The properties of the marine boundary layer produced by the National Center for Atmospheric Research (NCAR) Community Climate Model version 3 (CCM3) are compared with observations from two experiments in the central and western equatorial Pacific. The main objective of the comparison is determining the accuracy of the ocean–atmosphere fluxes calculated by the model. The vertical thermodynamic structure and the surface fluxes calculated by the CCM3 have been validated against data from the Central Equatorial Pacific Experiment (CEPEX) and the Tropical Ocean Global Atmosphere–Tropical Atmosphere Ocean (TOGA–TAO) buoy array. The mean latent heat flux for the TOGA–TAO array is 92 W m−2, and the model estimate of latent flux is 109 W m−2. The bias of 17 W m−2 is considerably smaller than the overestimation of the flux by the previous version of the CCM. The improvement in the latent heat flux is due to a reduction in the surface winds caused by nonlocal effects of a new convective parameterization. The agreement between the mean sensible heat flux for the TOGA–TAO array and the model estimate has also been improved in the new version of the model. The current version of the CCM overestimates the sensible heat flux by 3.4 W m−2. The atmospheric temperature and water vapor mixing ratio from the lowest model layer are within 0.3 K and 0.4 g kg−1 of measurements obtained from radiosondes. The mean model value of the boundary layer height is within 13 m of the average height derived from a Raman lidar on board a ship in the CEPEX domain. There is some evidence that the biases in the model can be reduced further by modifying the bulk formulation of the surface fluxes.
Evaluation of the performance of global climate models is essential for the continued improvement in climate prediction. Current modeling activity is focused on the coupling of global atmospheric models with either regional ocean models and/or global ocean circulation models. A region of particular interest is the tropical Pacific basin. The tropical Pacific sea surface temperatures exhibit large interannual variability. It has been established that variations in the Pacific SSTs and precipitation patterns have profound climate implications for the Tropics and most likely the extratropics. Thus, it is of great importance for climate models to realistically simulate the state of the atmosphere in this region. The forcing of the ocean–atmosphere system is accomplished through three processes: the flux of energy (most importantly solar and latent heat), the flux of momentum, and the flux of freshwater. The physical realism of a coupled atmosphere–ocean model rests, to a large extent, on the ability of the coupled model to realistically simulate these fluxes of energy, momentum, and freshwater.
In the present study, we present a comparison of the simulated climate from the latest version of the National Center for Atmospheric Research (NCAR) Community Climate Model version 3 (CCM3) with observational data from the tropical Pacific. The measurements are obtained from the Central Equatorial Pacific Experiment (CEPEX) field campaign and the Tropical Ocean Global Atmosphere–Tropical Atmosphere Ocean (TOGA–TAO) array. The CEPEX data offer a unique collection of observations on the structure of the marine boundary and surface layers in the central Pacific. A companion study will address the accuracy of the radiative budget of CCM3 in the equatorial Pacific region. CEPEX and the TOGA Coupled Ocean–Atmosphere Response Experiment (TOGA COARE) offer comprehensive data for the evaluation of the important physical processes that operate in the Pacific. We show that these data can be used to quantitatively assess climate simulations. We also compare results from a previous version of the CCM that employed a different convection parameterization. Thus, we show the important role the convective processes play in determining the surface and boundary layer structure within the model.
The paper is organized as follows: Section 2 briefly describes the CEPEX and TOGA–TAO observations. Section 3 describes the physical processes in the CCM3 and outlines the framework for the model simulations. Section 4 presents comparisons of the observed and modeled surface fluxes of moisture, sensible heat, and momentum. Section 5 describes the vertical structure of the marine boundary layer from both observations and the CCM3. Section 6 summarizes the remaining biases in CCM3 and discusses the probable cause of these remaining biases. Finally, section 7 draws general conclusions from this study and suggests future areas of improvement in the CCM3.
2. CEPEX observations
The observations used in this analysis were collected during CEPEX. The scientific objectives and experimental design are described in detail in Ramanathan et al. (1993). The principal goal of CEPEX was to test the cirrus thermostat hypothesis (Ramanathan and Collins 1991) and other proposed mechanisms for regulation of tropical SSTs [e.g., Wallace (1992); Hartmann and Michelsen (1993); for a summary see Waliser (1996)]. The experiment began on 7 March 1993 and ended on 6 April 1993. This analysis is based upon experimental estimates of the sensible and latent heat fluxes at the ocean surface, measurements of the dynamic and thermodynamic fields used to calculate these fluxes, and vertical profiles of the atmospheric boundary layer. The surface fluxes were derived from measurements from the ship R/V John Vickers and from the TOGA–TAO buoy array (McPhaden 1995). Vertical profiles of wind, temperature, and humidity were measured with several types of radiosondes launched from the R/V Vickers. High-resolution profiles of water vapor mixing ratio were also derived from a Raman lidar system on the R/V Vickers. The R/V Vickers traversed the CEPEX experimental domain as it sailed from 160°E to 160°W at 2°S. A detailed description of the location and times of the aircraft and ship observations is given in Williams (1993).
