEP data.

The 2.0° daily data are downscaled using a probabi-listic approach that relates the fractional area of pre-cipitation with the precipitation intensity at 2.0°. FromBayesian theory, the probability of occurrence of pre-cipitation within a 2.0° grid cell with fractional coverage

(A) for a given grid cell average precipitation intensity(I) can be written as

p�A|I� �p�I|A�p�A�

p�I�, �1�

where p(I|A) is the conditional probability of an inten-sity I given a fractional area A. Probabilities for eachterm on the right-hand side of Eq. (1) are generated foreach month and grid cell using data from the GPCPdaily dataset, which has global and multiple year cov-erage. The NCEP daily data are then downscaled to1.0° by sampling at random from the resultant condi-tional probability distribution p(A|I) to determine thespatial coverage of precipitation in terms of the numberof 1.0° grid cells.

This disaggregation method was validated by recon-structing the 1.0° GPCP dataset from a 2.0° aggregatedversion using the probability distributions from Eq. (1).Figure 3 shows an example of the spatial statistics forthe GPCP dataset and three different reconstructedversions over the North American continent. Thesethree versions were created, respectively, by (i) distrib-uting the 2.0° precipitation value uniformly over all 1.0°

FIG. 2. Fractional area of precipitation as a function of spatial scale for mild, midlatitude climateregions: (a) mean and (b) standard deviation for January; (c), (d) same as in (a), (b), but for July. Solidlines are the TRMM data; dashed lines are the GPCP data. The spatial scale is relative to the resolutionof the precipitation datasets (TRMM � 0.25°; GPCP � 1.0°).


cells within; (ii) using the probabilistic approach to de-termine the fractional area of precipitation (number of1.0° cells) within a 2.0° cell and distributing the 2.0° gridcell precipitation uniformly within these cells; and (iii)as for (ii), but weighting the precipitation among thewet 1.0° grid cells based on the precipitation in neigh-boring 2.0° cells. This final method assumes that pre-cipitation occurrence has some spatial coherence andthat the wet cells are deemed to have some simple con-nectivity with neighboring regions of precipitation. Fig-ure 3 indicates that using the distributed method is littlebetter than using a uniform approach, although the ef-fect on land surface states may be quite different. How-ever, weighting the distribution of the precipitationgives values of spatial variability that are consistentwith the original GPCP data, although slightly higher.Similar results apply for other regions across the globe.The local autocorrelation of the original and recon-structed datasets was also calculated at various lagtimes to see whether the correction methods preservedthe temporal characteristics of precipitation at eachgrid cell (Fig. 4). In general, the errors decrease withincreasing lag time for all three methods. Again, theweighted method shows the least error, except for Eu-

rope and North America where all methods performessentially the same at longer lag times.

2) METEOROLOGICAL VARIABLES

The other meteorological variables (downwardshortwave and longwave radiation, surface pressure,specific humidity, and wind speed) were disaggregatedfrom 2.0° to 1.0° using bilinear interpolation but withadjustments for differences in elevation between thetwo grids. The effects of elevation on near-surface me-teorology have been well documented and the differ-ence in elevation between the two grids, as shown inFig. 5, can be significant. The differences are mostprominent in the foothills of mountain ranges whereelevation may change by a few thousand meters withina 2.0° grid cell. Maximum differences are approxi-mately 3000 m in the Himalayas, 1300 m in U.S. Rock-ies, and up to 3700 m in the Andes. To account for thedifferences in elevation, air temperature is first ad-justed to the new grid elevation using the environmen-tal lapse rate (6.5°C km�1). Following the methods ofCosgrove et al. (2003), which assume that the relativehumidity is constant to avoid the possibility of super-saturation, the specific humidity, surface air pressure,

FIG. 3. Average monthly distribution of the coefficient of variability for North America forthe original daily, 1.0° GPCP dataset and three datasets that were downscaled from a 2.0°aggregated version of the GPCP data to 1.0° using various downscaling methods. The uniformmethod assigns precipitation values uniformly to the higher-resolution cells. The distributedapproach uses a probabilistic method to determine the number of 1° grid cells within a 2° cellin which it is raining and distributes the 2° grid cell precipitation uniformly within these cells.The distributed with weighting method is the same as the distributed approach but weights theprecipitation among the 1° grid cells based on the precipitation in neighboring cells. Similarresults apply for the other continents.


and downward longwave radiation are also correctedfor elevation changes to ensure consistency. These cor-rections were applied whenever a dataset (reanalysisand observational) was interpolated from one grid toanother, whether for upscaling or downscaling, usingthe following method: first, the data were elevation ad-justed to sea level (0.0-m elevation) on their native grid;then the data were interpolated to the new grid reso-lution and elevation adjusted to the topography of thenew grid. This ensures that the interpolation procedureis free of any elevation effects on the data. For theinterpolation between the 2.0° and 1.0° grids, these el-evation adjustments resulted in significant changes insome regions, with a maximum change of approxi-mately 25°C for temperature, 160 W m�2 for longwaveradiation, 0.013 g g�1 for specific humidity, and 38 KPafor surface air pressure.

c. Temporal downscaling

1) PRECIPITATION

The diurnal variation of precipitation is generally sig-nificant over land areas, especially during the summermonths where diurnal amplitudes can be greater than

50% of the daily mean value (Dai 2001). High temporalresolution precipitation data (6 hourly or higher) arenecessary to describe the diurnal cycle and are desir-able for a multitude of hydrologic applications. Landsurface hydrological processes are governed not only bythe total amount of precipitation but also by the tem-poral structure of the precipitation, that is, the stormduration, intensity, and interstorm length. Marani et al.(1997) showed the effect of the temporal structure ofprecipitation on land surface hydrological processes tobe considerable because of the nonlinear processes in-volved in partitioning precipitation. In the context ofremotely sensed precipitation, which may suffer fromundersampling of the diurnal cycle, similar conclusionshave been reached (e.g., Soman et al. 1995; Salby andCallaghan 1997; Nijssen and Lettenmaier 2004). Yet theavailability of subdaily precipitation data is intermittentin time and space, whether from gauges, radar, or re-mote sensing, and thus downscaling is required forlarge-scale applications. Direct use of the highest-resolution NCEP precipitation data (6 hourly) is un-warranted because it is acknowledged as being unreli-able at less than monthly scales (Kalnay et al. 1996) andmoreso given the biases in the rain day statistics as

FIG. 4. Rmse over the six continents in autocorrelation for various daily lag lengths between theoriginal daily, 1.0° GPCP dataset and three datasets that were downscaled from a 2.0° aggregated versionof the GPCP data to 1.0° using various downscaling methods.


described in section 3a. The biases in storm durationand storm frequency have been partially accounted forthrough the correction of monthly rain day frequencies(section 3a). However, to obtain reasonable estimatesof the diurnal cycle of precipitation, disaggregation ofthe daily values to a 3-hourly time step is necessary.

