Comparing MM5 radiative fluxes with observations gathered during the 1995 and 1999 Nashville southern oxidants studies



[1] The impact of radiative transfer processes on MM5 air-quality forecasts during the Nashville Southern Oxidants Experiments is addressed. We find that an incorrect specification of aerosol scattering, aerosol absorption, and ozone absorption in the model can lead to positive errors in the instantaneous total solar irradiance at the surface of 100 W m−2 using the Dudhia solar parameterization. This increased solar irradiance upsets the surface energy balance in the model causing errors in the surface heat fluxes and mixing depth estimation. We also show that a systematic 80 W m−2 increase in the forecast downwelling long-wave radiation can inhibit the formation of ground-based nocturnal inversions in the model when the Dudhia long-wave radiation routine is used. Thus we find that accurate specification of radiative processes, including some driven by local pollution, can be important for mesoscale meteorological forecasting. A method for including ozone absorption in MM5 is introduced, and optimum choices for running the model in aerosol rich geographic regions are discussed.

1. Introduction

[2] Radiation is the primary physical process that drives the atmosphere-ocean circulation. The climate modeling community has gone to great lengths in verifying the accuracy of their model radiation parameterizations, given the fundamental role radiative transfer plays in their modeling efforts [Ellingson and Fouquart, 1991]. What is perhaps not quantitatively so well understood is the key role radiative forcing can play in the evolution of the atmospheric boundary layer (ABL) on timescales shorter than 24 hours. In this paper, we examine the impact that radiative transfer processes can have on ABL forecasts used in air quality research.

[3] During the 1995 and 1999 Southern Oxidants Studies (SOS), the NOAA Environmental Technology Laboratory (ETL) deployed an array of in situ and remote sensing systems that obtained detailed observations of the wind, thermodynamic, and turbulent flux distributions within the ABL surrounding the Nashville, Tennessee metropolitan area. We compared the observed fluxes of heat and moisture and diagnosed mixing depths with forecast values derived using the Blackadar [1979] ABL parameterization embedded in the Pennsylvania State University (PSU) -National Center for Atmospheric Research (NCAR) Mesoscale Model (MM5) [Grell et al., 1994].

[4] Zhang and Anthes [1982] and Oncley and Dudhia [1995] have evaluated the Blackadar ABL scheme used to predict the surface fluxes in MM5. Zhang and Anthes noted the large effect latent heating had on ABL development. Reducing latent heating makes more energy available for sensible heating and increased ABL growth. Oncley and Dudhia found that the MM5 surface flux parameterization was most sensitive to the soil moisture availability parameter (M). Given a proper choice of M, they found reasonable agreement with the model-predicted fluxes and fluxes measured by both ground-based and airborne observing systems.

[5] The preliminary conclusions reached after running MM5 and comparing it with the 1995 field data were that MM5 systematically over-predicted the ABL depth, and that the turbulent latent heat fluxes were much higher than the observed fluxes. However, when we adjusted the soil moisture in the manner suggested by Oncley and Dudhia [1995] we found that the discrepancy between the observed and predicted ABL depth increased dramatically. Initially we suspected that using a more sophisticated land surface model in MM5 such as the Pan and Mahrt scheme [Pan and Mahrt, 1987; Chen and Dudhia, 2001] would produce better agreement between the observed latent heat fluxes and MM5. Subsequent tests suggested that the model overestimated the net radiation used in the surface-energy budget calculation when the Dudhia radiative transfer parameterization was employed [Dudhia, 1989]. Given the large net radiative input to the flux routines, it was not surprising that the mixed layer depths grew unrealistically when we reduced the amount of energy available for evaporation.

[6] We also noted that boundary layer profilers observed a low-level nocturnal jet in the lowest few kilometers in the atmosphere over the Nashville region on many nights during the 1995 study. Spectral analysis performed by Gupta et al. [1997] on the SOS 99 wind profiler data sets revealed a pronounced peak at the inertial period, suggesting that the classical mechanism of nocturnal low-level jet formation proposed by Blackadar [1957] was responsible for the observed jets. When we compared MM5 forecast soundings for Nashville with the National Weather Service (NWS) observations, we noted that surface based nocturnal inversions were not developing in the model because the surface was too warm at night. The lack of low-level stability in the model prevented the wind field decoupling needed for low-level jet formation.

