1. Introduction
Investigations related to climate variations, climate changes, and other climate diagnostics, such as decadal variations and temperature trends, demand long-term datasets with homogeneous quality. Studies of such topics using the analyses from operational global data assimilation systems (GDASs) have been subject to various quality issues, such as periodic and nonperiodic system changes (e.g., observational satellites changes, surface observational data changes, NWP model improvements, etc.), as well as incomplete injection of observational data from the delayed transmissions. For such reasons, studies of one subject with different datasets could lead to vastly different conclusions. As an example, Bengtsson and Shukla (1988) show that the generation of global eddy-available potential energy was of opposite signs when the calculations were made from two different versions of the European Centre for Medium-Range Weather Forecasts First GARP Global Experiment datasets for February 1979. In order to enhance our understandings of the earth climate system, several meteorological centers have thus undertaken the projects to reanalyze the global observational data, using a fixed, state-of-the-art, data assimilation system, The result is a long-term internally consistent, homogeneous, multivariate dataset.
Recently, the National Centers for Environmental Prediction (NCEP) and National Center for Atmospheric Research have completed a 40-yr reanalysis project (Kalnay et al. 1996), hereafter called the reanalysis. This reanalysis makes use of most of the observational data that was available at any time, including both conventional and remotely sensed data. The GDAS, which consists of an analysis scheme and a global forecast model, is identical to the NCEP operational system implemented on 11 January 1995, except that a coarse horizontal resolution of T62 (approximately 210 km) is used. Posting the designated project period of 1957–96, the same frozen process is continued, such that an expanding time series with the same homogeneous quality for current state can be obtainable.
The reanalysis data output is rather comprehensive; from the primary parameters of temperature, u, υ, humidity, and geopotential height, to the diagnostic variables, such as vertical diffusion meridional acceleration, etc. (Kalnay et al. 1996). This long-period dataset contains several episodes of El Niño–Southern Oscillation (ENSO) and may contain some information about recent climate changes. A dataset as inclusive as this one invariably can be utilized for revisiting a number of fundamental issues about the general circulation (e.g., Hartmann et al. 1986; Ramanathan 1987), such as the global energy balance and energy transports related to the earth radiation budget (e.g., Oort and Vonder Haar 1976; Kann et al. 1994).
The components of the reanalysis earth radiation budget are computed by the radiative transfer algorithm in the global model. The computation requires the prognostic parameters of temperature and moisture profiles, the diagnostic parameters of clouds, and the prescribed parameters for surface properties and ozone. It is a coherent, dynamical–physical, processing system, whose output reflects the state of the assimilated atmospheric–earth system. Thus an evaluation of the reanalysis radiation budget data can serve two objectives. It can document the data quality for this aspect of the reanalysis, which scientific users may find useful in many applications. It can also identify general problems of GDAS, leading to future system improvement.
This study compares the monthly reanalysis to the National Aeronautics and Space Administration’s (NASA) Earth Radiation Budget Experiment (ERBE) data (Barkstrom et al. 1989), and to ERBE-derived surface albedo. For the period investigated, the ERBE data are measured from broadband scanning radiometers on board two satellites, and are generally recognized as the high quality radiation budget data. In section 2, the major modules of the GDAS, especially those related to the earth radiation budget are discussed. Section 3 describes the datasets used in this study, including the top-of-the-atmosphere (TOA) outgoing longwave radiation (OLR) and TOA reflected shortwave radiation (RSW) for both total-sky and clear-sky conditions. In addition, surface albedo is also evaluated. The results of the comparison are presented in section 4, while a summary is stated in section 5.
2. GDAS reanalysis system
The NCEP–NCAR reanalysis project involves observational data preparation, data assimilation, and data archival/distribution. As a whole, it is a complex operation that has taken substantial coordination of personnel, facilities, data, and algorithms from various institutions. A more detailed description of the reanalysis project is reported in Kalnay et al. (1996). The reanalysis archive contains components of the earth radiation budget, including both TOA total-sky fluxes and cloud forcing, which can be utilized for deriving clear-sky (CS) fluxes. In this study we focus on TOA OLR and RSW for both the total-sky and CS conditions, as well as surface albedo.
The fields of the radiation fluxes are computed in the global model (Kanamitsu 1989), which has a spectral truncation of T62, approximately 2° lat × 2° long horizontal resolution, and 28 vertical layers. These radiative transfer computations use the prognostic parameters of temperature and moisture, the diagnostic parameters of clouds, and the prescribed parameters of surface and atmospheric properties (e.g., ozone and cloud emissivity, etc.). The radiative transfer algorithm is called once every hour instead of every numerical time step, for computational efficiency. This frequency is sufficient to simulate the diurnal cycles for the subject horizontal resolution. The reanalysis archives the output once every 6 h. Thus, the radiation fluxes are averages from a 6-h forecast period. Monthly means are computed from the 6-h averages from these archives.
The 6-h forecast from the global model serves as a first guess for the data analysis. NCEP uses the spectral statistical interpolation (SSI), a three-dimensional variational analysis scheme of Parrish and Derber (1992), instead of the conventional optimal interpolation scheme. SSI imposes a balanced analysis over the globe, such that a nonlinear normal-mode initialization in the model forecast is no longer needed. Since the implementation in June 1991, SSI has improved forecasts and significantly reduced precipitation “spinup” (Kalnay and Jenne 1991; Kalnay et al. 1996), which implies improved model water vapor data is being passed to the radiative transfer computations.