Temporal downscaling of precipitation has been at-tempted by many authors using a variety of techniques,including the use of probability distributions of precipi-tation statistics (e.g., Hershenhorn and Woolhiser 1987;Connolly et al. 1998), multifractal cascade methods(e.g., Olsson 1998; Gütner et al. 2001), and rectangular-pulses, stochastic rainfall generators (e.g., Bo et al.1994; Cowpertwait et al. 1996). Here, a simple stochas-tic sampling approach is used based on 3-hourly pre-cipitation distributions extracted from the TRMM real-time dataset. This product provides one of the fewlarge-scale, observation-based, gridded precipitationdatasets at subdaily resolution. The original TMI datasuffer from undersampling of the diurnal cycle becauseof orbit characteristics and can only adequately de-scribe the diurnal cycle at coarse time and space reso-lutions (Negri et al. 2002). However, the real-timeproduct used here combines the TMI data with IRdata to produce near-continuous coverage in time andspace. Other alternative datasets could be used, includ-ing the TRMM-based Precipitation Estimation fromRemotely Sensed Information Using Artificial Neural

Networks (PERSIANN) analysis (Hsu et al. 1997) andmodel-based products such as those from NASA’sGoddard Earth Observing System (GEOS), NCEP’sGlobal Data Assimilation System (GDAS), and theECMWF.

The precipitation is downscaled from the daily NCEPproduct (with corrected rain day frequencies; section3a) to a 3-hourly time step using a probabilistic ap-proach based on sampling from the remote sensing–based TRMM dataset. The TRMM dataset consists of3-hourly data covering the latitude band 50°S–50°N(see section 2). Monthly joint probability density func-tions (PDFs) of 3-hourly and daily precipitationamounts are derived from this dataset for each 1.0° gridcell using information from the surrounding 2.0° win-dow. Three-hourly precipitation amounts are thensampled at random from these distributions for eachNCEP daily total and then the eight 3-hourly values ineach day are scaled to match this daily total. For regionsoutside of 50°S–50°N where TRMM precipitation dataare not available, it is assumed that the PDFs are uni-form across regional climate zones. Thus, joint PDFswere created for each continent and climate zone[based on the Koppen climate classification; see Critch-field (1983)] and these were used to downscale the dailyNCEP data outside of 50°S–50°N, within the same cli-mate zone and continent. This method was not feasiblefor regions within polar climate zones because there are

FIG. 5. Difference in elevation (m) between the 2.0° and 1.0° grids. Elevation adjustments are madeto air temperature, surface pressure, specific humidity, and downward longwave radiation wheneverdatasets are interpolated between grids.


Fig 5 live 4/C

no such regions within the 50°S–50°N latitude band. Inthis case, the probability distributions derived fromcold climate zones were assumed to be representativeof polar climates in the same continent.

The disaggregation method forces the statistics of thedisaggregated data to match those of the TRMMdataset, while retaining the NCEP daily totals. Themethod was validated by recreating the TRMM prod-uct from its daily totals and indicated good perfor-mance in recreating the mean monthly diurnal cycle fordifferent seasons and regions. The application of thedisaggregation method is, however, dependent on theaccuracy of the PDFs in representing actual diurnalcycles, and so is limited by the amount of data thatcontribute to them. The TRMM dataset used here haslimited temporal coverage and may itself contain biases(Gottschalck et al. 2005). Updates from the TRMMreal-time product and additional data from the retro-spective version that started in 1998 will be added in thefuture to increase confidence in the PDFs and thus, theresulting disaggregated values. Data from gauge-baseddatasets (Dai 2001) that may be more reliable at re-gional scales could also be used.

2) METEOROLOGICAL VARIABLES

The meteorological variables are simply downscaledfrom 6-hourly to 3-hourly resolution using linear inter-polation. It is assumed that the diurnal cycle of thesevariables is represented adequately in the reanalysis,although the diurnal temperature range is adjusted toremove biases at the monthly scale [see section 3d(2)].No attempt is made to adjust these variables to makethem consistent with the disaggregated 3-hourly pre-cipitation, as the relationships between precipitationand other meteorological variables are often weak.Downward solar radiation is interpolated in regard tothe solar zenith angle to give a more realistic represen-tation of the diurnal path of the sun. The type of re-analysis variable (downward shortwave and longwaveradiation are time average values; air temperature,pressure, humidity, and wind speed are instantaneousvalues) is taken into account during the interpolationand all variables are converted into time average val-ues.

d. Monthly bias corrections

As described in the introduction, systematic biasesare inherent in the NCEP reanalysis (and other re-analysis products) at the monthly and seasonal scale.These biases are seasonally and regionally variable andwill filter down into simulations of the land surface wa-ter and energy budgets. Adjustments are made to the

reanalysis data (after downscaling and elevation correc-tions) so that the mean monthly values match thosefrom available observation-based datasets. Adjust-ments are not made to the specific humidity, air pres-sure, and wind speed because global-scale, observation-based datasets for these variables do not exist.

1) PRECIPITATION

The NCEP reanalysis precipitation is completelygenerated by the atmospheric forecast model and assuch is acknowledged as being somewhat unreliable atthe submonthly and local scale (Kalnay et al. 1996),although it does reveal useful information at largerspace and time scales (Kalnay et al. 1996; Janowiak etal. 1998; Kistler et al. 2001). Biases in the NCEP re-analysis precipitation have been studied by many au-thors (e.g., Janowiak et al. 1998; Trenberth andGuillemot 1998; Serreze and Hurst 2000). Figure 6shows the time series of global and continental averageprecipitation for the NCEP reanalysis and CRU

FIG. 6. Annual time series of precipitation averaged over globaland continental land areas excluding Antarctica for the NCEPand CRU datasets. NCEP global mean precipitation � 2.2 mmday�1, CRU global mean precipitation � 2.0 mm day�1, and glob-al mean bias in NCEP precipitation � 0.19 mm day�1 (70 mmyr�1).


datasets. The NCEP dataset is biased by 0.193 mmday�1 over global land areas excluding Antarctica,which is equivalent to about 70 mm yr�1. Errors in theNCEP reanalysis precipitation at monthly scales trans-late into errors in land surface fields like evapotranspi-ration, soil moisture, and snow cover (Sheffield et al.2004). The effect on runoff generation has been inves-tigated by Ngo-Duc et al. (2005), who found that biasesin the NCEP reanalysis precipitation contributed thelargest errors in resultant large-basin river dischargewhen compared to biases in air temperature and radia-tion. To remove the biases in the NCEP product, thedaily values are scaled so that their monthly totalsmatch those of the CRU dataset before disaggregationto a 3-hourly time step as follows:

P*NCEP,3hr �PCRU,MON

PNCEP,MON� PNCEP,3hr, �2�

where the asterisk indicates a corrected value and thesubscripts indicate the data source (NCEP or CRU)and the temporal resolution (3 hourly, daily, or

monthly). Gauge-based precipitation measurementsare often subject to losses from wind and wetting lossesand due to solid precipitation (Goodison et al. 1998).Adam and Lettenmaier (2003) describe a global datasetof adjustment ratios that can be used for correctinggauge undercatch and can result in an increase in pre-cipitation of about 12% globally. These catchment ra-tios can be applied to precipitation climatologies or toindividual years in the reference period of the dataset(1979–98; see Adam and Lettenmaier 2003). For thisstudy, the monthly CRU precipitation dataset is ad-justed using these catchment ratios before being used toscale the NCEP daily totals.

Figure 7 shows the effect of the monthly bias correc-tions on the NCEP reanalysis precipitation. These ad-justments result in changes in global terrestrial precipi-tation (excluding Antarctica) of �8.8% (�0.19 mmday�1 or �70.3 mm yr�1) after scaling to the CRUmonthly values and �1.7% (�0.037 mm day�1 or 13.7mm yr�1) after also adjusting for gauge undercatch.Although the reduction in global precipitation by scal-ing to the CRU values is offset by the undercatch ad-

FIG. 7. Average DJF precipitation (mm day�1) for (a) NCEP, (b) NCEP scaled with the CRU dataset and adjusted for gaugebiases, (c) the difference between CRU and NCEP, and (d) the difference between (b) and the CRU dataset.