[7] Questions about the SOS95 radiative flux measurements made it difficult to conclude that the over prediction of mixing depth and the lack of nocturnal surface cooling was caused by errors in the MM5 radiation parameterization. ETL returned to the Nashville region in 1999 with an extensive suite of surface radiation, soil moisture, and soil heat flux instrumentation. We obtained a detailed data set that could be used for a deeper evaluation of the MM5 ABL and radiation routines.

[8] In addition to the enhanced instrumentation we also ran the latest version of the MM5 modeling system, MM5V3.3. When the model predicted nocturnal boundary layer thermal structure and radiative fluxes were compared with the observations we found the same biases discovered in the 1995 studies.

[9] We will show that during stagnation events over the southern U.S., more accurate predictions of the net radiative flux can be made by MM5 using the Rapid Radiative Transfer Model (RRTM) [Mlawer et al., 1997] long-wave radiation parameterization and modifications to the current short-wave [Dudhia, 1989] parameterization. We will also illustrate the key role of the aerosol representation in radiative transfer physics.

[10] Section 2 summarizes the observational networks used during the 1995 and 1999 studies. The MM5 model configuration is discussed in section 3. Section 4 highlights the meteorological conditions present during the 1995 and 1999 case studies. The comparisons between the model runs and the observations are examined in section 5. The MM5 short-wave and long-wave radiative transfer parameterization tests are shown in sections 6 and 7. Our final results and conclusions are presented in section 8.

2. Observational Networks

[11] Figure 1 shows the locations of the boundary layer profiler and surface flux measuring sites during SOS95 and SOS99. The surface flux sites in 1995 were located approximately 15 and 20 km northeast of downtown Nashville at Dupont (DUP) and New Hendersonville (HEN). The surface heat fluxes were estimated using the eddy correlation method and data gathered by an Atmospheric Technology Incorporated (ATI) sonic anemometer/thermometer and an Ophir IR-2000 fast infrared hygrometer. Both the hygrometer and anemometer were mounted atop a 10-m tower at Dupont and 18 m above ground at New Hendersonville. The net solar flux was measured at 2 m using a Radiation Energy Balance Systems (REBS) Q7.1 radiometer. A LICOR LI-200SZ solar pyranometer measured the incoming solar radiation at 2 m. Mean measurements of wind speed and direction, station pressure, temperature, and relative humidity were made at the Dupont site and the other ETL profiler sites. Mixed boundary layer depths were derived hourly from the profiler reflectivity data [White, 1993; Angevine et al., 1994].

Figure 1.

Locations of the observational sites during the SOS 1995 (crosses) and 1999 (triangles) field campaigns.

[12] The sonic anemometer and hygrometer were separated by approximately 1.0 m on the tower. This separation can cause a loss of covariance when significant energy is present on scales smaller than 1.0 m. We estimate that the magnitude of the latent heat flux error is between 10 to 15% based on the analysis of Kristensen et al. [1997].

[13] During the 1999 field study, the surface radiative, latent, and sensible heat flux measurements were co-located with the surface chemistry measurements 10 km northeast of downtown Nashville at the Cornelia Fort Airpark (CFA) from 15 June to 15 July (Figure 1). The instrumentation was supplemented with upward and downward looking Eppley Precision Spectral Pyranometers (PSPs) and Eppley Precision Infrared Radiometers (PIRs) or pyrgeometers, ground temperature sensors buried at 6.0, 8.0, and 10.0 cm below the surface, a REBS soil heat flux plate buried at 2.0 cm and a Campbell Scientific CSI 615 soil moisture time domain reflectometer (TDR).

[14] The upward and downward pyrgeometers and the upward looking pyranometer were mounted 2.0 m above ground. The downward looking pyranometer was mounted 9 m above ground level on a small tower at the end of a 1.5-m boom. Black body pyrgeometer calibrations were made before and after the experiment at the NOAA Climate and Diagnostics Laboratory (NOAA/CMDL) calibration facility. The calibrations were applied using the Albrecht-Cox [Albrecht and Cox, 1977; Philipona et al., 1995; Fairall et al., 1998] method. The pyrgeometer IR fluxes are considered accurate to within ±5.0 W m−2. The pyranometers were calibrated at the NOAA/CMDL Solar Calibration facility. The solar calibration facility is a WMO World Region IV Center. Given the cosine response of the individual pyranometers used during SOS99, the broadband solar fluxes are accurate within ±5.0 W m−2.