For the terrestrial longwave computation, the efficient algorithm of Schwarzkopf and Fels (1991, hereafter SF) is used, which incorporates the effects of water vapor e-type continuum, water–carbon dioxide overlap, and Voigt line shape, etc. For CS OLR, SF agrees very well with the line-by-line benchmark calculations on the standard atmospheric profiles obtained for the Intercomparison of Radiation Codes in Climate Models (ICRCCM; Ellingson and Fouquart 1991). The largest difference occurs in the tropical profile, which SF slightly underestimates by 1.37 W m−2; see Table 1. This error is smaller than the uncertainty of ERBE data to be discussed in section 3. The reanalysis uses 330 ppmv for carbon dioxide concentration. The seasonal ozone profiles are from the zonal-mean climatologies constructed from Hering and Borden (1965) and London (1962). Sea surface temperature is from the optimal interpolation reanalysis of Reynolds and Smith (1994). A surface emissivity of 1 is used throughout the whole globe.
For the solar, shortwave, computation, the global model uses a scheme strongly based on one developed by Lacis and Hansen (1974). This parametric scheme accounts for multiple scattering. In addition to absorption by ozone and carbon dioxide (Sasamori et al. 1972), the effects of water vapor, clouds, and surface albedo are also considered. The monthly surface albedo over land is prescribed using data from Matthews (1985), and ocean surface albedo is prescribed from the parameterization of Payne (1972). No effect of aerosols is considered.
The earth radiation budget is highly modulated by the clouds. Although not within the scope of the current study, it is relevant to mention an issue related to the reanalysis cloud data. In the NCEP global model, radiatively active clouds are generated either from the Arakawa–Schubert-type convection scheme of Grell (1993), or from the diagnostic stratus parameterization of Campana et al. (1994a,b), both of which are similar to Slingo (1987). This stratus parameterization uses relative humidity, RH, as a predictor. Since the NCEP global model uses mixing ratio, Q, as the prognostic parameter in the hydrological cycle, RH is converted from Q. In this conversion, however, the saturation vapor pressure is calculated with respect to liquid phase only, even when the temperature is below freezing and vapor pressure over ice should have been calculated. This simplification may have introduced some errors to the reanalysis clouds.
3. Comparison datasets
a. ERBE TOA fluxes
To evaluate the reanalysis data, TOA fluxes measured by the NASA ERBE (Barkstrom et al. 1989) are used. During the 2-yr period studied, 1985–86, the monthly mean data are derived from the measurements sampled by two satellites, the sun-synchronous NOAA-9 and the Earth Radiation Budget Satellite. The latter was a dedicated satellite with 57° inclination and precession 4.95° west in the ascending node. The combination of the two satellites with varying sampling time provides virtually complete diurnal variation and global coverage.
Two types of instruments were on board each of the satellites: scanning radiometers (scanners) and nonscanning radiometers (nonscanners). Scanners had three channels: total (0.2–200 μm), shortwave (reflected solar radiation, 0.2–5 μm), and longwave (thermal emission, 5–200 μm). Nonscanners had four channels: total spectral and shortwave for both wide field of view and medium field of view. To ensure continued accuracy was maintained, all the channels were subject to in-flight calibration. For the current study, only the higher-resolution scanner data were used, 40 km at nadir. The measurements were mapped to a 2.5° latitude–longitude horizontal grid. The reanalysis flux data were converted to the same resolution for comparison.
A special feature of ERBE products is the inclusion of the clear-sky estimates, which are derived from the samples of cloud-free scenes. This parameter is useful in deriving cloud forcing (Ramanathan 1987), which yields the direct effect of clouds on the atmosphere. The clear-sky fluxes are also useful for other applications, such as the assimilation of radiative transfer calculations by the general circulation models (Webb et al. 1993), or inferring surface albedo (Staylor and Wilbur 1990) from the clear-sky RSW. Other monthly CS OLR climatologies that can be referred are the compilation of Yang (1987), which uses atmospheric circulation statistics and a parameterization of Thompson and Warren (1982), and the estimates of Ardanuy et al. (1989) from Nimbus-7 satellite observations.
The archived ERBE data are generated from a complex process that includes many algorithms. Thus, the overall uncertainties are difficult to characterize (Barkstrom et al. 1989). The major sources of the uncertainties are from sampling deficiencies and from misclassification of scene type. The errors of selecting scene type would lead to using the incorrect anisotropic models for data inversion. Based on a detailed study of April 1985, Barkstrom et al. (1989) estimate that the uncertainties of ERBE data are about 1% for longwave and 2% ∼ 3% for shortwave. These translate to ±5 W m−2 for the monthly averages for both total-sky OLR and RSW. Other studies show better results. Harrison et al. (1990) reports that the rms errors appear to be about ∼2 W m−2 for the monthly means CS OLR and CS RSW, and about 3 W m−2 and 5 W m−2 for total-sky OLR and RSW, respectively.
b. Surface albedo
The global coverage from the ERBE clear-sky product makes it possible to estimate the broadband surface albedo from space observations. Staylor and Wilber’s (1990) estimates of the surface albedo are based on the TOA-to-surface conversion of Koepke and Kriebel (1987), with the effect of solar zenith angle, water vapor, aerosol absorption, and scattering taken into account. In compiling this surface albedo dataset, ancillary water vapor and ozone data are taken from TOVS observations, and aerosol data are from the World Climate Program reports (WCRP 1993).
4. Results
a. TOA global synopsis
The global average differences between the ERBE observations and the reanalysis in Figs. 1 and 2 show the time series of global means for both CS fluxes and total-sky fluxes for OLR and RSW, respectively. The seasonal variations of TOA fluxes are modulated by the response of landmass to the solar cycle, which is dominant in the Northern Hemisphere. The minimum in the northern winter OLR is due to the lower surface temperature and to the increased cloudiness. RSW has a maximum in the same season due to the brighter surface and the increased cloudiness, in addition to the higher TOA insolation.