Fig 7 live 4/C

justment, there are substantial regional changes. Figure7c shows the December–February (DJF) biases in theNCEP dataset when compared to the CRU dataset.The largest biases in the reanalysis are over Greenland,the central and southeast United States (for DJF only),and in northern India during June–August (JJA; notshown). There are large positive biases in mid- and highnorthern latitudes during the summer (not shown),most notably in Canada and Alaska, central Europe,and throughout Eurasia to China. In the Tropics, thebiases are spatially variable and seasonally dependent.For example, in Amazonia, the biases in the NCEPprecipitation tend to be negative in the southwest dur-ing September–November (SON) and DJF and shiftnorthward during the other part of the year. Con-versely, large positive biases generally occur in the eastin an opposite pattern. This indicates the poor repre-sentation of the seasonal cycle of the tropical moisturepatterns in the NCEP dataset (Trenberth and Guille-mot 1998). Of note is the correction to the spuriouswavelike pattern in high northern latitudes as describedin section 3a. Figure 7d shows the mean DJF map ofadjustments for gauge biases, which are generally posi-tive, with the largest increases in Greenland, the centraland northeast United States, parts of northern Eurasia,and scattered regions in the Tropics.

2) TEMPERATURE

The NCEP air temperature is calculated from themodeled atmospheric variables, which are constrainedby upper air observations and surface pressure, but noassimilation of screen-level observations is carried out.It is a B-class variable (Kalnay et al. 1996) in the re-analysis classification, as it is strongly influenced by themodel parameterization of surface energy fluxes. Kal-nay and Cai (2003) compared NCEP surface air tem-perature with station-based observations over theUnited States and found that the interannual variationwas well represented, although the upward trend overtime was significantly less than that observed. Similarresults were found by Kistler et al. (2001) at globalscales. Simmons et al. (2004) looked at continental andregional scales and again found good agreement withinterannual variability and generally lower warmingtrends in the Northern Hemisphere but distinct andprobably incorrect regions of cooling in Australia andsouthern South America.

Figure 8 shows the mean annual time series of 2-m airtemperature for the NCEP and CRU datasets for glob-al and continental land areas excluding Antarctica. Theaverage annual global bias in the NCEP dataset is�0.56°C. Comparison of the seasonal average air tem-peratures shows much larger regional and seasonal dif-

ferences (see Fig. 9). Most notably, in Siberia and west-ern Canada, and Alaska in the Northern Hemispherewinter, biases in the NCEP reanalysis can reach in ex-cess of 5°C. Low biases are evident in the Himalayanrange and Greenland, again of the order of 5°C, withsmaller biases throughout the Tropics and scattered ar-eas in northern Africa and central Asia. Biases in airtemperature can be directly linked to changes in theland surface water budget through modifications ofevaporation and thus soil moisture (e.g., Qu et al.1998). To remove these biases, the NCEP temperaturedata were adjusted to match the CRU monthly valuesby shifting the NCEP values by the difference betweenthe NCEP and CRU monthly average values:

T*NCEP,3hr � TNCEP,3hr � �TCRU,MON � TNCEP,MON�.

�3�

In addition to scaling the 3-hourly values so that theirmonthly mean matched the CRU monthly values, thediurnal cycle of temperature for each day was scaled sothat the monthly mean diurnal temperature range

FIG. 8. Same as in Fig. 6, but for air temperature (°C). NCEPglobal mean air temperature � 7.6°C, CRU global mean airtemperature � 8.1°C, and global mean bias in NCEP air tempera-ture � �0.6°C.


(DTR) matched the CRU monthly DTR values but thedaily average value was unchanged as follows:

T*NCEP,3hr � T*NCEP,DAILY �DTRCRU,MON

DTRNCEP,MON

� �T*NCEP,3hr � T*NCEP,DAILY�. �4�

Adjustments were made to the specific humidity, sur-face air pressure, and downward longwave radiation asoutlined in section 3b(2) to make them consistent withthe new temperature values.

3) DOWNWARD SHORT- AND LONGWAVE SURFACE

RADIATION

Incoming shortwave radiation incident at the earth’ssurface is the primary energy source for the land sur-face and drives evapotranspiration and snowmelt.Therefore, accurate specification of these forcing fluxesis essential for land surface modeling. Snow accumula-tion and melt are particularly sensitive to incominglongwave radiation (e.g., Schlosser et al. 2000), al-though Morrill et al. (1999) found that energy and wa-

ter budgets were not sensitive to the diurnal cycle oflongwave radiation. Downward surface short- and long-wave radiation are completely predicted by the NCEPreanalysis forecast model, and, as with precipitationand air temperature, contain systematic biases at sea-sonal time scales. Local-scale comparisons indicate bi-ases in both the long- and shortwave products that maybe systematic across geographic regions. Brotzge (2004)found that the NCEP dataset consistently overesti-mated downward surface shortwave radiation by 17%–27% over 2000–01 when compared to two OklahomaMesonet sites. Longwave radiation was underesti-mated; but to a lesser degree. Betts et al. (1996) foundsimilar results when making a comparison with datafrom the First ISLSCP Field Experiment (FIFE) for1987 and concluded that these problems are generallyattributed to the NCEP model atmosphere being tootransparent and to too few clouds being produced,which may be systematic of large-scale atmosphericmodels in general. At larger scales, comparisons withremote sensing–based data have revealed large-scalebiases. For example, Berbery et al. (1999) found posi-

FIG. 9. Average seasonal difference in near-surface air temperature between the NCEP and CRU datasets (°C).


Fig 9 live 4/C

tive biases of 25–50 W m�2 over the United States whencompared to the Geostationary Operational Environ-mental Satellite (GOES)-based product of Pinker andLaszlo (1992).

Several global SRB datasets have been developed inrecent years including the Global Energy and WaterCycle Experiment (GEWEX)–NASA Langley SurfaceRadiation Budget Project 1984–95 product (Stackhouseet al. 2004) and the ISCCP global 1983–2000 product(Zhang et al. 2004). These datasets provide surfaceshort- and longwave fluxes that have been validatedagainst ground measurements. The latest version of theNASA Langley SRB product (release 2.0) is used here.Comparisons with ground-based measurements fromthe Baseline Surface Radiation Network (BSRN) indi-cate that errors are within measurement uncertainty. Acomparison of the SRB and NCEP downward long-wave data is shown in Fig. 10 as seasonal averages for1984–95. The mean bias in the NCEP dataset is 15.8 Wm�2 over global land areas excluding Antarctica. Thereare large regional biases of the order of 50–100 W m�2

across the Sahara, Middle East, central Asia, theAndes, and to a lesser extent, in the western United

States and Australia. The biases tend to be highest inthe Northern Hemisphere spring and summer. Thecomparison of downward shortwave radiation is sum-marized in Fig. 11. The mean bias in the NCEP short-wave data is �41.5 W m�2 over global land areas. Thebiases tend to be larger in the spring and summer ofeach hemisphere in mid- to high latitudes. These ex-ceed �60 W m�2 across the northern United States andCanada, northern Europe, Siberia, and central Asiaduring the boreal summer and in the southern part ofSouth America in the austral summer. In the Tropics,there is reasonable agreement throughout the year.