The buoy data used in this study are from the TOGA–TAO buoy array in the equatorial Pacific (McPhaden 1993). For the 31 buoys from 10°N to 10°S and 120°E to 140°W, we have extracted the segment of the observational time series corresponding to the CEPEX field campaign. The surface sensible and latent heat are estimated from bulk formulas using daily mean SST, surface air temperature, wind speed, and relative humidity using the procedures in Zhang and McPhaden (1995). The instrumental errors in wind, air temperature, relative humidity, and SST measurements are 0.2 m s−1, 0.2°C, 2%–4%, and 0.03°C, respectively. The largest uncertainties in the latent heat flux derived from these measurements is 12–18 W m−2 and is associated with the upper range of errors in relative humidity (Zhang and McPhaden 1995). Fluxes computed with hourly mean observations are generally larger than fluxes computed from daily mean fields, but the differences do not significantly affect the comparison with the model results. Monthly average fluxes computed from hourly mean and daily mean fields differ by less than 10% in the CEPEX region (Zhang and McPhaden 1995). The effects of fluctuations in the wind velocity and thermodynamic fields on timescales shorter than 1 h are negligible (Esbensen and McPhaden 1996). The differences between fluxes calculated with hourly and daily mean fields are considered further in section 4a.
The vertical profiles of temperature, humidity, and wind are from Vaisala sondes launched approximately every 6 h (at 0000, 0600, 1200, and 1800 UTC) from the R/V Vickers. A detailed description of the sonde instrumentation, calibration, and processing is given in Kley et al. (1997). The data are accurate to 1 mb in pressure, 1 K in temperature, and 5% in relative humidity up to 250 mb (Weaver et al. 1994). The humidity in the boundary layer has been corrected for heating of the sensors by incident sunlight that results in an underestimate of the true humidity. The correction increases the humidity by approximately 10% near the surface and decays to zero between 900 and 950 mb. The resulting values of specific humidity in the boundary layer are 2–3 g kg−1 larger than those determined without the correction. Based upon an intercomparison of two independently calibrated sets of soundings launched from the R/V Vickers, the instrumental uncertainty in estimates of water vapor mixing ratio is approximately 1 g kg−1 in the boundary layer.
A Raman lidar system on the R/V Vickers was used to measure the vertical profile of water vapor mixing ratio with high spatial and temporal resolution. The height of the boundary layer (or mean entrainment zone height) has been derived from the variation of water vapor with altitude above the ocean surface (Cooper et al. 1996). The boundary layer heights are derived from profiles averaged over 45 min of data. Due to the experimental nature of the lidar installation on the R/V Vickers, it was not possible to determine a continuous time series of boundary layer heights. The lidar provided 66 estimates of the boundary layer height between 11 March and 21 March 1993. The design and performance of the Raman lidar are given in Eichinger et al. (1994). The lidar vertical resolution was 1.5 m in the near field (altitudes below 2 km) and 75 m in the far field (altitudes between 2 and 10 km). The duration for each measurement ranged from 3.5 to 7 s. Agreement between values of the vapor mixing ratio from coincident radiosonde and Raman lidar data was generally within 0.25 g kg−1. Within the boundary layer, the systematic bias of the lidar relative to the radiosondes is −0.26 g kg−1.
The outgoing longwave radiation (OLR) has been derived from the infrared imaging instrument on board the Japanese geostationary meteorological satellite GMS-4. The estimation of OLR is based upon a calibration of the satellite against measurements of the upwelling hemispherical longwave flux from the NASA ER-2 aircraft flying in the lower stratosphere (Collins et al. 1997). The spatial resolution of the OLR is 0.25°, and the temporal resolution is hourly. The mean bias in the OLR relative to the aircraft data is less than or equal to 4 W m−2, and the instantaneous rms error is less than approximately 10 W m−2.