3. MM5 Model Configuration

[15] We used the non-hydrostatic PSU/NCAR Mesoscale Model MM5 version 2.9 in our 1995 case studies. MM5 versions 2.9 and 3.3 were tested on the 1999 cases. Boundary conditions were specified using the NOAA National Center for Environmental Prediction (NCEP) global-spectral model. Initial condition, were derived using objective analyses. The Cressman-based method used all the available surface and upper-air sounding data to update the initial first guess field also obtained from NCEP global spectral model. Twenty-seven kilometer mother grids and 9-km inner nests were used in the 1995 and 1999 studies. The grids were centered over Nashville, TN. Land use specification came from the NCAR 19 km global land use data tapes during the 1995 experiments and the U.S. Geological Survey (USGS) 24-category land use data for the 1999 studies. We configured the model with 27 vertical levels. The first nine predictive levels were located in the lowest 1.6 km of the model atmosphere. The Grell cumulus, moist shallow convection, and the simple ice explicit moisture parameterizations [Dudhia, 1989] were used on both the 27.0 and 9.0 km grids.

[16] Both long and short-wave radiative effects were simulated [Dudhia, 1989]. During the 1995 case studies, only the Dudhia cloud radiative scheme was used. For the 1999 case studies, we also tested the so-called Simple radiative transfer scheme [Grell et al., 1994] along with the NCAR Community Climate Model (CCM-2) and RRTM radiative parameterizations. We used the Blackadar boundary layer parameterization [Blackadar, 1979; Zhang and Anthes, 1982]. We also tested higher-order ABL schemes including the Burk-Thompson (B-T) [Burk and Thompson, 1989] and Medium Range Forecast (MRF) [Hong and Pan, 1996]. Although the mixing depths from the Blackadar routine were systematically over-predicted, the Blackadar scheme still gave better predictions of mixed layer depth than either the B-T or MRF parameterizations over central Tennessee. The ground temperature was predicted using the surface energy budget and the multilevel soil model. In all our test cases, we used static initial conditions in forecast mode.

4. Meteorological Conditions

[17] McNider et al. [1998] summarizes the meteorological conditions present over Tennessee and the surrounding states during the 84-h period beginning at 1200 UTC, 10 July 1995. The synoptic scale meteorological flow was dominated by a large, persistent subtropical high-pressure system centered over the southeastern US during the time period. Weak synoptic scale pressure gradients and, hence, weak northeasterly winds were found in the lower 2.0 km of the atmosphere. Precipitation occurred before the stagnation period, but no rain was observed during 11–13 July in the region surrounding and including Nashville. Skies were clear for the entire period.

[18] During the 1999 SOS field season, the conditions in late June and early July were much wetter and cloudier. The only clear sky day during the entire experiment occurred on 18 June. During the 3–8 July tropospheric ozone event, thunderstorm activity was observed over Nashville on 3 July.

5. Comparing the Model Results and the Observations

[19] We ran a total of four cases for SOS 95 and five cases for the SOS 99 study. The first SOS 95 case chosen for study was the three-day stagnation event of 11–13 July 1995. The mixing depths forecast by MM5 for each day are at least 200 m higher than the profiler observations (Figure 2). The comparison between the modeled and observed sensible heat fluxes (Figure 3) indicates that the model overestimates the flux by about 100 W m−2 at the peak of the first diurnal heating cycle and then systematically overestimates the fluxes through the last 40 h of the simulation. The latent heat fluxes are 280 W m−2 higher than the observations at the peak of each heating cycle (Figure 4).

Figure 2.

Mixing depth predicted by MM5 (solid line) and wind profiler observations (dashed line) at New Hendersonville, Tennessee as a function of time for 11–14 June 1995.

Figure 3.

Sensible heat flux predicted by MM5 (solid line) and measured (dashed line) at New Hendersonville, Tennessee as a function of time for 11–14 June 1995.

Figure 4.

Latent heat flux predicted by MM5 (solid line) and measured (dashed line) at New Hendersonville, Tennessee as a function of time for 11–14 June 1995.