Both CS OLR and CS RSW show very good agreement, Figs. 1 and 2, with the reanalysis only slightly greater than ERBE. The only exception is that the CS OLR bias is closer to zero in about half of 1985. The 1985 annual-mean CS OLR for ERBE is only 0.1 W m−2 greater than that of reanalysis. The 1985–86 mean differences, see Table 2, are 1.1 W m−2 for CS OLR, and 1.8 W m−2 for CS RSW, respectively. These differences are within the uncertainties of ERBE.
However, the differences are larger for the total-sky cases: 3.1 W m−2 for OLR and 12.6 W m−2 for RSW. The latter is well beyond the uncertainties of ERBE (Barkstrom et al. 1989; Harrison et al. 1990). This persistent overestimation of RSW throughout the subject period indicates that the global energy budget from the reanalysis is not balanced (Yang et al. 1997), as less solar energy is absorbed and slightly more terrestrial energy exits the earth–atmosphere. Since there is good agreement for CS cases, but not for total-sky cases, the results also suggest that the reanalysis may contain shortcomings in the hydrological cycles related to cloudiness.
b. OLR
1) Clear-sky OLR
In estimating the clear-sky fluxes, the reanalysis employs method II computation procedures of Cess and Potter (1987), while ERBE uses method I. Method I samples the “drier” clear part of the atmosphere, in contrast to method II, which calculates the flux by removing cloud and maintaining the atmospheric profile. Method II tends to underestimate method I, although the magnitudes of the differences are difficult to summarize as they vary with cloud conditions. This difference has little effect in the comparison of the global means. However, it could affect the zonal means, especially on the high latitudes of the winter seasons (Cess and Potter 1987). As discussed in the global averages, the reanalysis OLR is larger than that of ERBE. This bias would have been even larger if the method I computation procedure had been used.
The time series of the zonal mean difference (ERBE minus reanalysis) is depicted in Fig. 3. It shows that reanalysis CS OLR overestimates ERBE for most latitudes. The difference pattern indicates a strong seasonality, especially at mid- to high latitudes. North of 45°N, the reanalysis is larger than ERBE by up to 15 W m−2 in January, while almost equal in July. From 30° to 40°N, the reanalysis is less than ERBE from October to April, although the magnitude of the difference is smaller than the region north of 45°N. There is similar seasonality in the Antarctic region with an amplitude of 20 W m−2; the phasing is, however, not as evenly spaced as in the Arctic. There is also a weak signal of the semiannual cycle over the Tropics.
Figure 4 shows the 2-yr average (1985–86) geographical distribution of the ERBE CS OLR (Fig. 4a), and the difference of ERBE minus the reanalysis (Fig. 4b). Land areas tend to have much larger differences than those shown over the oceans. North Africa is the largest region where the reanalysis substantially overestimates ERBE by more than 10 W m−2. A large portion of this region is sandy desert, where surface observations are scarce. It is possible that the reanalysis overestimates the surface temperature. The other possible cause is that the surface emissivity (1.0) prescribed by the reanalysis is substantially overestimated. The emissivity of sand can be as low as 0.90 (Sellers 1965). Similar arguments can also be applied to the northwest part of Australia and the Great Sandy Desert.
From the Himalayas eastward to central China, the reanalysis underestimates ERBE by about 20 W m−2. East and west of the Andes, the reanalysis overestimates ERBE by 5–10 W m−2, and underestimates at the Andes. It appears these types of wavelike bias are from Gibbs oscillation (Lindberg and Broccoli 1996) caused by the complex terrain that cannot be resolved by the T62 spatial resolution. Although the reanalysis overestimates ERBE by more than 10 W m−2 poleward of 60°, the magnitude is seasonally dependent, as shown in Fig. 3.
The reanalysis overestimates ERBE over most of the oceans, which is the largest contribution to the global-mean bias. However, the reanalysis underestimates ERBE in the subsidence regions of the central Pacific, as well as from 15°S to the equator in the Indian Ocean. The difference for these two large regions ranges from 1 to more than 5 W m−2. Other than the small errors from sea surface temperature (SST) analysis, the most probable cause is that the reanalysis moisture is overestimated, to be discussed in section 4b(2).
From 45°S to Antarctic, the reanalysis overestimates ERBE by about 5 W m−2. This is the region where the surface transits from water to ice. The reanalysis uses Reynolds SST reanalysis and National Environmental Satellite, Data and Information Service weekly snow cover analysis (Kalnay et al. 1996). If ERBE is considered as truth, the differences indicate that the surface temperature is overestimated by approximately 2.5°C. At these temperatures, CS OLR is nearly linear to the surface temperature, with a slope of about 2 W m−2 per degree change, assuming that the column humidity is not overly dry (Thompson and Warren 1982; Yang et al. 1987).
2) Total-sky OLR
Total-sky OLR, which reflects the integral of the earth–atmosphere state, has been one of the most used space-viewed parameters in climate diagnostics, especially for studying tropical phenomena. As discussed in the global synopsis, section 4a, the global annual-mean difference is about 1.5% with the reanalysis slightly higher than ERBE (237.1 vs 234.0 W m−2). The total-sky OLR difference of 3.1 W m−2, substantially larger than the 0.9 W m−2 of CS OLR, reflects the difficulties in modeling the moisture/cloud-related processes by GDAS.