3. Description of CCM3
We provide a brief description of the NCAR CCM3;a more detailed description can be found in Kiehl et al. (1996). A number of changes in the physical parameterizations have occurred in the development of CCM3. Many of these changes are in parameterizations of moist physical processes and radiative processes. A major difference between CCM3 and CCM2 is the parameterization of deep convection. CCM3 employs the deep convective scheme of Zhang and McFarlane (1995). This scheme is a penetrative mass flux scheme that includes the effects of saturated downdrafts. The convective-scale updraft is comprised of an ensemble of plumes, and the occurrence of convection is conditioned upon the local convectively available potential energy (CAPE). CCM3 also includes a revised planetary boundary layer scheme, which accounts for the dependence of the ocean roughness length on surface wind speed (Bryan et al. 1996). The calculation of the boundary layer height has been modified to provide more realistic estimates of boundary layer depth (Vogelezang and Holtslag 1996).
The CCM3 also includes a parameterization of convective cloud cover that depends on cloud mass flux. CCM2 employed a specified distribution of cloud water (Kiehl et al. 1994). CCM3 now includes a locally diagnostic formulation. The improvements to the radiation parameterizations include a differentiation of cloud drop size between maritime and continental conditions (Kiehl 1994); an explicit representation for cloud ice radiative properties; the radiative effects of CH4, N2O, minor CO2 bands, CFC-11, and CFC-12; and inclusion of a background aerosol in the boundary layer.
The dynamical framework in CCM3 remains the same as that in CCM2. Thus, the model solves the dynamical equations in spectral space, while moisture advection is handled with a semi-Lagrangian transport method. The nominal resolution of the model is a horizontal spectral truncation of T42 (equivalent Gaussian grid of 2.9° × 2.9°) and 18 levels in the vertical.
Model results employed in the present study used observed SSTs through the TOGA COARE and CEPEX time periods. The model simulation began in 1979 and was run through September of 1993. Thus, model results for the time period of this study are not affected by initial conditions. The focus of the present work is on the time mean state of the marine boundary layer. Thus, we do not attempt to address issues related to the transient response of the simulated climate to transient behavior in SSTs.
4. Comparison of surface fluxes
In this section, fluxes computed by the CCM3 are compared with CEPEX observations. A comparison of the latent heat fluxes is presented in section 4a, and the sensible and momentum fluxes are discussed in section 4b and 4c. In section 4d, the variation of the surface fluxes with a measure of convective activity is derived from the CEPEX data and compared with the corresponding variation computed by the CCM3.
a. Surface latent heat flux
In the CCM3 parameterization for the atmospheric boundary layer, the surface latent heat flux (Fq) is calculated using a bulk formula given by Kiehl et al. (1996):
Here, Lυ is the latent heat of evaporation, ρ is air density, Ce is the turbulent moisture exchange coefficient, V is the horizontal wind speed, and q̃ is the specific humidity. The subscripts s and 1 refer to values at the surface and at the lowest model level, respectively. The saturated q̃ at SST is denoted by q̃*s. A similar bulk thermodynamic formula is used to calculate Fq for the buoy data (Zhang and McPhaden 1995):
Here, q denotes the water vapor mixing ratio. In order to simplify the comparison of the model with observations, humidity is expressed in terms of vapor mixing ratio in the remainder of the analysis. For the buoy data, q is measured at 3 m from sea surface, and wind speed V is measured at 4 m. As noted in section 2, a recent study by Zhang and Grossman (1996) indicates that the estimates of Fq from (2) are consistent with estimates derived from eddy correlation methods for CEPEX and TOGA COARE and with the COARE bulk flux algorithm (Fairall et al. 1996).
The exchange coefficient Ce is determined in slightly different ways in (1) and (2). The roughness length for momentum is a different function of velocity for the model and observations. The roughness length for moisture in the model is approximately 5 times larger than the corresponding roughness length used for the buoy data. Finally, the stability functions denoted by ψm and ψh in Zhang and McPhaden (1995) are 40% larger than the corresponding functions in the model under stable conditions. The velocity and thermodynamic fields are scaled to the same reference height of 10 m in both bulk formulations.
The meridionally averaged Fq from the CCM3 and buoy observations is shown in Fig. 1 as a function of longitude. For the buoy data, the average at each longitude includes the observations collected between 10°N and 10°S during the CEPEX time period. For the CCM3, the averages include the daily mean model fields closest to the individual buoy measurements sites and coincident with the observation period. From Fig. 1, it is evident that the model overestimates Fq by an average of 17 W m−2 between 155°E and 150°W relative to the buoy observations. This overestimate of Fq increases to approximately 80 W m−2 around 140°W. The average value of Fq from the buoy measurements is 92.0 W m−2, and the average value of the coincident estimates of Fq from the CCM3 is 109.0 W m−2. The magnitude of the differences decreases slightly if Fq is computed from hourly buoy measurements. However, the sign of the difference between the fluxes from the model and observations is unaffected by the temporal resolution of the buoy data. The annual mean estimate of Fq increases by 2.8 W m−2 at 140°W and by 13.6 W m−2 at 165°E, relative to estimates derived with daily mean measurements (Zhang and McPhaden 1995). As shown in Fig. 2, the offsets between fluxes computed using hourly and daily mean observations for the CEPEX time period are comparable.