[20] Oncley and Dudhia [1995] found a similar difference when they compared the MM5 latent heat flux estimated using the Blackadar parameterization with tower and aircraft measured fluxes in the winter of 1992 over northeastern Colorado and in the spring of 1993 over northeastern Kansas using the default soil moisture availability (M). When the Blackadar “Slab”, Land Surface Model (LSM) is used, the soil moisture availability at a grid point is a fixed fraction based on the land use for the grid point. The latent heat flux is parameterized in the Blackadar ABL scheme as

equation image

where M is the moisture availability parameter, ρa is the air density at the lowest model layer, I−1 is the resistance to moisture transfer, qvs(Tg) is the mixing ratio at the skin surface, and qva(Ta) is the mixing ratio at the lowest model layer. Oncley and Dudhia showed that better agreement could be found with the observed latent heat flux when the M used in the model was reduced. When we reduced M, the difference between the modeled and observed latent heat fluxes became much smaller. Unfortunately, the energy that was previously used in evaporation now increased the sensible heat flux, giving us an even larger discrepancy between the modeled and observed sensible heat fluxes.

[21] The surface energy balance in the Blackadar parameterization is estimated using the “Force-restore” method [Zhang and Anthes, 1982]. The budget is written as

equation image

Cg is the thermal capacity of the ground slab per unit area, equation image is the skin temperature tendency, Rn is the net radiative flux, Hm is the heat flow into the substrate, Hs is the sensible heat flux, and Lv is the latent heat of vaporization. Scale analysis of equation (1) tells us that under typical soil conditions, the heat flow into the substrate is one order of magnitude smaller than the other three terms on the right hand side. It became clear that arbitrarily changing M only changed the partitioning of the net radiation between the latent and sensible heating. This led us to compare the net radiative flux into the slab with the observations.

[22] The comparisons between the MM5 predicted net radiative fluxes and the 11–14 July 1995, observations (Figure 5) suggested that the surface was not cooling enough during the night, and that it was too warm during the day. The solar radiative flux predicted by MM5 for the period was also biased high by over 100 W m-2 when compared with the observations (Figure 6). Further examination of the NWS Nashville morning sounding taken for 1200 UTC 11 July and the MM5 predicted sounding (Figure 7) showed that MM5 failed to produce a ground-based nocturnal inversion. These results suggested that both the downwelling shortwave and longwave radiative fluxes were overestimated in the MM5 forecast. Note that during the stagnation event, no clouds were observed over the observational site, and no clouds formed in the simulation.

Figure 5.

Net radiative flux predicted by MM5 (solid line) and measured (asterisks) at New Hendersonville, Tennessee as a function of time for 11–14 June 1995.

Figure 6.

Solar radiative flux predicted by MM5 (solid line) and measured (asterisks) solar radiative flux at New Hendersonville, Tennessee as a function of time for 11–14 June 1995.

Figure 7.

Temperature as a function of height predicted by MM5 using the Dudhia longwave parameterization (solid) and the Nashville rawinsonde observation (dashed) for 1200 UTC 11 July 1995.

[23] Although similar results were obtained for the 18–21 July and 7–9 July 1995, case studies, shortcomings in the quality of the radiative flux observations made it difficult to conclude that the radiative parameterizations in MM5 were responsible for the discrepancy between the model and the observations. The LICOR LI-200SZ solar pyranometers utilize a silicon photodiode that does not respond uniformly over the full solar spectral range. The spectral response is very low at 0.4 μm, with a cutoff near 1.1 μm. Calibration constants are determined for each sensor using an Eppley PSP which has a flat spectral response between 0.28 μm and 2.8 μm. On days when large amounts of water vapor absorption are occurring in the near-infrared region of the spectrum, a LI-2000SZ solar pyranometer can produce biased output because it was calibrated to agree with an Eppley PSP under calibration atmospheric conditions. Later side by side comparisons done during SOS 99 with the LICOR LI-200SZs and the Eppley Precision Spectral Pyranometers suggest that the LICOR solar pyranometer can overestimate solar irradiance by as much as 50 W m−2 under hazy sky conditions. This suggests that the solar flux bias found during the 1995 model validations could have been larger than 100 W m−2.