The time series of the zonal mean difference of total-sky OLR, ERBE minus the reanalysis in Fig. 5, also exhibits a strong seasonality as shown in CS OLR. However, the features are fairly different. There is more positive bias at the poles for CS OLR, and this positive bias prevails in the summer season. From the extratropics to the midlatitudes, the seasonal variation is more pronounced. Interestingly, the variation peaks in April and October, instead of January and July, with a magnitude of 10 W m−2 for positive bias. In the Tropics, just north and south of the equator, is where the strongest negative bias occurs, to be discussed further below.
Figure 6a shows the 2-yr average geographical distribution of the ERBE total-sky OLR, which essentially is the modulation of cloudiness over the clear-sky OLR (Fig. 4a). The regions with large annual cloudiness, such as Indonesia, African Congo, and the Amazon are where the OLR minimums are located. Figure 6b shows CS OLR differences, ERBE minus reanalysis. The reanalysis is negatively biased in the subtropical regions over the central Pacific, the Indian Ocean, and the Amazon Basin. The 2-yr mean differences in these regions are larger than 20 W m−2.
The regions where the reanalysis is positively biased are located over the Sahara Desert with a magnitude of more than 10 W m−2, and over Indonesia by more than 20 W m−2. The CS OLR difference over Indonesia is only about 5 W m−2, whereas it is more than 10 W m−2 for the Sahara Desert. Part of this can be explained by the lower surface albedo in GDAS, to be discussed in section 4e.
One of the intriguing features is comparing the two active convection regions, the Amazon and Indonesia, where there is reasonably good agreement in CS condition, with less than 5 W m−2 difference. In the total-sky conditions, however, the reanalysis has +20 W m−2 bias for the Amazon, and −20 W m−2 bias for Indonesia. The differences of the two regions are of opposite signs.
The causes for such total-sky differences of opposite sign could be multifactorial as OLR reflects the integral state of the atmosphere. The disagreement in the total-sky condition suggests that the problems are moisture related. In an evaluation of the NCEP global atmospheric moisture budgets produced from a similar GDAS, Trenberth and Guillemot (1995) point out that none of the global analyses produce very reliable moisture budgets. The source of problems is in the large-scale analyzed fields of moisture and divergence, which are in turn related to the parameterization of moist processes in the global model. It is suspected that the moisture-related parameterizations over the Amazon Basin generate excessive cold cloud top and precipitation. OLR is highly correlated with precipitation in the Tropics and the extratropics. In studying the reanalysis precipitation, Higgens et al. (1996) also found a similar bias when compared to the estimate of Xie and Arkin (1996).
The other regions of negative bias for the reanalysis are over the subsidences regions of the Pacific and Indian Oceans just north and south of the ITCZ. The biases of more than 20 W m−2 are rather substantial, while the CS biases are not significant. These results suggest that the reanalysis cloud amounts are overestimated, or some of the cloud properties are incorrectly prescribed. This is consistent with the comparison of RSW, as the reanalysis overestimates ERBE RSW by more than 50 W m−2 in these specific regions.
In the regions northeast of the Himalayas and east of the Andes, the reanalysis shows positive biases in the range of 25 W m−2. The bias is similar to that of CS for the Andes, but opposite for the region northeast of the Himalayas. This may be attributed to the complex topography unresolved by the T62 model resolution and/or other causes.
c. RSW
1) Synopsis of RSW
Reflected shortwave radiation is not as often discussed in the climate studies as the absorbed solar radiation (ASR), defined as the difference between the TOA insolation and RSW. For the studies like earth–atmosphere energetics (e.g., Oort and Vander Haar 1976; Kann et al. 1994), ASR is considered as the forcing to the general circulation and, thus, can be directly used. However, for the purpose of diagnosing the reanalysis, a comparison of RSW is more convenient.
RSW is highly modulated by the cloudiness as OLR. However, RSW is anticorrelated with OLR in the lower latitudes, as clouds emit less thermal radiation and reflect more solar radiation than the background surface. The argument does not hold in the high latitudes due to the fact that TOA insolation diminishes much faster than OLR. Because of the earth–sun geometry, the seasonal cycle of RSW is much larger than that of OLR, and meridional gradients in the winter hemisphere are significantly more pronounced.
CS RSW is governed by TOA insolation and surface albedos. The latter have large seasonal cycles due to vegetation or snow/ice cover. For such a reason, the pattern of the CS RSW tends to follow the topography over land and is stratified by latitude over ocean. Figure 8a shows the CS RSW distribution for the ERBE 2-yr mean, and Fig. 8b is the difference from ERBE minus reanalysis. The NCEP global model employs the latitudinal-dependent sea surface albedo of Payne (1972). As a result, the reanalysis RSW data over the ocean tends to have less wavy features than that of ERBE.
In the low and midlatitudes, deserts tend to have the highest CS RSW. Figure 8a shows that more than 120 W m−2 is reflected over the Sahara and the Arabian Peninsula by ERBE. Tall mountain ranges, such as the Himalayas and Andes, are covered with snow all year-round and exhibit higher RSW than the surrounding plains. Still, the dominant effect of snow and ice occurs in the high latitudes, where the low solar angle significantly increases albedo (Larson and Barkstrom 1977) and enhances the CS RSW to large values, up to 300 W m−2 in the summer hemisphere.
2) Clear-sky RSW
As discussed in regard to the global average, section 4c(1), the reanalysis CS RSW is larger than ERBE by about 3.5%. Figure 7 depicts the time series of the zonal mean difference for CS RSW, ERBE minus reanalysis. It shows a strong seasonality in the higher latitudes, with ERBE larger than the reanalysis in the summer by up to 25 W m−2. The seasonality in the midlatitudes is also very evident with the reanalysis showing positive biases through the year. The 2-yr average (1985–86) geographical distribution in Fig. 8b further shows that most of this difference is over landmasses, either positive or negative biases, with regional biases ranging from −25 to 25 W m−2.