If we neglect the difference in air density between 3 m above sea level and the lowest model level (approximately 67 m above sea level), it is clear from (1) and (2) that three factors can contribute to the Fq difference between the model and buoy observations. These factors are differences in wind speed V, differences in the exchange coefficient Ce, and differences in the humidity deficit (q*s − q) between the model and observations. It is important to note that the lowest atmospheric layer in the model and various sensors on the buoys are not at the same height above sea level. The lowest model level in the atmosphere is located at about 67 m above the sea surface, while the buoy measurements are collected between 3 and 4 m above the sea surface. The model temperature and humidity fields have been evaluated using radiosonde measurements at the same altitude as the lowest model layer. No reliable vertical profiles of horizontal winds are available that coincide with the ship and buoy observations. The model winds have been scaled to the height of the buoy wind sensors using Monin–Obukhov scaling theory to facilitate comparison.
The model produces a reasonable surface wind speed compared to the buoy observations over most of the CEPEX domain (Fig. 3). The average value of the wind speed V1 in the lowest atmospheric layer in CCM3 is 4.75 m s−1, and the average wind speed measured by the buoys is 4.64 m s−1. The mean model wind scaled to the same height as the buoy wind sensors is 4.01 m s−1. The model wind speed is much higher than the buoy observations in the eastern Pacific near 140°W, but this is the only region where the difference in the mean wind speeds is statistically significant. The fact that model wind speed is larger than the buoy observations by 7 m s−1 is the main reason why the model value of Fq is much higher than observations around 140°W (Fig. 1). But at other longitudes, the differences in surface wind speed do not contribute significantly to the overestimation of Fq by the CCM3.
The variation of the moisture exchange coefficient Ce with wind speed is shown in Fig. 4 for buoy observations and collocated CCM3 output. The Ce coefficient for CCM3 is derived by applying the model formulation for the surface fluxes to the model output. The Ce coefficient for buoy observations is derived using the formulae given in Zhang and McPhaden (1995). The buoy coefficient is evaluated at a height of 67 m to ensure compatibility with the model. From Fig. 4, it can be seen that the Ce for the buoy data and the model diverge for wind speeds below 5 m s−1, where Ce for the model increases more rapidly with decreasing wind speed than Ce for the buoys. The differences between the model and buoy values of Ce are statistically significant at all wind speeds measured by the buoy array during CEPEX. The differences in the Ce coefficients tend to increase the Fq calculated by the model relative to the estimates from the TAO array.
Since Ce depends on both the surface wind speed and the stability of the surface air, two factors can lead to differences in the exchange coefficients even if the surface wind fields are identical. These factors are different formulas for calculating Ce and different static stability of the surface air in the model and observations. In order to examine the effects of the static stability, we have calculated Ce from the output of CCM3 using the bulk formulation of Zhang and McPhaden (1995). Comparison of these test coefficients with the Ce derived for the buoy observations for the same surface wind speeds yields the effects on Ce of different static stability in the model and observations. It is found (not shown here) that the test values Ce are almost identical to the Ce derived from the buoy measurements. This suggests that the differences in the buoy and model exchange coefficients shown in Fig. 4 are caused mainly by differences in the bulk formula for Ce.
To estimate the effect of different bulk formulas for Ce on the calculation of Fq, the Fq calculated using the CCM3 fields in (2) but with the formula of Zhang and McPhaden (1995) is compared with the latent heat flux from the buoys in Fig. 5. The figure shows that the differences between the model and observed Fq are reduced when we use the formula of Zhang and McPhaden (1995), instead of the current CCM3 scheme, to calculate Fq. The mean value of Fq computed by applying (2) to the CCM3 fields is 99.3 W m−2. The overestimation of Fq has been reduced from 18.5% to 7.9% by changing the bulk formulation. It should be noted, however, that the improvement in Fq may not be reproduced exactly if the modifications in the bulk formulation are run interactively with the remainder of the model physics. It is likely that coupling between changes in the latent heat flux and the large-scale circulation will alter Fq from the values reported here.