[24] Given the limitations of the LICOR solar pyranometers, we returned to Nashville during the summer of 1999 and deployed two Eppley Precision Spectral Pyranometers (PSPs) and two Eppley Precision Infrared Radiometers (PIRs) at the Cornelia Fort Airpark surface chemistry observational site, greatly reducing the possible instrumental errors. One of the major tropospheric ozone events occurred over Nashville between 3–8 July 1999. Comparisons between the MM5 predicted solar fluxes using the Dudhia atmospheric radiation scheme and the Eppley PSP measured fluxes for the 72-hr period beginning at 0000 UTC 4 July are shown in Figure 8. The presence of clouds in both the MM5 simulation and the observations complicates the interpretation of the comparison, but it is still clear that the model produced more solar radiation than was observed. The net radiation for the same period (Figure 9) also shows the same bias as the SOS 95 comparisons, including the positive bias during the night. We found that the model forecasts for 11–13 July and 18–19 June 1999, also had the same biases. Given the improved radiation measurements obtained during SOS 99, it became apparent that a problem or problems in the Dudhia MM5 radiation parameterization were responsible for the disagreement between the simulated net radiation and the observations.

Figure 8.

Solar radiative flux predicted by MM5 (solid line) and measured (asterisks) solar radiative flux at Cornelia Fort Airpark, Tennessee as a function of time for 4–6 July 1999.

Figure 9.

Net radiative flux predicted by MM5 (solid line) and measured (asterisks) net radiative flux at Cornelia Fort Airpark, Tennessee as a function of time for 4–6 July 1999.

6. Testing the Solar Radiation Parameterization

[25] The solar irradiance parameterization utilized when the Dudhia and RRTM atmospheric radiation packages are run in MM5 is based on the Lacis and Hansen [1974] parameterization. Solar absorption varies with cloud amount and composition, humidity, and the zenith angle of the sun. The implementation of the scheme in the current version of MM5 neglects stratospheric ozone absorption. Ozone absorption, Rayleigh scattering, aerosol absorption, and upward aerosol scattering are accounted for using a single model parameter XSCA. Initially, agreement between the model and observations was forced using this parameter. In some sense, XSCA can be regarded as a “catch-all” for any physical attenuation process not explicitly described in the parameterization.

[26] The vertical computational domain of MM5 rarely extends beyond 10 kPa in typical applications of the modeling system. It is well known that absorption of solar radiation by ozone occurs at and above 10 kPa, and that under typical midlatitude atmospheric conditions, the absorption is between 20 and 30 W m−2. Therefore, one is tempted to conclude that ignoring ozone absorption in the solar stream will have a negligible effect on the model solution for the majority of fluid dynamics problems studied using the MM5 modeling system.

[27] However, in our studies it appears that additional solar absorption is also taking place in the atmosphere over Nashville, TN that is not being accounted for in MM5. The additional energy input into the surface leads to high ABL heat fluxes. Particulate scattering, particulate absorption, gas scattering, and gas absorption are all important processes that reduce the amount of solar radiation transmitted to the earth's surface. Any physical process that reduces the solar radiation received at the surface by an amount larger than the observational uncertainty should be included in the MM5 solar flux parameterization under these conditions.

[28] As a test, we initially modeled the 18 June 1999, clear sky case using the Dudhia atmospheric radiation scheme. The comparison between the model solar fluxes and the observations shows again that the model overestimates the incoming solar irradiance (Figure 10). In this case, the difference is 80 W m−2 at solar noon.

Figure 10.

Solar radiative flux predicted by MM5 (solid line) and measured (asterisks) solar radiative flux at Cornelia Fort Airpark, Tennessee as a function of time for 18 June 1999.

[29] Next, we ran the model over Boulder, CO with and without stratospheric ozone absorption and compared the model solar radiative outputs with detailed measurements taken by NOAA/CMDL and a line by line (LBL) radiative transfer model. Our goal in choosing Boulder was to seek a less polluted atmosphere and to compare to the suite of highly accurate permanent instruments at the Boulder radiation facility. 12 July 1999, was chosen for the test case. Ozone absorption in the Dudhia short-wave scheme was included using the Lacis and Hansen [1974] parameterization. The details are discussed in the Appendix. No clouds were observed over Boulder on 12 July. NOAA/CMDL Dobson spectrophotometer measurements over Boulder found a column ozone value of 306 DU. A multifilter rotating shadowband radiometer (MFRSR) measured an aerosol optical depth at 500 nm of 0.125. The spectrophotometer and shadowband radiometer measurements were used in the LBL model calculations. The ozone absorption functions in the modified Dudhia short-wave parameterization were specified using the spectrophotometer measurements.