Analysis of station data has shown that shortwaveradiation at the earth’s surface has decreased over largeregions during 1960–90 (Gilgen et al. 1998), which hasbeen attributed to increases in cloud cover. More re-cently, studies of station data (Wild et al. 2005) andsatellite measurements (Pinker et al. 2005) indicate thatthese downward trends have reversed over the past de-cade or so, possibly due to reductions in aerosols. How-ever, the trend in global terrestrial shortwave radiationfrom the reanalysis shows a spurious upward trend (Fig.12b). Therefore, the reanalysis shortwave radiation is

FIG. 10. Average seasonal difference in downward longwave radiation (W m�2) between the NCEP and SRBdatasets for 1984–94.


Fig 10 live 4/C

adjusted so that first, systematic biases are removed atthe monthly scale so that it matches the mean of theSRB data for 1984–94, and second, trends over the full50-yr period are consistent with observations.

Using the relationship between cloud cover and sur-face downward shortwave radiation (Thornton andRunning 1999), a new time series of radiation is con-structed that is consistent with observed trends. A lin-ear regression was developed at each grid cell betweenthe monthly anomalies of reanalysis cloud cover andshortwave radiation. This relationship was then used topredict monthly anomalies of shortwave radiation fromobservation-based estimates of cloud cover anomaliesfrom the CRU dataset. The resultant time series wasthen converted to actual values whose monthly clima-tology over 1984–94 matched that of the SRB dataset.This was done by subtracting the mean monthly clima-tology for 1984–94 from the time series of anomaliesand then adding the mean climatology of the SRBdataset. In this way, the new time series is consistentwith the SRB data over the limited period of overlapbut imposes the long-term trends as derived from ob-

served cloud cover. The NCEP 3-hourly values are thenscaled so that their mean values match this newmonthly time series as follows:

SW*NCEP,3hr �SWSRB�CRU,MON

SWNCEP,MONSWNCEP,3hr, �5�

where the subscript SRB�CRU indicates the time se-ries of monthly SW values derived from the CRU cloudcover and scaled to the SRB dataset.

Downward longwave radiation is bias corrected usinga probability matching method that scales the reanaly-sis monthly values to match the mean and variability ofthe SRB values but retains the year-to-year variation ofthe NCEP data. Figure 12c shows no apparent globaltrend in the NCEP data, which is consistent with sta-tion-based observations of long-term trends that arewithin the bounds of measurement error (Wild et al.2001). Therefore, no attempt is made to alter the long-term trends in the NCEP monthly values. The probabil-ity matching method replaces each of the NCEPmonthly mean values over 1948–2000 with a monthlyvalue from the SRB time series that has the same cu-

FIG. 11. Same as in Fig. 10, but for shortwave radiation.


Fig 11 live 4/C

mulative probability as the NCEP value. The cumula-tive probabilities were calculated from PDFs of theNCEP and SRB monthly time series. The new monthlytime series was then used to scale the NCEP 3-hourlyvalues as follows:

LW*NCEP,3hr �LWSRB,MON

LWNCEP,MONLWNCEP,3hr. �6�

Figure 12 shows the global mean monthly time seriesof downward short- and longwave radiation for theNCEP, SRB, and the scaled NCEP datasets.

4. Discussion and conclusions

The goal of this study is to provide a global dataset offorcings that has long temporal and global coverage andis consistent in time and space. In this respect, thedataset makes use of the latest available global meteo-rological datasets and combines them with state-of-the-art reanalysis to form a consistent, high-quality dataset.Nevertheless, an essential part of the development ofany dataset is validation against independent datasources, which will quantify the errors and known bi-ases and hopefully instill confidence in the use of the

FIG. 12. (a) Annual anomalies of global mean cloud cover for the CRU dataset (dark solidline) and cloud cover (solid line) and downward shortwave radiation (dashed line) from theNCEP dataset. (b) Annual time series of global mean downward shortwave radiation for theNCEP (solid line), SRB-QCSW (dark solid line), and NCEP (dashed line) corrected datasets.The corrected dataset has been scaled to be consistent with the SRB data and the long-termvariation of the CRU cloud cover. (c) Same as in (b), but for longwave radiation for theNCEP, SRB-LW, and NCEP corrected datasets. The corrected dataset has been scaled usingthe probability swap method to be consistent with the mean and variability of the SRB datawhile retaining the year-to-year variation of the NCEP dataset. Global means are calculatedover terrestrial areas excluding Antarctica.


data. This is a difficult task given the general lack oflarge-scale observations and the fact that potential vali-dation datasets are better utilized in the developmentof the forcing dataset to produce the highest-qualitydataset possible.

The intended application of the forcing dataset is forlong-term, large-scale modeling, where the focus of in-terest is on the variation of the land surface over sea-sonal to annual time scales and across regional and con-tinental space scales. Here, it is more important to en-sure that the statistics of the forcing data are correctrather than trying to replicate the finescale features ofthe historic record. For example, the forcing dataset isunlikely to recreate actual historic storm events partlybecause the correction of the daily precipitation fre-quencies may disrupt spatial coherence at the daily timescale. This is because the disaggregation and correctionmethodologies are designed to match the observed dataonly in a statistical sense while providing consistencyamong variables where possible, and any detrimentaleffects on the terrestrial water budget will be small atseasonal and regional scales.

The dataset can be evaluated by forcing a land sur-face model and comparing the resultant water and en-ergy fluxes and states with observations, such asstreamflow records, snow cover extent, and in situ soilmoisture measurements. Several studies have shownthat evaluating land surface model simulations overlarge areas requires a detailed examination of all as-pects of the modeling process (e.g., Nijssen et al.2001a,b; PILPS2-E Experiment, Nijssen et al. 2003;Bowling et al. 2003; the series of NLDAS papers, K. E.Mitchell et al. 2004). In addition to the errors in theforcings, there are also uncertainties in the land surfacemodel structure, physical parameterizations, and inputparameters (vegetation, soils, etc.) as well as in the ob-servations themselves. The relative contribution ofthese factors to the differences from observations isdifficult to discern without a detailed examination of allaspects of the modeling process and is a work in prog-ress.

Nevertheless, it is possible to evaluate the datasetagainst similar bias-corrected forcing products. This isdone next by comparing it to the GSWP2 forcingdataset, which uses a similar strategy to combine re-analysis with observations, although for a much shortertime period.

a. Comparison with GSWP2 forcing dataset

The goal of GSWP2 is to develop global datasets ofsoil moisture and other hydrologic variables from mul-tiple land surface models and to investigate the differ-ences and sensitivities of these models. The GSWP

forcing dataset has the same temporal (3 hourly) andspatial (1.0°) resolution but for a shorter time period(1986–95), is based on the NCEP–DOE reanalysis, andis described in detail by Zhao and Dirmeyer (2003).This section compares monthly mean values of precipi-tation, temperature, and radiation from the twodatasets as absolute differences and using the nonpara-metric Wilcoxon-signed rank test of differences. Themonthly mean diurnal temperature range and daily pre-cipitation frequencies are also examined, as these gen-erally have a significant impact on the hydrologic cycle.Figure 13 shows the mean annual differences and sta-tistical significance of differences in the monthly meansfor these variables.