The reanalysis is higher than ERBE for most of the landmass between 30° and 70°N, and is lower between the equator and 30°N. Northern Africa, especially the Sahara Desert, is the most substantial landmass of the latter latitudinal belt, and the average difference is more than 20 W m−2. Also noted is that the reanalysis CS OLR in this region overestimates ERBE by more 15 W m−2 (Fig. 4b).
The combined information from CS OLR and CS RSW suggests two possible causes for the differences. 1) The absorption of solar radiation from the model desert surface may be too high, that is, the model-prescribed surface albedo is too low, such that the thermal emission is increased by the model to balance the absorbed flux. This point will be discussed further in section 4c on surface albedo. 2) Longwave emissivity may be too high in the reanalysis. The emissivity of sandy desert can be as low as 0.90, while the model uses 1.0. Reducing the model emissivity can substaintially decrease the reanalysis OLR. However, correct emissivity for the region has to be determined first.
Similarly, the reanalysis overestimates ERBE in CS OLR and underestimates CS RSW at Antarctica. At both poles, the reanalysis tends to underestimate ERBE in the summer seasons (Fig. 7) with a magnitude larger than 20 W m−2. The exact causes for these differences are yet to be determined. There are significant missing ERBE CS RSW data flagged out by the ambiguous scene identification between polar cloud and ice surfaces.
Other than the circumpolar ocean adjacent to the Antarctic region, the reanalysis agrees well with ERBE over the ocean. At the circumpolar ocean, the difference is generally less than 2 W m−2 for the annual means. The differences can be caused by inaccuracies in the sea ice and/or snow cover analysis used by the reanalysis (Kalnay et al. 1996).
3) Total-sky RSW
Figure 9 depicts the time series of the zonal mean difference for total-sky RSW, ERBE minus reanalysis. It shows a strong seasonality for every latitude. At the poles, the reanalysis underestimates ERBE by up to 40 W m−2 at summer. At the the lower latitudes, the reanalysis overestimates ERBE with the similar magnitudes. The sinusoidal pattern suggests that the biases are dependent on solar zenith angle.
The 2-yr average geographical distribution of total-sky RSW is mostly anticorrelated with OLR in the tropical oceans (Fig. 10a vs Fig. 6a), as both are modulated by the clouds. A similar argument can also be applied to the Amazon and the Congo. However, there is positive correlation between OLR and RSW in deserts, such as in northern Africa and northwest Australia. The positive correlation in these regions are due to the high reflectivities of the surface and high thermal emissions from the generally cloud-free surface.
The patterns in Fig. 10b of the RSW difference in mid- and low latitudes, ERBE minus the reanalysis, are also similar to that of OLR (Fig. 6b), but with opposite sign. This type of bias, that is, similarity in difference pattern but OLR (RSW) underestimated (overestimated) by the reanalysis, is an indication of excessive model clouds. The presence of cloud enhances RSW and reduces OLR. The regions, such as the central Pacific, the Atlantic just north and south of the ITCZ, the west-central Indian Ocean, and the Amazon, are where this type of bias is particularly pronounced.
At Indonesia, the reanalysis OLR overestimates ERBE (Fig. 6b) and the reanalysis RSW overestimates ERBE (Fig. 10b). The possible cause for such a combination of biases is that the reanalysis cloud in this region is too reflective and/or is misplaced to a lower altitude than observed.
d. Interannual variation
Since the objective of the reanalysis is to provide a long-term coherent database for studying climate variation, it is useful to demonstrate that true climate signals are well portrayed in the radiation flux data. An example of Kalnay et al. (1996) is followed for showing the interannual OLR and RSW differences from the 1985 to 1986 (Figs. 11a,b). In contrast to the 1987–88 case of Kalnay et al. (1996) with ENSO signal, there is no strong climate event for the current case. Thus this comparison reveals more subtle characteristics of databases that are not overwhelmed by the climate signals.
The comparison of OLR (Fig. 11a), shows that there is reasonable agreement in the synoptical-scale features. The central and western Pacific has the highest concentration of interannual variations. The reanalysis tends to show better-organized features with larger magnitudes than that of ERBE. It appears that the model resolution of T62 has filtered some of the smaller features observed by ERBE. The Indian subcontinent and Arabian Sea are regions where the reanalysis depicts a similar pattern, but with weaker magnitude of interannual variation than that of ERBE.
Driven by the interannual variations of the cloudiness, similar features to OLR also occur for RSW (Fig. 11b). However, the reanalysis RSW has less magnitude than ERBE, which may be caused by the specifications of the model cloud properties, or the cloud parameterization as a whole. Generally, both OLR and RSW from the reanalysis do clearly present the signals of interannual variation.
e. Surface albedo
In balancing the earth radiation budget, the surface albedo is as important as cloud reflectivity. It determines the solar energy reflected and absorbed by the surface. Part of the absorbed solar energy is then reemitted to the atmosphere as thermal radiation. The NCEP global model employs Matthews’s (1985) surface albedo over land, and Payne’s (1972) surface albedo over the oceans. Matthews’s surface albedo is based on the compilation of vegetation distribution and land-use data. Payne’s ocean surface albedo is an empirical derivation from the surface point observations.
The comparison dataset of Staylor and Wilbur’s (1990) surface albedo is derived from the clear-sky observations of ERBE, hereafter denoted as SW. Globally, the reanalysis surface albedo is positively biased over most of the land area. This bias also accounts for the positive biases in RSW in the global mean. Over the oceans, the surface albedo difference is within 0.03 (in albedo units) in the mid- and low latitudes. The differences near the poles are not shown due to substantial missing data from SW. It also appears that the spatial variation from the reanalysis is modulated by the surface elevation. The SW surface albedo is derived from space observation with less variance. Significant differences of this type can be detected around the Andes.