[30] We found excellent agreement between the ozone compensated MM5 estimated solar irradiance and the NOAA/CMDL observations. There is a slight lag between the observations and the model solar peak output caused by the timing at which the radiation routines are called in the model cycle (Figure 11). As expected, ozone accounts for 28.3 W m−2 additional solar absorption. We also checked the initialization of the model water vapor column with the nearest radiosonde observations and found good agreement between the model column water vapor and the observations.

Figure 11.

Solar radiative flux predicted by MM5 (solid line) and measured (crosses) solar radiative flux at Boulder, CO as a function of time for 12 July 1999.

[31] We also compared the NOAA/CMDL Boulder solar irradiance observations with a (LBL) model [Portmann et al., 2001]. The scattering profile for the LBL model was chosen based on the NOAA/CMDL MFRSR observations. The observations and the model total solar flux agree within 20 W m−2 (Figure 12a). These results are similar to other studies including Halthore et al. [1998] and Mlawer et al. [2000]. The LBL results also reveal that solar absorption by ozone maximized at solar noon with a value of 27.4 W m−2 (Figure 12b). The LBL ozone results suggest that the simple ozone parameterization we included in MM5 is reasonably accurate.

Figure 12.

NOAA/CMDL observed solar flux and line-by-line model results as a function of time for 12 July 1999.

[32] The major radiative difference between the Colorado and Tennessee air masses is the aerosol optical depth. In contrast to Boulder, the atmosphere over central Tennessee during stagnation events has higher aerosol loading. NOAA Aeronomy Laboratory personnel estimated an aerosol optical depth of 0.2 below 5 km at 550 nm over Nashville on 4 July 1999, using data gathered by the NOAA WP3 aircraft.

[33] The performance of the Dudhia MM5 solar radiation routine in simulating solar irradiance for the Colorado and Tennessee locations implies that the largest errors in the forecast solar irradiance are found when aerosol scattering and absorption in the atmosphere exceeds the climatological value used in MM5. Over Boulder where the aerosol optical depth is 0.1 the stratospheric ozone adjusted solar fluxes produced by the Dudhia parameterization show excellent agreement with the observations. When the aerosol optical depth doubles, as it does over Nashville the same scheme overestimates the solar irradiance by ∼100 W m−2. Of that excess only 30 W m−2 can be attributed to stratospheric ozone absorption based on the Nashville spectrophothometer measurements. Assuming observational errors of 10.0 W m−2, this leaves a 60 W m−2 difference that we believe must be due to insufficient aerosol scattering and absorption. Research carried out during the Indian Ocean Experiment (INDOEX) [Satheesh et al., 1999] shows that the global solar fluxes decrease by 50 to 80 W m−2 as the observed aerosol optical depth varies from 0.1 to 0.4.

[34] The LBL results and the 12 July 1995 MM5 run with ozone absorption over Boulder suggest that the default Rayleigh scattering, aerosol directional scattering, and aerosol absorption used in the Dudhia solar parameterization are quite reasonable so long as the aerosol optical depths are small. Running the model in geographic regions where the aerosol optical depths are large requires using observations or using detailed aerosol physics to get the correct solar irradiance at the surface.

[35] Westphal and Toon [1991] using the MM4 modeling system [Anthes et al., 1987], along with the NASA/Ames aerosol model [Toon et al., 1988], simulated an extreme case of aerosol loading in the atmosphere caused by a forest fire over western Canada. They show that radiative feedbacks occurred well downstream of the fire region over New England even after significant scavenging had taken place.

[36] In the absence of a sophisticated aerosol physics package we suggest tuning the XSCA parameter in the Dudhia parameter to force agreement with the observed clear-sky solar fluxes or developing a relationship between XSCA and the observed aerosol optical depth. Great care must be taken when adopting either approach. If the XSCA parameter is empirically adjusted to compensate for a stagnant, aerosol loaded, air mass and the large scale synoptic scale weather pattern brings in a relatively cleaner air mass the modeled solar fluxes will still be in error.