1) MONTHLY MEAN TEMPERATURE

Comparison of the monthly temperatures revealeddifferences that are consistent with differences in theobservations used to create the two datasets. Both uselong-term monthly temperature from CRU, but differ-ent versions (GSWP uses version 1.0 and this study usesthe updated and extended version 2.0). Using the Wil-coxon-signed rank test, the null hypothesis that the me-dian difference of monthly means between the twodatasets is zero for each grid cell at the 95% confidencelevel was tested. Figure 13 shows that the null hypoth-esis can be rejected in the majority of regions, possiblydue to changes in contributing gauges between the twoversions of the CRU dataset and to the effects of usingdifferent datasets to correct for elevation effects. TheGSWP uses the ISLSCP elevation product and thisstudy uses the National Geophysical Data Center(NGDC) 2-min elevation dataset aggregated up to 1.0°resolution. The mean difference is 0.0965°C over globalland areas and maximum monthly differences of up to4°–5°C occur in parts of the Himalayas and TibetanPlateau, Northern Greenland, and in small isolated re-gions scattered across the globe.

2) MONTHLY MEAN PRECIPITATION

Monthly precipitation shows widespread differencesthat are statistically significant, which is to be expectedgiven the independent sources of observation data usedby each dataset. The GSWP data are based on GlobalPrecipitation Climatology Center (GPCC) monthlydata, which are used to scale the NCEP–DOE sub-monthly precipitation amounts. Corrections are alsoapplied for gauge undercatch. Data from the GlobalPrecipitation Climatology Project (GPCP) are blendedin for regions where the density of contributing gaugesfor the GPCC product is low. For this study, the CRUmonthly means are used, with corrections for gauge


undercatch based on the analysis of Adam and Letten-maier (2003). Several regions stand out as being coher-ently biased one way or the other. The GSWP datasetis generally greater across midlatitudes in both hemi-

spheres, with larger differences greater than 1.0 mmday�1 in Scandinavia, the Pacific Northwest, Alaska,the eastern United States, and southern South America.It is generally lower in the Tropics and high northern

FIG. 13. Difference of monthly mean values of air temperature, precipitation, downward short- and longwave radiation, diurnaltemperature range, and wet day frequency averaged over 1986–95 between the GSWP2 forcing dataset and this study. Color shadingrepresents critical values of the Wilcoxon-signed rank test statistic at the 95% level. Red shading is where the GSWP data are greater;blue shading is where GSWP data are less. Regions where the two datasets are statistically similar are unshaded.


Fig 13 live 4/C

latitudes, most notably in Central America, the Ama-zon basin, northern Canada, and Greenland. The globalmean annual bias in the GSWP is 0.0661 mm day�1

(24.1 mm yr�1).

3) MONTHLY MEAN DOWNWARD SHORT- AND

LONGWAVE RADIATION

For downward short- and longwave radiation, bothforcing datasets use SRB products, although differentversions (this study uses the SRB-QCSW and SRB-LWdatasets and GSWP uses the SRB-SW and SRB-QCLW datasets). Additionally, this study uses ob-served cloud cover data to adjust the interannual vari-ability of the shortwave monthly means, whereas theGSWP uses the data as is because of the direct overlapwith the SRB time period. The differences between thetwo datasets are consistent with the differences be-tween the SRB-SW and SRB-QCSW datasets (globalmean bias for GSWP is �8.2 W m�2). The GSWP isgenerally smaller, with the greatest differences across aband stretching from northern Africa to Japan, and alsoin the western and northeast United States. For long-wave, the differences are again consistent with the dif-ferences between the SRB-LW and SRB-QCLW. TheSRB-QCLW (GSWP) values tend to be larger (globalmean bias of 6.0 W m�2) and distributed similarly to theshortwave differences, but are lower at high latitudes.

4) DIURNAL TEMPERATURE RANGE

Both datasets use CRU DTR products to correct theair temperature. As for mean temperature, the GSWPdataset uses CRU version 1.0 and version 2.0 is usedhere. For the GSWP dataset, the 3-hourly air tempera-ture values are scaled by the ratio of the CRU toNCEP–DOE reanalysis DTR values but with restric-tions on the size of the ratio to avoid excessive values.A similar method is used here [section 3d(2)] but withno restriction on the ratio of CRU to NCEP values. Themean global bias for GSWP is �1.4°C and maximummonthly differences can exceed 5°C. GSWP is generallylarger in the Western Hemisphere and the Tropics, dur-ing the wet season, and is lower predominantly overcentral Asia and other dry regions worldwide.

5) DAILY PRECIPITATION FREQUENCY

The distribution of precipitation within a month interms of the number of wet and dry days plays an im-portant role in partitioning the monthly precipitationinto runoff and evaporation (Sheffield et al. 2004). TheGSWP dataset makes no adjustment to daily precipita-tion frequencies and therefore uses the NCEP–DOE as

is. For this study, the NCEP–NCAR reanalysis dailydata are resampled to match observation-baseddatasets of precipitation frequencies. The GSWP hason average 0.48 fewer precipitation days per month glo-bally. GSWP is generally greater in the humid Tropics(except for northwest South America) and southwestSouth America and is smaller in most other regions,especially in higher northern latitudes. Maximummonthly differences are generally less than 4 precipita-tion days but can reach 10 precipitation days in smallregions in Greenland, central Siberia, and the humidTropics.

b. Future improvements

The emphasis has been on using global-scale obser-vation datasets to ensure consistency in space; never-theless, better quality datasets exist in terms of spatialand temporal resolution but with smaller spatial andtemporal extents. For example, for the temporal disag-gregation of precipitation at high northern latitudes, itwas assumed because of the lack of coverage by theTRMM dataset that the diurnal cycle in cold, midlati-tude climates is representative of neighboring polar re-gions. Subdaily station data from Canadian surface air-ways products and the former Soviet Union (Razuvaevet al. 1998) are available for a significant number ofhigh-latitude locations and can be used to derive theprobability distributions used for the disaggregation.Furthermore, most monthly gridded precipitationdatasets also do not allow for orographic effects. As thenetwork of rain gauges that contributes to thesedatasets is not generally located in regions of complexand elevated topography, this usually results in an un-derestimation of precipitation, by as much as 3 times(Adam et al. 2006). The correction method of Adam etal. (2006) uses a simple catchment water balancemethod to calculate adjustments to precipitation. Thesechanges may be incorporated into new versions of thedataset in the future, although concerns over consis-tency in time and space may make this somewhat coun-terproductive. In addition, as improved and extendedversions of observation-based datasets used in thisstudy become available, they will be incorporatedwhere applicable.

c. Dataset availability

The forcing dataset will be made available over theInternet (see online at http://hydrology.princeton.edu)in the Assistance for Land-surface Modeling activities(ALMA) netcdf format (version 3), which is a standarddata exchange format for land surface scheme forcingand output data. The development of the final 1.0°,


3-hourly dataset has gone through a number of inter-mediate stages in terms of spatial and temporal resolu-tion and these intermediate products will also be addedto the archive. The following products are available:

• Global, 2.0°, 1948–2000, daily• Global, 1.0°, 1948–2000, daily• Global, 2.0°, 1948–2000, 3 hourly• Global, 1.0°, 1948–2000, 3 hourly

The variables are precipitation, air temperature, down-ward short- and longwave radiation, surface pressure,specific humidity, and wind speed. Global coverage in-dicates terrestrial regions excluding Antarctica.

d. Concluding remarks

This paper describes a long-term, high-resolution,near-surface meteorological dataset that can be usedfor forcing hydrologic simulations of the land surfacewater and energy budgets. The necessity for accurateestimates of the spatial and temporal variation in ter-restrial water and energy fluxes and states is evidentand is the driving force in the development of high-resolution and long-term hydroclimatological datasets.The development of the highest-quality forcing datasetsis a first and vital step toward this. Through researchinitiatives such as the World Climate Research Pro-gram (WCRP) Climate Variability and Predictability(CLIVAR) Program and the Global Energy and WaterCycle Experiment (GEWEX), the emphasis has beenon the development and enhancement of large-scaledatasets, through the use of increasingly better obser-vational datasets and new assimilation and modelingtechniques. This study is intended to form a part of thisprocess by providing a benchmark forcing dataset thatcombines state-of-the-art reanalysis products with themost recent observation-based datasets. The goals inthe development of this dataset are to provide consis-tency in time and space among variables from contrib-uting datasets while trying to achieve the highest reso-lution that can be supported by the data. This datasetprovides a significant improvement over the originalreanalysis variables and can be used for a wide varietyof applications and diagnostic studies in the climato-logical, hydrological, and ecological sciences.