In the previous discussion, it has been suggested that some of the reanalysis biases in RSW and OLR are caused by the model surface albedo. This is notable in northern Africa, where RSW is negatively biased and OLR is positively biased. The comparison of the surface albedo, shown in Fig. 12, substantiates this argument. From SW, the maximum surface albedo in this region is more than 0.40 (in albedo units), while the reanalysis is about 0.30 (in albedo units), a difference of 25%. Other regions with similar types of biases, but with smaller magnitudes are northeast Australia and the Amazon. An adjustment of the model-prescribed albedo, with the corresonding correction of surface emissivity, could remedy the biases.
5. Summary
This study presents an evaluation of the radiation budget data from the NCEP–NCAR reanalysis. A comparison of TOA OLR and TOA RSW for 1985 and 1986 to the satellite measurements from the NASA ERBE is performed for the monthly data. The ERBE-derived data of Staylor and Wilbur (1990) are also utilized for comparing surface albedo. The study serves two objectives:(i) to document the general data quality of the reanalysis radiation budget, which scientific users may find useful in many applications, and (ii) to identify some of the general problems related to the reanalysis GDAS, which could be useful for future system improvement.
From the OLR comparison, it shows that the global annual-mean difference is only about 1.5%, with the reanalysis slightly higher than ERBE (237 vs 233.7 W m−2). The zonally averaged difference is highly seasonal dependent and is particularly evident at the high latitudes for the CS OLR. The seasonality also prevails at most latitudes for total-sky OLR. For the geographical distribution, the synoptic patterns are well defined. Yet there are regions with significant systematic biases, such as the Amazon Basin (+20 W m−2), Indonesia (−20 W m−2), and the subsidence regions of the Pacific and the Indian Oceans just north and south of the ITCZ (−20 W m−2). Possible causes are from shortcomings in the cloud/moisture parameterizations of the reanalysis GDAS. The complex topography, unresolvable by the T62 model, could also be cause for biases northeast of the Himalayas and east of the Andes.
The global-mean CS RSW from the reanalysis (54.9 W m−2) agrees well with that of ERBE (53.1 W m−2). However, the total-sky RSW from the reanalysis overestimates ERBE globally by a significant amount: 12.6 W m−2 (115.3 W m−2 for reanalysis vs 102.7 W m−2 for ERBE). The persistent overestimations of RSW throughout the subject period indicates that the global energy budget for the reanalysis is not balanced (Yang et al. 1997), as less solar energy is absorbed and slightly more terrestrial energy, OLR, exits the earth–atmosphere.
Since there is good agreement for CS cases, but not for total-sky cases, the results are also consistent with the above OLR comparison; that is, the reanalysis GDAS has possible shortcomings in the cloud/moisture parameterizations. Differences shown for RSW may also result from a deficiency in the shortwave parameterization in the reanalysis GDAS. NCEP is currently evaluating these results and experimenting with new parameterizations that will improve the future GDAS system, including prognostic cloud modeling using liquid water content and more spectral bands for solar radiation calculations.
Over the Sahara Desert, the reanalysis underestimates RSW and overestimates OLR, both in the clear-sky and total-sky conditions. This indicates that the prescribed surface albedo in the reanalysis is too low This result is confirmed by a comparison to the Staylor and Wilber (1990) derivation of surface albedo from ERBE clear-sky measurements. It suggests that GDAS surface albedo in this region can be increased by up to 0.1 (in albedo units).
The radiation budget data of the reanalysis portrays realistic climate signals as shown by a comparison of the interannual variations for the boreal summer. The reasonable OLR agreement was shown in the synoptic features, although the strength of the signal varied. The RSW result is similar to OLR, but with weaker signal than that of ERBE.
Acknowledgments
The authors appreciate Dr. Charles Woodlock and Ms. Nancy Rich of NASA/Langley Research Center for providing surface albedo data. Discussions with Drs. M. Kanamitsu and Julian Wang, and encouragement from Chet Ropelewski, are also appreciated. Comments from two anonymous reviewer and Mel Gelman are valuable for refining the original manuscripts. This work was supported by NASA/Langley Space Flight Center Grant L90988C.
REFERENCES
Ardanuy, P. E., L. L. Stowe, A. Gruber, M. Weiss, and C. L. Long, 1989: Longwave cloud radiative forcing as determined from Nimbus-7 observation. J. Climate,2, 766–799.
Barkstrom, B. R., E. Harrison, G. Smith, R. Green, J. Kibler, R. Cess, and ERBE Science Team, 1989: Earth Radiation Budget Experiment (ERBE) archival and April 1985 results. Bull. Amer. Meteor. Soc.,70, 1254–1262.
Bengtsson, L., and J. Shukla, 1988: Integration of space and in situ observations to study global climate change. Bull. Amer. Meteor. Soc.,69, 1130–1143.
Campana, K. A., Y.-T. Hou., K. E. Mitchell, S.-K. Yang, and R. Cullather, 1994a: Improved diagnostic cloud parameterization in NMC’s global model. Preprints, 10th Conf. on Numerical Weather Prediction, Portland, OR, Amer. Meteor. Soc., 324–325.
——, ——, ——, and ——, 1994b: Use of cloud analyses to validate and improve the global model cloud. Proc. ECMWF/GEWEX Workshop on Modeling, Validation and Assimilation of Clouds, Reading, United Kingdom, ECMWF/GEWEX, 207–231.