7. Testing the Long-Wave Radiation Parameterizations

[37] As mentioned earlier, we tested the Simple, RRTM, Dudhia, and CCM-2 long-wave radiative transfer schemes in MM5 after finding that nocturnal inversions were not being produced for the case studies analyzed during the 1995 and 1999 SOS experiments. We measured both the downwelling and upwelling IR as described in section 2 for the 1999 SOS cases and compared those measurements with the MM5 predicted IR components.

[38] For the 3–8 July case study, all of the radiation parameterizations except the Dudhia long-wave scheme produced a nocturnal inversion for each diurnal cycle. The NWS Nashville rawinsonde observation launched at 1100 for 1200 UTC 3 July, is shown in Figure 13. In contrast, the sounding predicted by MM5 using the Dudhia atmospheric radiation package has a weak stable layer located above a 50 kPa surface-based dry adiabatic layer (Figure 13). We examined the MM5 predicted soundings at hourly intervals through the period from 0000 UTC 3 July to 1200 UTC 3 July making sure that a nocturnal surface based inversion did not form and then quickly mix out to form the sounding shown in Figure 13. There was no indication that an inversion formed, and in fact, the dry adiabatic layer persisted through the night. The MM5 predicted low-level winds were from the southwest while the Nashville rawinsonde winds showed south southeasterly flow in the near surface layer (not shown). This finding suggests that the low-level wind field was coupled to the mid-tropospheric wind field in the model through the entire night, preventing the development of the low-level southerly jet.

Figure 13.

Temperature as a function of height predicted by MM5 using the Dudhia (solid) and RRTM (short dashed) longwave parameterizations and the Nashville rawinsonde observation (long dashed) for 1200 UTC 3 July 1999.

[39] Our first impression was that either turbulent mixing or numerical diffusion might be too strong in the model, and this could inhibit the development of a low level inversion in the run. Given that the other three radiation schemes did develop an inversion or at least a stable region, it seems unlikely that either numerical diffusion or the turbulence parameterization is responsible for the results we show. By comparing the measured downwelling IR fluxes with those predicted by MM5 using the Dudhia IR routine, we found that the model consistently over predicted the downwelling flux by approximately 80 W m−2 (Figure 14). The comparison between the downwelling fluxes produced by the RRTM method and the observations indicates better agreement in a mean sense but the presence of clouds in both the simulation and the observations makes the interpretation of the details difficult (Figure 15). However, the RRTM scheme coupled with the Blackadar ABL parameterization did produce a low-level stable layer at the 1200 UTC, 3 July 1999, verification time (Figure 13). The net radiation produced using the Dudhia solar and RRTM IR schemes again shows better overall agreement during the evening hours and before cloud formation on 3 July (Figure 16). The comparison also shows the 100 W m−2 solar bias we found earlier near solar noon.

Figure 14.

Downwelling IR radiative flux predicted by MM5 using the Dudhia long-wave parameterization (solid line) and measured (asterisks) at Cornelia Fort Airpark, Tennessee as a function of time for 3–4 July 1999.

Figure 15.

Downwelling IR radiative flux predicted by MM5 using the RRTM long-wave parameterization (solid line) and measured (asterisks) at Cornelia Fort Airpark, Tennessee as a function of time for 3–4 July 1995.

Figure 16.

Net radiative flux predicted by MM5 using the RRTM long-wave parameterization (solid line) and measured (asterisks) at Cornelia Fort Airpark, Tennessee as a function of time for 3–4 July 1999. (asterisks).

[40] We made one additional run using the Dudhia long-wave parameterization for the 3–5 July 1999 time period. For this run, we subtracted the 80 W m−2 bias from the downwelling IR stream at the surface at each radiation time step. A surface based nocturnal inversion formed each night during this run suggesting that an error of 80 W m−2 in the downwelling IR stream at the surface can have a significant impact on the ABL thermal structure and can inhibit the formation of nocturnal inversions. These findings are similar to those of Guichard et al. [2002]. They found a similar result when comparing the Dudhia downwelling IR estimates with the Southern Great Plains Cloud and Radiation Testbed (CART) observations.

[41] It should be pointed out that both the Dudhia longwave and shortwave radiative transfer schemes were developed to simulate cloud-radiation interactions in monsoon flow. [Dudhia, 1989]. Extensive testing of the radiative flux schemes in mesoscale models has not been carried out because high-quality observational radiative fluxes are only now becoming available.