Acknowledgments. This study was carried out withfunding from NASA Grants NAG5-9414 and NAG8-1517. The NCEP–NCAR reanalysis data were providedby the NOAA–CIRES Climate Diagnostics Center inBoulder, Colorado, from their Web site (see online athttp://www.cdc.noaa.gov/). The GPCP dataset wasdownloaded from the World Data Center for Meteo-

rology (see online at http://lwf.ncdc.noaa.gov/oa/wmo/wdcamet-ncdc.html). The TRMM dataset was alsodownloaded (available online at http://trmm.gsfc.nasa.gov). The CRU datasets were obtained from Dr. TimMitchell at the Climatic Research Unit (see online athttp://www.cru.uea.ac.uk) at the University of East An-glia, United Kingdom. The SRB dataset was down-loaded from the NASA Langley Research Center At-mospheric Science Data Center (see online at http://eosweb.larc.nasa.gov).

REFERENCES

Adam, J. C., and D. P. Lettenmaier, 2003: Adjustment of globalgridded precipitation for systematic bias. J. Geophys. Res.,108, 4257, doi:10.1029/2002JD002499.

——, E. A. Clark, D. P. Lettenmaier, and E. F. Wood, 2006: Cor-rection of global precipitation products for orographic ef-fects. J. Climate, 19, 15–38.

Barkstrom, B. R., E. F. Harrison, and G. L. Smith, 1989: Resultsfrom the Earth Radiation Budget Experiment (ERBE). Adv.Space Res., 9, 775–782.

Berbery, E. H., K. E. Mitchell, S. Benjamin, T. Smirnova, H.Ritchie, R. Hogue, and E. Radeva, 1999: Assessment of land-surface energy budgets from regional and global models. J.Geophys. Res., 104, 19 329–19 348.

Berg, A. A., J. S. Famiglietti, J. P. Walker, and P. R. Houser, 2003:Impact of bias correction to reanalysis products on simula-tions of North American soil moisture and hydrologicalf luxes. J. Geophys. Res. , 108, 4490, doi :10.1029/2002JD003334.

Betts, A. K., S.-Y. Hong, and H.-L. Pan, 1996: Comparison ofNCEP–NCAR reanalysis with 1987 FIFE data. Mon. Wea.Rev., 124, 1480–1498.

——, P. Viterbo, and E. Wood, 1998: Surface energy and waterbalance for the Arkansas Red River basin from the ECMWFreanalysis. J. Climate, 11, 2881–2897.

——, J. H. Ball, M. Bosilovich, P. Viterbo, Y. C. Zhang, and W. B.Rossow, 2003a: Intercomparison of water and energy budgetsfor five Mississippi subbasins between ECMWF reanalysis(ERA-40) and NASA Data Assimilation Office fvGCM for1990–1999. J. Geophys. Res., 108, 8618, doi:10.1029/2002JD003127.

——, J. H. Ball, and P. Viterbo, 2003b: Evaluation of the ERA-40surface water budget and surface temperature for the Mac-kenzie River basin. J. Hydrometeor., 4, 1194–1211.

Bo, Z., S. Islam, and E. A. B. Eltahir, 1994: Aggregation–disaggregation properties of a stochastic rainfall model. Wa-ter Resour. Res., 30, 3423–3435.

Bowling, L. C., and Coauthors, 2003: Simulation of high latitudehydrological processes in the Torne-Kalix basin: PILPSPhase 2(e) 1: Experiment description and summary intercom-parisons. Global Planet. Change, 38, 1–30.

Brotzge, J. A., 2004: A two-year comparison of the surface waterand energy budgets between two OASIS sites and NCEP–NCAR reanalysis data. J. Hydrometeor., 5, 311–326.

Connolly, R. D., J. Schirmer, and P. K. Dunn, 1998: A daily rain-fall disaggregation model. Agric. For. Meteor., 92, 105–117.

Cosgrove, B. A., and Coauthors, 2003: Real-time and retrospec-


tive forcing in the North American Land Data AssimilationSystem (NLDAS) project. J. Geophys. Res., 108, 8842,doi:10.1029/2002JD003118.

Cowpertwait, P. S. P., P. E. O’Connell, A. V. Metcalfe, and J. A.Mawdsley, 1996: Stochastic point process modeling of rain-fall. II. Regionalisation and disaggregation. J. Hydrol., 175,47–65.

Critchfield, H. J., 1983: General Climatology. 4th ed. PrenticeHall, 453 pp.

Cullather, R. I., D. H. Bromwich, and M. C. Serreze, 2000: Theatmospheric hydrologic cycle over the Arctic Basin from re-analyses. Part I: Comparison with observations and previousstudies. J. Climate, 13, 923–937.

Dai, A., 2001: Global precipitation and thunderstorm frequencies.Part II: Diurnal variations. J. Climate, 14, 1112–1128.

Darnell, W. L., W. F. Staylor, S. K. Gupta, N. A. Ritchey, andA. C. Wilber, 1992: Seasonal variation of surface radiationbudget derived from International Satellite Cloud Climatol-ogy Project C1 data. J. Geophys. Res., 97, 15 741–15 760.

Dirmeyer, P. A., X. Gao, M. Zhao, Z. Guo, T. Oki, and N. Ha-nasaki, 2005: The Second Global Soil Wetness Project(GSWP-2): Multi-model analysis and implications for ourperception of the land surface. COLA Tech. Rep. 185, 46 pp.[Available online at ftp://grads.iges.org/pub/ctr/ctr_185.pdf.]

Eltahir, E. A. B., and R. L. Bras, 1993: Estimation of the frac-tional coverage of rainfall in climate models. J. Climate, 6,639–644.

Fekete, B. M., C. J. Vörösmarty, J. O. Roads, and C. J. Willmott,2004: Uncertainties in precipitation and their impacts on run-off estimates. J. Climate, 17, 294–304.

Fu, Q., K. N. Liou, M. C. Cribb, T. P. Charlock, and A. Grossman,1997: Multiple scattering parameterization in thermal infra-red radiative transfer. J. Atmos. Sci., 54, 2799–2812.

Gibson, J. K., P. Kållberg, S. Uppala, A. Hernandez, A. Nomura,and E. Serrano, 1997: ERA description. ECMWF ReanalysisProject Report Series 1, 72 pp.

Gilgen, H., M. Wild, and A. Ohmura, 1998: Means and trends ofshortwave irradiance at the surface estimated from globalenergy balance archive data. J. Climate, 11, 2042–2061.

Gong, G., D. Entekhabi, and G. D. Salvucci, 1994: Regional andseasonal estimates of fractional storm coverage based on sta-tion precipitation observations. J. Climate, 7, 1495–1505.