Cess, R., and J. Potter, 1987: Exploratory studies of cloud radiative forcing with a general circulation model. Tellus,39A, 460–473.
Ellingson, R. G., and Y. Fouquart, 1991: The intercomparison of radiation codes in climate models: An overview. J. Geophys. Res.,96 (D5), 8925–8928.
Grell, A., 1993: Prognostic evaluation of assumptions used by cumulus parameterizations. Mon. Wea. Rev.,121, 764–787.
Harrison, E. F., P. M. Minnis, B. R. Barkstrom, V. Ramanathan, R. D. Cess, and G. G. Gibson, 1990: Seasonal variation of cloud radiative forcing derived from the Earth Radiation Budget Experiment. J. Geophys. Res.,95 (D11), 18 687–18 703.
Hartmann, D. L., V. Ramanathan, A. Berroir, and A. Hunt, 1986: Earth radiation budget data and climate research. Rev. Geophys.,24, 439–468.
Hering, W. S., and T. R. Borden Jr., 1965: Mean distribution of ozone density over North America, 1963–1964. Environmental Research Rep. 162, USAF Cambridge Research Laboratory.
Higgens, R. W., Y. Yao, M. Chelliah, W. Ebisuzaki, J. E. Janowiak, C. F. Ropelewski, and R. E. Kistler, 1996: Intercomparison of the NCEP/NCAR and the NASA/DAO Reanalyses (1985–1993). NCEP/Climate Prediction Center Atlas 2, U.S. Dept. of Commerce/NOAA/National Weather Service, 169 pp.
Kalnay, E., and R. Jenne, 1991: Summary of the NMC/NCAR reanalysis workshop of April 1991. Bull. Amer. Meteor. Soc.,72, 1897–1904.
——, and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc.,77, 437–471.
Kanamitsu, M., 1989: Description of the NMC global data assimilation and forecast system. Wea. Forecasting,4, 334–342.
Kann, D. M., S.-K. Yang, and A. J. Miller, 1994: Mean meridional transport of energy in the earth-atmosphere system using NMC global analyses and ERBE radiation data. Tellus,46A, 553–565.
Koepke, P., and K. T. Kriebel, 1987: Improvements in the shortwave cloud-free radiation budget accuracy. Part I: Numerical study including surface anisotropy. J. Climate Appl. Meteor.,26, 374–395.
Lacis, A. A., and J. E. Hansen, 1974: A parameterization for the absorption of solar radiation in the earth’s atmosphere. J. Atmos. Sci.,31, 118–133.
Larson, J. C., and B. R. Barkstrom, 1977: Effects of realistic angular reflection laws for the earth’s surface upon calculation of the earth-atmosphere albedo. Proc. Symp. on Radiation in the Atmosphere, Science Press, 451–453.
Lindberg, C., and A. J. Broccoli, 1996: Representation of topography in spectral climate models and its effect on simulated precipitation. J. Climate,9, 2641–2659.
London, J., 1962: Mesospheric dynamics 3: The distribution of total ozone in the Northern Hemisphere. Dept. of Meteorology/Oceanography Final Rep., New York University.
Matthews, E., 1985: Atlas of archived vegetation, land-use and seasonal albedo data sets. NASA Tech. Memo. 86199, 53 pp. [Available from Goddard Space Flight Center, Greenbelt, MD 20771.].
Oort, A. H., and T. H. Vonder Haar, 1976: On the observed annual cycle in the ocean–atmosphere heat balance over the Northern Hemisphere. J. Phys. Oceanogr.,6, 781–800.
Parrish, D. F., and J. C. Derber, 1992: The National Meteorological Center’s spectral statistical interpolation analysis system. Mon. Wea. Rev.,120, 1747–1763.
Payne., R., 1972: Albedo of the sea surface. J. Atmos. Sci.,29, 959–970.
Ramanathan, V., 1987: The role of earth radiation budget studies in climate and general circulation research. J. Geophys. Res.,92 (D4), 4071–4095.
Reynolds, R. W., and T. M. Smith, 1994: Improved global sea surface temperature analyses using optimum interpolation. J. Climate,7, 929–948.
Sasamori, T., J. London, and D. Hoyt, 1972: Radiation budget of the Southern Hemisphere. Meteorology of the Southern Hemisphere, Meteor. Monogr., No. 35, Amer. Meteor. Soc., 9–23.
Schwarzkopf, M. D., and S. B. Fels, 1991: The simplified exchange method revisited: An accurate, rapid method of computation of integrated cooling rates and fluxes. J. Geophys. Res.,96 (D5), 9075–9096.
Sellers, W. D., 1965: Physical Climatology. University of Chicago Press, 272 pp.
Slingo, J. M., 1987: The development and verification of a cloud prediction scheme for the ECMWF model. Quart. J. Roy. Meteor. Soc.,113, 899–927.
Staylor, W. F., and A. C. Wilber, 1990: Global surface albedos estimated from ERBE data. Proc. Seventh Conf. on Atmospheric Radiation, San Francisco, CA, Amer. Meteor. Soc., 231–236.
Thompson, S., and S. Warren, 1982: Parameterization of outgoing infrared radiation derived from detailed radiative calculations. J. Atmos. Sci.,39, 2667; Addendum and Corrigendum, 40, 1859.
Trenberth, K. E., and C. J. Guillemot, 1995: Evaluation of the global atmospheric moisture budget as seen from analyses. J. Climate,8, 2255–2272.
Webb, M. J., A. Slingo, and G. L. Stephens, 1993: Seasonal variation of the clear-sky greenhouse effect: The role of changes in atmospheric temperatures and humidities. Rep. CRTN 41, Hadley Centre, U.K. Met Office, Bracknell, United Kingdom, 11 pp.