Differences between the simulated and observed vapor mixing ratio can also introduce biases in the model value of Fq. The mean value of q between 60 and 70 m above sea level from the Vickers radiosondes is 17.1 ± 1.0 g kg −1, and the mean model value is 17.5 ± 0.2 g kg−1. Errors in the simulation of q itself are not directly contributing to the overestimation of Fq by the model. However, the mean q*s for the buoys is 25.19 ± 0.03 g kg−1 and for the model is 21.7 ± 0.03 g kg−1. This difference in q*s between model and observations is a result of using an approximate expression for the saturation mixing ratio over ocean (Bryan et al. 1996) that was implemented in CCM3 to facilitate coupling within the NCAR Climate System Model. The CCM3 employs an accurate table lookup for q*s, and in future versions of the CCM this table will also be used for surface saturation mixing ratios, as it was in previous versions of the CCM.
b. Surface sensible heat flux
In the CCM3, the surface sensible heat flux (Fh) is calculated with the bulk formula
Here, cp is the specific heat at constant pressure, Ch is the turbulent heat exchange coefficient, θ is potential temperature, and the remainder of the notation is identical to (1). The subscripts s and 1 refer to values at the surface and at the lowest model level, respectively. The sensible heat flux is estimated from the buoy measurements using
where Ts is the sea surface temperature and T is the surface air temperature, which is measured at 3 m on the TAO buoys. Since the moisture and thermal roughness constants have been set to the same value, the exchange coefficients Ce and Ch in (2) and (4) are identical (Zhang and McPhaden 1995). The thermal roughness length in CCM3 has two values depending on stability. Under unstable conditions, the roughness length is approximately 2.5 times larger than the roughness length used for the buoys. The bulk formula for the buoy data is expressed in terms of temperature because, to a very good approximation, θ ≃ T at the height of the buoy instruments above sea level. For comparison with the CCM3, θ is used in place of T in (4).
The meridional average values of Fh derived from buoy observations and collocated CCM3 output are shown in Fig. 6. At most longitudes, the model overestimates Fh by 3 to 4 W m−2 compared to the buoy observations. The mean value of Fh from the buoy observations is 6.1 W m−2, and the mean value of Fh calculated by the CCM3 is 9.5 W m−2. The effects of differences in ρ, Ts, and surface pressure between CCM3 and buoy observation are negligible. Since the model underestimates the mean surface wind by 0.63 m s−1, the differences in V cannot explain the overestimation of Fh. The principal factors that determine the offset between the buoy and model values of Fh are differences in the bulk formulations and differences in air temperature T.
The effects of the bulk formulation are examined first. If we derive the flux using the bulk formula applied to the buoys [(4)], the model flux decreases to 8.0 W m−2. The reduction in Fh is related to the differences in the turbulent drag coefficients discussed in section 4a. However, the model value of Fh is still approximately 30% larger than the buoy value of Fh. Two factors related to the temperature difference between CCM3 and buoy observations contribute to the offset remaining between the model and observed Fh. The CCM3 produces a temperature profile that is less than the measured profile in the boundary layer (section 5). At the lowest model level, the temperature calculated by the CCM3 is about 0.3 K colder than the collocated radiosonde observations. For a fixed value of SST, colder air temperature will increase Fh. In addition, because the temperature at the lowest model level is colder than buoy measurements of T, the surface layer in the model is more unstable. The greater instability also tends to increase the surface sensible heat flux.
c. Momentum flux
In CCM3, the surface momentum flux (Fm) is computed from the meridional and zonal components of the flux:
where Cd is the momentum exchange coefficient and the remainder of the notation is identical to (1). The magnitude of the surface momentum flux is given by Fm = (Fmx)2 + (Fmy)2. The flux is computed from the buoy observational data using the same formula but with a different exchange coefficient adopted from Zhang and McPhaden (1995).
A comparison of the averaged surface momentum flux from buoy observations and the collocated CCM3 is shown in Fig. 7. It can be seen from Fig. 7 that the surface momentum flux of CCM3 is in reasonable agreement with buoy observations over most of the CEPEX region. The main exception is in the eastern equatorial Pacific near 140°W, where the momentum flux calculated by the model is considerably higher than the observations. The main reasons for the large bias in Fm in the eastern Pacific are the overestimation of the wind speed (section 4a) and the quadratic dependence of Fm on V. The mean value of Fm from the buoy observations is 42.64 N m−2, and the mean value of Fm from the CCM3 is 38.08 N m−2, a difference of 10.7%. The factors contributing to the difference between the mean values of Fm from the CCM3 and observations are differences in the air temperature T, wind speed V, and the calculation of the coefficient Cd. The effects of differences in ρ, Ts, q, and surface pressure between CCM3 and buoy observations are negligible and will not be examined further.