Goodison, B. E., P. Y. T. Louie, and D. Yang, 1998: WMO solidprecipitation measurement intercomparison, final report.World Meteorological Organization Tech. Doc. WMO TD-872, Geneva, Switzerland, 212 pp.

Gottschalck, J., J. Meng, M. Rodell, and P. Houser, 2005: Analysisof multiple precipitation products and preliminary assess-ment of their impact on Global Land Data Assimilation Sys-tem land surface states. J. Hydrometeor., 6, 573–598.

Gupta, S. K., W. L. Darnell, and A. C. Wilber, 1992: A param-eterization for longwave surface radiation from satellite data:Recent improvements. J. Appl. Meteor., 31, 1361–1367.

——, N. A. Ritchey, A. C. Wilber, C. H. Whitlock, G. G. Gibson,and P. W. Stackhouse, 1999: A climatology of surface radia-tion budget derived from satellite data. J. Climate, 12, 2691–2710.

Gütner, A., J. Olsson, A. Calver, and B. Gannon, 2001: Cascade-based disaggregation of continuous rainfall time series: Theinfluence of climate. Hydrol. Earth Syst. Sci., 5, 145–164.

Hahmann, A. N., 2003: Representing spatial subgrid-scale precipi-tation variability in a GCM. J. Hydrometeor., 4, 891–900.

Hall, F. G., G. Collatz, S. Los, E. Brown de Colstoun, and D.

Landis, Eds., 2005: ISLSCP Initiative II. NASA, DVD/CD-ROM.

Hershenhorn, J., and D. A. Woolhiser, 1987: Disaggregation ofdaily rainfall. J. Hydrol., 95, 299–322.

Hsu, K., X. Gao, S. Sorooshian, and H. V. Gupta, 1997: Precipi-tation estimation from remotely sensed information using ar-tificial neural networks. J. Appl. Meteor., 36, 1176–1190.

Huffman, G. J., and Coauthors, 1997: The Global PrecipitationClimatology Project (GPCP) combined precipitation dataset.Bull. Amer. Meteor. Soc., 78, 5–20.

——, R. F. Adler, M. M. Morrissey, D. T. Bolvin, S. Curtis, R.Joyce, B. McGavock, and J. Susskind, 2001: Global precipi-tation at one-degree daily resolution from multisatellite ob-servations. J. Hydrometeor., 2, 36–50.

——, E. F. Stocker, D. T. Bolvin, and E. J. Nelkin, 2003: Analysisof TRMM 3-hourly multi-satellite precipitation estimatescomputed in both real and post-real time. Preprints, 12thConf. on Satellite Meteorology and Oceanography, LongBeach, CA, Amer. Meteor. Soc., CD-ROM, P4.11.

Janowiak, J. E., A. Gruber, C. R. Kondragunta, R. E. Livezey,and G. J. Huffman, 1998: A comparison of the NCEP–NCARreanalysis precipitation and the GPCP rain gauge–satellitecombined dataset with observational error considerations. J.Climate, 11, 2960–2979.

Johnson, K. D., D. Entekhabi, and P. S. Eagleson, 1993: Theimplementation and validation of improved land-surface hy-drology in an atmospheric general circulation model. J. Cli-mate, 6, 1009–1026.

Kalnay, E., and M. Cai, 2003: Impact of urbanisation and land-usechange on climate. Nature, 423, 528–531.

——, and Coauthors, 1996: The NCEP/NCAR 40-Year Reanaly-sis Project. Bull. Amer. Meteor. Soc., 77, 437–471.

Kanamitsu, M., W. Ebisuzaki, J. Woollen, S.-H. Yang, J. J. Hnilo,M. Fiorino, and G. L. Potter, 2002: NCEP–DOE AMIP-IIReanalysis (R-2). Bull. Amer. Meteor. Soc., 83, 1631–1643.

Kistler, R., and Coauthors, 2001: The NCEP–NCAR 50-Year Re-analysis: Monthly means CD-ROM and documentation. Bull.Amer. Meteor. Soc., 82, 247–267.

Lau, K.-M., V. M. Mehta, Y. C. Sud, and G. K. Walker, 1994:Climatology and natural variability of the global hydrologiccycle in the GLA atmospheric general circulation model. J.Geophys. Res., 99 (D1), 1329–1345.

Lenters, J. D., M. T. Coe, and J. A. Foley, 2000: Surface waterbalance of the continental United States, 1963–1995: Re-gional evaluation of a terrestrial biosphere model and theNCEP/NCAR reanalysis. J. Geophys. Res., 105 (D17),22 393–22 425.

Levis, S., M. T. Coe, and J. A. Foley, 1996: Hydrologic budget ofa land surface model: A global application. J. Geophys. Res.,101 (D12), 16 921–16 930.

Liang, X., D. P. Lettenmaier, E. F. Wood, and S. J. Burges, 1994:A simple hydrologically based model of land surface waterand energy fluxes for general circulation models. J. Geophys.Res., 99 (D7), 14 415–14 428.

Marani, M., G. Grossi, M. Wallace, F. Napolitano, and D. En-tekhabi, 1997: Forcing, intermittency, and land surface hy-drological partitioning. Water Resour. Res., 33, 167–175.

Maurer, E. P., G. M. O’Donnell, D. P. Lettenmaier, and J. O.Roads, 2001: Evaluation of the land surface water budget inNCEP/NCAR and NCEP/DOE Reanalyses using an off-linehydrologic model. J. Geophys. Res., 106 (D16), 17 841–17 862.

——, A. W. Wood, J. C. Adam, D. P. Lettenmaier, and B. Nijssen,


2002: A long-term hydrologically-based data set of land sur-face fluxes and states for the conterminous United States. J.Climate, 15, 3237–3251.

Meeson, B. W., F. E. Corprew, J. M. P. McManus, D. M. Myers,J. W. Closs, K.-J. Sun, D. J. Sunday, and P. J. Sellers, 1995:ISLSCP Initiative I-Global Data Sets for Land-AtmosphereModels, 1987–1988. Vols. 1–5, NASA, CD-ROM.

Mitchell, K. E., and Coauthors, 2004: The multi-institution NorthAmerican Land Data Assimilation System (NLDAS) project:Utilizing multiple GCIP products and partners in a continen-tal distributed hydrological modeling system. J. Geophys.Res., 109, D07S90, doi:10.1029/2003JD003823.

Morrill, J. C., R. E. Dickinson, and A. N. Hahmann, 1999: Sensi-tivity of a land surface model to the diurnal distribution ofdownward longwave radiation. J. Meteor. Soc. Japan, 77 (1B),265–279.

Negri, A. J., T. L. Bell, and L. Xu, 2002: Sampling of the diurnalcycle of precipitation using TRMM. J. Atmos. Oceanic Tech-nol., 19, 1333–1344.

New, M. G., M. Hulme, and P. D. Jones, 1999: Representing twen-tieth-century space–time climate variability. Part 1: Develop-ment of a 1961–90 mean monthly terrestrial climatology. J.Climate, 12, 829–856.

——, ——, and ——, 2000: Representing twentieth-centuryspace–time climate variability. Part II: Development of 1901–1996 monthly grids of terrestrial surface climate. J. Climate,13, 2217–2238.

Ngo-Duc, T., J. Polcher, and K. Laval, 2005: A 53-year forcingdata set for land surface models. J. Geophys. Res., 110,D06116, doi:10.1029/2004JD005434.

Nijssen, B., and D. P. Lettenmaier, 2004: Effect of precipitationsampling error on simulated hydrological fluxes and states:Anticipating the Global Precipitation Measurement satellites.J. Geophys. Res., 109, D02103, doi:10.1029/2003JD003497.