8. Summary and Conclusions

[42] Our testing of the Dudhia radiative transfer modules in MM5 shows that during summer stagnation events over the southeastern U.S., significant biases can exist in the predicted solar and infrared radiative fluxes. For the cases we examined, including ozone absorption in the Dudhia solar irradiance parameterization, and using the RRTM long wave scheme produced the best agreement between the model and the observations.

[43] The ultimate goal of the meteorological modeling component of the Southern Oxidant Studies is to produce forecast wind fields and mixing depths to drive air chemistry models. The MM4/MM5 modeling systems were initially developed to address mesoscale atmospheric fluid dynamics problems where the primary source of energy in the atmosphere is at the synoptic scale. During stagnation events, synoptic scale and microscale forcing mechanisms interact strongly and produce the observed wind and thermal fields. Thus, radiative forcing on both large and small scales, land-use specification, and turbulent diffusion become crucial components of an accurate MM5 forecast. In this paper, we have explored the radiative transfer parameterization issue in some detail. Our initial focus has been radiation rather than the land-surface model because the primary input to the land-surface scheme is net radiation. Our study shows that careful tuning of the land-surface scheme can mask problems with the radiative transfer modules that only become apparent when extreme events such as stagnation episodes are examined.

[44] These findings must be viewed with caution. No observational data set is without error. Obtaining high quality radiative flux measurements requires great care and attention to detail. In addition, there are many possible choices that one can make when configuring the MM5 modeling system. We have focused on the MM5 configuration that was initially used by the Environmental Protection Agency (EPA) to drive the Models 3 Community Multiscale Air Quality (CMAQ) system. Different combinations of model physics and different choices for the MM5 initial and boundary conditions could produce results that are very different from what we have shown in this work.

[45] However, we have shown that an incomplete specification of the net radiative input in the model can have a strong impact on the boundary temperature forecasts. Once the radiative transfer solution begins diverging, feedbacks between the turbulent fluxes and clouds interact and degrade the model forecast. Aerosol scattering and absorption can produce a mid-day 60–70 W m−2 solar irradiance reduction in MM5 during stagnation events and are at least partially responsible for the net radiative biases we have shown. Our work demonstrates that a crucial feedback loop exists between air chemistry and the meteorological forecast. This illustrates that there is a need for detailed aerosol observations within the ABL and a better treatment of aerosol scattering and absorption in the solar parameterization.

Appendix A

[46] Ozone absorption can be included in the Dudhia MM5 solar parameterization using the simple absorption functions for the Huggins, Hartley, and Chappius bands proposed by Lacis and Hansen [1974]. For simplicity in the ozone calculations, we assume that the atmosphere is a pure absorbing layer above 10 kPa and a pure scattering layer below 10 kPa. The total column ozone is placed in the layer above 10 kPa, and the absorption for each band is then computed as a function of zenith angle as soon as the zenith angle is calculated in subroutine SWRAD.F. Once the total ozone absorption is calculated for the column, that amount is subtracted from the MM5 solar constant. The remainder of the solar flux calculations are accomplished using this effective meteorological solar constant. The method requires minimal changes to subroutine SWRAD.F and the addition of two short subroutines. This approach avoids explicitly solving the MM5 governing equations above 10 kPa while still accounting for stratospheric ozone absorption. The method could be implemented using a climatological value for the total ozone amount as a function of latitude. In our case, we assumed that the column ozone amount was uniform in the stratosphere over the model domain based on the daily NOAA/CMDL Dobson observations. The column ozone amount for each the grid point in question could also be specified using the satellite borne Total Ozone Mapping Spectrometer (TOMS).


[47] The authors would like to thank Scott Abbott, Dr. Clark King, Daniel Gottas, Jesse Leach, Brian Templeman, and Catherine Russell (NOAA/ETL) for their efforts deploying and maintaining the instrumentation and data output during the 1995 and 1999 SOS field program. Samuel Oltmans (NOAA/CMDL) supplied the total ozone measurements taken for both Nashville, TN and Boulder, CO. Dr. Ola Persson, Dr. Jimy Dudhia, and the anonymous reviewers provided us with thoughtful, careful reviews of the manuscript. This work was supported by the NOAA Health of the Atmosphere Program. The support of Drs. Daniel Albritton, James Meagher and William D. Neff is gratefully acknowledged.