The average momentum exchange coefficient Cd for buoy observations and collocated CCM3 output is plotted as a function of wind speed in Fig. 8. The buoy coefficient is evaluated at a height of 67 m to ensure compatibility with the model. From Fig. 8, it can be seen that the Cd calculated for the CCM3 and buoys begin to diverge from when the surface wind speed falls below approximately 8 m s−1. The value of Fm calculated using the CCM3 fields in the Zhang and McPhaden (1995) bulk formulation is 37.03 N m−2, or 11% smaller than the buoy estimate. The CCM3 surface wind speed contributes a tendency to underestimate Fm while the momentum exchange coefficient and surface air temperature contribute a tendency to overestimate Fm.
d. Relations between surface fluxes and convective activity
The relation between surface fluxes and convection may be examined using the OLR as an index for convective activity. The observational study of Gaynor and Ropelewski (1979) for Global Atmospheric Research Program Atlantic Tropical Experiment (GATE) data indicated that there is a clear increase in Fh during the transition from undisturbed to convectively disturbed conditions. Analysis of a COARE pilot cruise by Young et al. (1992) has shown that Fm and Fq also increase under disturbed conditions. The physical explanation for the increase in Fh can be attributed to downdrafts, which cool the PBL. As noted in section 4b, lowering the temperature in the PBL increases the ocean–air temperature differential and the instability of the atmosphere near the ocean surface. Both of these effects tend to increase Fh. One of the main mechanisms for the increase in Fq are strong gusts associated with mesoscale convective downdrafts (Jabouille et al. 1996). It is important to note that the results of Gaynor and Ropelewski (1979), Young et al. (1992), and Jabouille et al. (1996) are for local spatial and temporal scales. When averaged over several years, the latent heat fluxes and wind speeds derived from the TOGA–TAO array are actually anticorrelated with OLR (Zhang et al. 1995). The CEPEX data has been analyzed to see if either of these effects can be observed in the buoy measurements collected during the experiment and, more importantly, if the model can reproduce these effects.
Figure 9 shows the averaged Fh versus OLR for CCM3 and observations. The observations show that Fh increases slowly with increasing convective activity (decreasing OLR). This relation is consistent with the results of Gaynor and Ropelewski (1979), although it is not evident in the CCM3 output. Figure 10 shows the mean values of Fq as a function of OLR from both the CCM3 and the CEPEX measurements. For the observations, Fq is calculated from the buoy data binned against the OLR from collocated GMS-4 satellite imagery. The relation of Fq and OLR for the CCM3 is derived from the model fields for the entire CEPEX domain and time period. Figure 10 indicates that Fq tends to increase with decreasing OLR in the observations, especially at lower OLR. However, the model indicates that latent heat flux decreases on average with increasing convective activity. One possible reason why the model produces a different local relation between OLR and surface fluxes is that the downdrafts in the convective parameterization do not directly alter the surface wind fields. However, the model results are in qualitative agreement with the long-term average of the buoy observations. There are some systematic differences in the relation between Fq and the dynamical fields for the CEPEX time period compared to longer observational records, and these differences have been discussed elsewhere (Zhang and Grossman 1996). Further analysis of the relation of Fq and convection produced by the CCM3 will the subject of a future study using data from the TOGA COARE Intensive Flux Array (IFA).
5. Analysis of the vertical structure of the boundary layer
Figure 11 shows the averaged vertical profiles of temperature in the PBL for Vickers radiosondes and the collocated CCM3 output. Compared to observations, CCM3 produces a vertical temperature profile 0.5 K cooler than the soundings throughout the PBL. The modeled and observed specific humidity agree to within the instrumental uncertainties in the lowest model layer (not shown). In the middle and upper portion of the PBL, the modeled q exceeds the observations by more than the instrumental uncertainty of 1 g kg−1. The largest offset between model and observations occurs in the middle level of the PBL.
The planetary boundary layer height (PBLH) is a very important quantity in the boundary layer parameterization of the atmosphere (Vogelezang and Holtslag 1996). CCM3 diagnostically calculates the PBLH as a function of the vertical wind shear and static stability of the PBL atmosphere. The updated PBLH is used to calculate a nonlocal eddy diffusivity for heat, water vapor, and passive scalars in the boundary layer (Holtslag and Boville 1993). This nonlocal diffusivity is introduced to account for transport under unstable or convective conditions when the scale of the largest turbulent eddies is comparable to the PBLH. During CEPEX, the PBLH was determined from lidar observations of the vertical profile of specific humidity (Cooper et al. 1996). The time series of the ship observations (not shown) shows that there is almost no diurnal variation of the PBLH. This is consistent with the small variations in the PBLH of 75 m (15% of the total) observed during TOGA COARE (Johnson and Dickey 1996). The fact that the diurnal variations are relatively small makes it possible to compare the 66 instantaneous PBLH observations with the daily mean estimates from the CCM3. Due to the small number of estimates of PBLH from the ship lidar, we have limited the comparison to the mean values of PBLH for the CEPEX time period. The average PBLH for the ship is 625 ± 12 m, and the average PBLH for the model is 604 ± 22 m. This represents a considerable improvement over the previous version of the model, as discussed in section 6. The agreement of the ship and model values suggests that biases in the boundary layer structure are probably not related to the nonlinear coupling between the eddy diffusivity, counter gradient transport, and PBLH.