WCRP, 1993: Experts meeting on aerosols and their climate effects. Rep. WCP-55, 107 pp. [Available from WMO, Case Postale No. 2300, CH-1211, Geneva 2, Switzerland.].
Xie, P., and P. A. Arkin, 1996: Analyses of global monthly precipitation using gauge observations, satellite estimates, and numerical model predictions. J. Climate,9, 840–858.
Yang, S.-K., G. L. Smith, and F. L. Bartman, 1987: An earth outgoing longwave radiation climate model. Part I: Clear sky radiation. J. Climate Appl. Meteor.,26, 1134–1146.
——, X. Wang, and A. J. Miller, 1997: A revisit of global energy balance using NCEP/NCAR reanalysis and satellite observations. Proc. First Int. Conf. on Reanlayses, Silver Spring, MD. WCRP/WCRP, WCRP-104, WMO/TD-No. 876, 37–39.
Global-mean clear-sky and total-sky OLR for 1985 and 1986 from ERBE and the reanalysis.
Citation: Journal of Climate 12, 2; 10.1175/1520-0442(1999)012<0477:EOTERB>2.0.CO;2
Global-mean clear-sky and total-sky TOA reflected shortwave radiation for 1985 and 1986 from ERBE and the reanalysis.
Citation: Journal of Climate 12, 2; 10.1175/1520-0442(1999)012<0477:EOTERB>2.0.CO;2
Zonally averaged time series of the clear-sky OLR difference from ERBE minus the reanalysis. Hatched area is positive; stippled area <−15 W m−2.
Citation: Journal of Climate 12, 2; 10.1175/1520-0442(1999)012<0477:EOTERB>2.0.CO;2
(a) ERBE geographical distribution of mean clear-sky OLR averaged from 1985 and 1986. (b) The difference of clear-sky OLR from ERBE minus the reanalysis. Orange area is > 5 W m−2; yellow area <−5 W m−2. Other than ±1 W m−2 contour lines, the contour interval is 5 W m−2.
Citation: Journal of Climate 12, 2; 10.1175/1520-0442(1999)012<0477:EOTERB>2.0.CO;2
Zonally averaged time series of the total-sky OLR difference from ERBE minus the reanalysis. Hatched area >5 W m−2; stippled area <−5 W m−2.
Citation: Journal of Climate 12, 2; 10.1175/1520-0442(1999)012<0477:EOTERB>2.0.CO;2
(a) ERBE geographical distribution of mean total-sky OLR averaged from 1985 and 1986. (b) The difference of total-sky OLR from ERBE minus the reanalysis. Orange area >10 W m−2; yellow area <−10 W m−2; the contour interval is 10 W m−2.
Citation: Journal of Climate 12, 2; 10.1175/1520-0442(1999)012<0477:EOTERB>2.0.CO;2
Zonally averaged time series of the clear-sky RSW difference from ERBE minus the reanalysis. Stippled area <−5 W m−2; hatched area >5 W m−2. Other than ±1 W m−2 contour lines, the contour interval is 5 W m−2.
Citation: Journal of Climate 12, 2; 10.1175/1520-0442(1999)012<0477:EOTERB>2.0.CO;2
(a) ERBE geographical distribution of mean clear-sky RSW averaged from 1985 and 1986; contour interval is 20 W m−2. (b) The difference of clear-sky RSW from ERBE minus the reanalysis. Orange area is larger than 5 W m−2; yellow area is less than −5 W m−2. Other than ±1 W m−2 contour lines, the contour interval is 5 W m−2.
Citation: Journal of Climate 12, 2; 10.1175/1520-0442(1999)012<0477:EOTERB>2.0.CO;2
Zonally averaged time series of the total-sky RSW difference from ERBE minus the reanalysis. Stippled area is less than −10 W m−2; hatched area is larger than 10 W m−2.
Citation: Journal of Climate 12, 2; 10.1175/1520-0442(1999)012<0477:EOTERB>2.0.CO;2
(a) ERBE geographical distribution of mean total-sky RSW averaged from 1985 and 1986; contour intervals is 15 W m−2. (b) The difference of total-sky RSW from ERBE minus the reanalysis. Hatched area is larger than 10 W m−2; stippled area is less than −40 W m−2. The contour interval is 10 W m−2.
Citation: Journal of Climate 12, 2; 10.1175/1520-0442(1999)012<0477:EOTERB>2.0.CO;2
Fig. 11a. Difference of average OLR during June–August 1986 and June–August 1987: (top) ERBE; (bottom) the reanalysis. Contour interval is 5 W m−2.
Citation: Journal of Climate 12, 2; 10.1175/1520-0442(1999)012<0477:EOTERB>2.0.CO;2
Fig. 11b. Same as Fig. 11a but for RSW.
Citation: Journal of Climate 12, 2; 10.1175/1520-0442(1999)012<0477:EOTERB>2.0.CO;2
July 1985 surface albedo (%): (top) ERBE, (center) the reanalysis, and (bottom) difference from ERBE minus the reanalysis.
Citation: Journal of Climate 12, 2; 10.1175/1520-0442(1999)012<0477:EOTERB>2.0.CO;2
OLR (in W m−2) for the ICRCCM atmospheric profiles, from Schwarzkopf and Fels (1991, SF) and line-by-line (LBL) calculations. The tropical case is denoted as T, the midlatitude summer case as MLS, the midlatitude winter case as MLW, the sub-Arctic summer case as SAS, and the sub-Arctic winter case as SAW.
Biannual global-mean TOA fluxes from the reanalysis and ERBE for 1985–86: LWCF: longwave cloud forcing, and SWCF:shortwave cloud forcing.