——, G. M. O’Donnell, D. P. Lettenmaier, D. Lohmann, andE. F. Wood, 2001a: Predicting the discharge of global rivers.J. Climate, 14, 3307–3323.

——, R. Schnur, and D. P. Lettenmaier, 2001b: Global retrospec-tive estimation of soil moisture using the Variable InfiltrationCapacity land surface model, 1980–93. J. Climate, 14, 1790–1808.

——, and Coauthors, 2003: Simulation of high latitude hydrologi-cal processes in the Torne-Kalix basin: PILPS Phase 2(e) 2:Comparison of model results with observations. GlobalPlanet. Change, 38, 31–53.

Olsson, J., 1998: Evaluation of a cascade model for temporal rain-fall disaggregation. Hydrol. Earth Syst. Sci., 2, 19–30.

Pan, M., and E. F. Wood, 2004: Water budget estimation by as-similating multiple observations and hydrological modelingusing constrained ensemble kalman filtering. Eos, Trans.Amer. Geophys. Union, 85 (Suppl.).

——, and Coauthors, 2003: Snow process modeling in the NorthAmerican Land Data Assimilation System (NLDAS): 2.Evaluation of model simulated snow water equivalent. J.Geophys. Res., 108, 8850, doi:10.1029/2003JD003994.

Pinker, R. T., and I. Laszlo, 1992: Modeling surface solar irradi-ance for satellite applications on a global scale. J. Appl. Me-teor., 31, 194–211.

——, and Coauthors, 2003: Surface radiation budgets in supportof the GEWEX Continental-Scale International Project

(GCIP) and the GEWEX Americas Prediction Project(GAPP), including the North American Land Data Assimi-lation System (NLDAS) project. J. Geophys. Res., 108, 8844,doi:10.1029/2002JD003301.

——, B. Zhang, and E. G. Dutton, 2005: Do satellites detecttrends in surface solar radiation? Science, 308, 850–854.

Qu, W., and Coauthors, 1998: Sensitivity of latent heat flux fromPILPS land surface schemes to perturbations of surface airtemperature. J. Atmos. Sci., 55, 1909–1927.

Razuvaev, V. N., E. G. Apasova, and R. A. Martuganov, 1998:Six- and three-hourly meteorological observations from 223U.S.S.R. stations. ORNL/CDIAC-108, NDP-048/R1, CarbonDioxide Information Analysis Center, Oak Ridge NationalLaboratory, Oak Ridge, TN. [Available online at http://cdiac.ornl.gov/ftp/ndp048/.]

Roads, J., M. Kanamitsu, and R. Stewart, 2002a: CSE water andenergy budgets in the NCEP–DOE Reanalysis II. J. Hydro-meteor., 3, 227–248.

——, E. Raschke, and B. Rockel, 2002b: BALTEX water andenergy budgets in the NCEP/DOE Reanalysis II. Boreal En-viron. Res., 7, 307–317.

——, and Coauthors, 2003: GCIP water and energy budget syn-thesis (WEBS). J. Geophys. Res., 108, 8609, doi:10.1029/2002JD002583.

Robock, A., and Coauthors, 2003: Evaluation of the NorthAmerican Land Data Assimilation System over the SouthernGreat Plains during the warm season. J. Geophys. Res., 108,8846, doi:10.1029/2002JD003245.

Rodell, M., and Coauthors, 2004: The Global Land Data Assimi-lation System. Bull. Amer. Meteor. Soc., 85, 381–394.

Rossow, W. B., and R. A. Schiffer, 1991: ISCCP cloud data prod-ucts. Bull. Amer. Meteor. Soc., 72, 2–20.

Salby, M. L., and P. Callaghan, 1997: Sampling error in climateproperties derived from satellite measurements: Conse-quences of undersampled diurnal variability. J. Climate, 10,18–36.

Schlosser, C. A., A. G. Slater, A. Robock, A. Pitman, K. Y. Vin-nikov, A. Henderson-Sellers, N. A. Speranskaya, and K.Mitchell, 2000: Simulations of a boreal grassland hydrology atValdai, Russia: PILPS phase 2(d). Mon. Wea. Rev., 128, 301–321.

Schubert, S. D., R. Rood, and J. Pfaendtner, 1993: An assimilateddata set for earth sciences applications. Bull. Amer. Meteor.Soc., 74, 2331–2342.

Serreze, M. C., and C. M. Hurst, 2000: Representation of meanArctic precipitation from NCEP–NCAR and ERA reanaly-ses. J. Climate, 13, 182–201.

Sheffield, J., A. D. Ziegler, E. F. Wood, and Y. Chen, 2004: Cor-rection of the high-latitude rain day anomaly in the NCEP–NCAR reanalysis for land surface hydrological modeling. J.Climate, 17, 3814–3828.

Simmons, A. J., and Coauthors, 2004: Comparison of trends andlow-frequency variability in CRU, ERA-40, and NCEP/NCAR analyses of surface air temperature. J. Geophys. Res.,109, D24115, doi:10.1029/2004JD005306.

Soman, V. V., J. B. Valdés, and G. R. North, 1995: Satellite sam-pling and the diurnal cycle statistics of Darwin rainfall data.J. Appl. Meteor., 34, 2481–2490.

Stackhouse, P. W., S. K. Gupta, S. J. Cox, J. C. Mikowitz, T.Zhang, and M. Chiacchio, 2004: 12-year surface radiation


budget data set. GEWEX News, Vol. 14, No. 4, InternationalGEWEX Project Office, Silver Spring, MD, 10–12.

Susskind, J., P. Piraino, L. Rokke, L. Iredell, and A. Mehta, 1997:Characteristics of the TOVS Pathfinder Path A Dataset. Bull.Amer. Meteor. Soc., 78, 1449–1472.

Thornton, P. E., and S. W. Running, 1999: An improved algorithmfor estimating incident daily solar radiation from measure-ments of temperature, humidity, and precipitation. Agric.For. Meteor., 93, 211–228.

Trenberth, K. E., and C. J. Guillemot, 1998: Evaluation of theatmospheric moisture and hydrological cycle in the NCEP/NCAR reanalyses. Climate Dyn., 14, 213–231.

Werth, D., and R. Avissar, 2002: The local and global effects ofAmazon deforestation. J. Geophys. Res., 107, 8087,doi:10.1029/2001JD000717.

Wild, M., A. Ohmura, H. Gilgen, J.-J. Morcrette, and A. Slingo,

2001: Evaluation of downward longwave radiation in generalcirculation models. J. Climate, 14, 3227–3239.

——, and Coauthors, 2005: From dimming to brightening: Dec-adal changes in solar radiation at Earth’s surface. Science,308, 847–850.

Zhang, Y., W. B. Rossow, A. A. Lacis, V. Oinas, and M. I. Mish-chenko, 2004: Calculation of radiative fluxes from the surfaceto top of atmosphere based on ISCCP and other global datasets: Refinements of the radiative transfer model and theinput data. J. Geophys. Res., 109, D19105, doi:10.1029/2003JD004457.

Zhao, M., and P. Dirmeyer, 2003: Production and analysis ofGSWP-2 near-surface meteorology data sets. COLA Tech.Rep. 159, 22 pp. [Available from the Center for Ocean–Land–Atmosphere Studies, 4041 Powder Mill Road, Suite302, Calverton, MD 20705.]


iri – international research institute for climate and society ...blyon/references/p30.pdfmodeling...

Documents