6. Summary of biases: Role of convection in surface fluxes
a. Effects of changing convective parameterizations
In order to illustrate the sensitivity of the surface fluxes to convection, we have also analyzed the simulation for the CEPEX region from a previous version of the CCM (CCM2+). One of the principal differences between CCM2+ (as well as CCM2) and CCM3 is the moist adjustment physics, specifically the parameterization for deep convection (section 3). The results from CCM2+ have been compared to CEPEX observations of wind speed and surface fluxes using the approach adopted in section 4. The change in the convective parameterization has significantly altered the large-scale circulation and surface winds over the central and western Pacific (Zhang et al. 1997, manuscript submitted to J. Climate).
The variation of surface wind speed V and latent heat flux Fq with longitude are shown in Figs. 12 and 13, respectively. When the model winds are scaled to 10 m, the model overestimates the surface wind speed by 2 to 3 m s−1 over most of the central equatorial Pacific, and it overestimates the surface moisture flux by 40 to 50 W m−2 relative to the buoy observations. Although the higher moisture exchange coefficient Ce in the model contributes to the overestimation of Fq, the high surface wind speed in the model is the main factor in the excess latent heat flux. The high surface winds also result in significant overestimation of the surface sensible heat flux and momentum flux in CCM2+ (not shown). These results are consistent with the stronger Walker circulation produced by the CCM2/2+ relative to CCM3.
The average vertical temperature profiles in the PBL from the CCM2+ and the Vickers radiosondes are shown in Fig. 14. A comparison of Fig. 14 with Fig. 11 shows that the vertical temperature profile in CCM2+ is approximately 0.5 K warmer than that of CCM3. One possible reason for this difference is the cooling effect associated with evaporation in convective downdrafts. A model of convective downdrafts is explicitly included in CCM3 but not in previous versions of the model. The effects of the downdrafts can also be detected in the depth of the PBL. The CCM2+ overestimates the PBL height by about 100 m relative to the lidar observations. Thus in the CEPEX region, the PBL height calculated by the CCM2+ is about 200 m higher than that calculated by the CCM3. This difference can be partially attributed to the convective downdrafts in CCM3, which tend to cool the PBL and lower the PBL height.
Comparison of the average vertical moisture profiles from CCM2+ and CCM3 shows that both models predict more moisture than is observed in the PBL, but the moisture bias is larger for the CCM2+. This may partially be due to the fact that the surface moisture flux in CCM2+ is larger than the moisture flux in CCM3. It may also show that the new parameterization for deep convection leads to more efficient transport of moisture out of the PBL than the previous moist adjustment scheme. The differences between the humidify profiles produced by the two models may also arise from modifications to the boundary layer parameterization.
b. Bulk parameterizations for free convection
Under conditions of low mean wind speed, the near- surface wind induced by large eddies in free convection (Beljaars 1995) and downdrafts associated with deep convection (Jabouille et al. 1996) can prevent the local wind speed from vanishing. Both Beljaars (1995) and Jabouille et al. (1996) suggest adding a correction to the horizontal wind speed in the lowest model layer V1 of the form
The CCM3 does not explicitly incorporate the effects of free convection and gustiness in the bulk flux parameterization in the current version. The changes in the simulation introduced by the Beljaars (1995) modification have been calculated for the CEPEX time period. When the Jabouille et al. (1996) parameterization is applied to European Centre for Medium-Range Weather Forecasts (ECMWF) analyses from the TOGA COARE Intensive Observing Period, Fq increases from 10 W m−2 in regions of strong surface winds up to 50 W m−2 in regions of light surface winds. The mean increase in Fq over the COARE Intensive Flux Array is approximately 20 W m−2. Since the CCM3 is presently overestimating Fq by an average of 17 W m−2, the addition of the Jabouille et al. (1996) gustiness correction would apparently increase the bias in Fq calculated by the model. Furthermore, the relationship between the GCM subgrid-scale wind gustiness and convection should be examined further before the local effects of convection on surface turbulent fluxes is incorporated in GCM bulk parameterizations. For this reason, the Jabouille et al. (1996) scheme has not been examined further.
When the Beljaars (1995) scheme is added to the bulk formula applied to the buoys, the mean latent heat flux in the CEPEX domain increases by 1.9 W m−2. When it is added to the CCM3 bulk formula and run offline against instantaneous model fields for a 5-day period, the mean latent heat flux increases by 1.5 W m−2. The 5-day period selected for this test is 23–26 March 1997, and the spatial domain spans 10°N to 10°S, 140°E to 140°W. However, when the Beljaars (1995) scheme is added to the full CCM3 and integrated over the same 5-day period, the mean latent heat flux decreases by approximately 21 W m−2. This change improves the agreement between the modeled and observed latent heat flux shown in Fig. 1. However, it also significantly degrades the agreement between the modeled and observed wind speeds. The mean modeled wind speed decreases by 0.9 m s−1 in the CEPEX domain. Since the winds produced by the standard CCM3 agree well with observed winds (Fig. 3), the modified model underestimates the observed wind speed by up to 1 m s−1. The decrease in latent heat flux is strongly correlated with reductions in the wind field (r = 0.97), and it is essentially decorrelated from changes in the boundary layer humidity and air temperature. A similar deceleration of the surface easterlies is obtained for the entire CEPEX period.
These results indicate that recommendations for changes in bulk flux parameterizations should be evaluated interactively with the full GCM. The sign and magnitude of changes in the bulk fluxes calculated offline may not be applicable when the modifications to the parameterization interact with the rest of the model physics. The results also indicate that at least one class of parameterizations for the effects of free convection improves the agreement between modeled and observed Fq at the expense of the fidelity of the simulated surface winds.
We have carried out a quantitative comparison of observed surface fluxes and the marine boundary layer structure with the NCAR CCM3. This study indicates the value of these observations for evaluation of global climate models. Extension of these types of observations to the eastern tropical Pacific would be of value to the climate modeling community. The results of this study indicate a significant improvement in the simulated surface fluxes from CCM3 over CCM2. The major reason for this is due to the implementation of the new convection scheme of Zhang and McFarlane (1995). CCM2 predicted excessively large latent heat fluxes in the tropical Pacific. The convective heating in CCM2 drove a vigorous tropical circulation that amplified the surface flux bias. The implementation of the new convection scheme has considerably reduced these biases in CCM3. The simulation of the boundary layer height is now in much better agreement with observations.
However, certain features of the simulated marine boundary layer have degraded in CCM3. A number of these problems are related to the particular surface exchange formulation used in CCM3. A future test will be to implement the Zhang and McPhaden (1995) surface exchange formulation into the CCM. In the new model, the boundary layer has cooled by 0.5 K, such that CCM3 now underestimates the thermal structure compared to observations. The moisture structure within the marine boundary layer still needs improvement. Although the vapor mixing ratio in the lowest model level is in good agreement with the observations, above this level the model mixing ratio is too high. This feature was in CCM2 and is most likely linked to entrainment mixing across the top of the marine boundary layer. It is hoped that continued use of observational data from CEPEX and TOGA COARE will help improve the vertical structure of temperature and moisture within the marine boundary layer simulated by the CCM3.
This research was supported by Grants NSF ATM89-20119 (WDC), NSF ATM94- 05024 (WDC, JK, and JW), and DOE DEFG 0391ER61198 (GJZ). CEPEX was funded by NSF and DOE. The TAO buoy data were obtained from the Project Office of the Pacific Marine Environmental Laboratory (PMEL) of NOAA, courtesy of M. J. McPhaden. We would like to thank the entire crew of the R/V Vickers, as well as U. Schmid and W. Biselli, for launching the radiosondes and maintaining the surface observations from the ship. The authors benefitted from discussions with K. Gage, R. Grossman, L. Hartten, P. Lemone, D. Parsons, and D. Rogers regarding measurements of winds in the lower boundary layer from TOGA COARE. The computer time for some of the integrations of the NCAR CCM3 was provided by the San Diego Supercomputer Center. The paper was improved considerably by reviews from two anonymous referees.
+ Additional affiliation: Center for Atmospheric Science, Scripps Institution of Oceanography, University of California, San Diego, La Jolla, California.
# Additional affiliation: National Center for Atmospheric Research, Boulder, Colorado.
Corresponding author address: Dr. W. D. Collins, NCAR/CGD, P.O. Box 3000, Boulder, CO 80307-3000.
* Center for Clouds, Chemistry, and Climate Report 164.