Influence of the prescribed solar spectrum on calculations of atmospheric temperature



[1] Significant differences in heating rates are found when two solar irradiance spectra are used in a line-by-line radiative transfer code. Compared with a spectrum of recent satellite data an older theoretical spectrum gives 20–40% more heating in the ozone Hartley band, important in the upper stratosphere. The spectra are implemented in a broad-band radiation code to which some improvements are also made to the ozone absorption parameterization. A widely-used spectrum of ground-based data from 1960s gives somewhat lower heating rates. The effects of the changes in the spectrum, and the broad-band scheme, on the temperatures simulated by a middle atmosphere GCM are investigated. The model has previously shown a warm bias, compared with climatology, around the stratopause but this is significantly reduced when the former spectrum is substituted for the latter, and the new ozone parameterization incorporated. The change in spectrum accounts for two-thirds of the improvement.

1. Introduction

[2] Solar radiation is the primary driving source of energy for atmospheric and oceanic circulations. An accurate solar irradiance spectrum, and its variability with time, is necessary to study the Earth's atmospheric thermal structure, composition, dynamics and climate [Thuillier et al., 2004; Lean et al., 2005]. Furthermore, an accurate assessment of the relationship between solar activity and climate will depend precisely on the input solar irradiance.

[3] The magnitude of the UV flux is much smaller than in the visible and near infrared regions but its absorption by ozone and molecular oxygen causes significant heating in the middle atmosphere. Absorption of visible radiation in the lower stratosphere, and of near infrared in the troposphere, by O3, H2O, CO2 and other minor gases result in smaller heating rates in these regions. At the top of the atmosphere the changes in incoming radiation in the UV (0.12–0.4 μm: 0.425 Wm−2) between periods of high and low solar activity over the 11-year cycle produce an impact on middle atmosphere temperature and composition which influences the radiative flux at the tropopause [Haigh, 1994].

[4] The use of different solar irradiance spectra in radiative transfer models can result in large changes in the calculated heating rate (J. Thelen, personal communication, 2006). The solar spectra of Kurucz [1995] have been extensively used in line-by-line radiative transfer models (e.g., LBLRTM [Clough et al., 2005]), and also in narrow-band models (e.g., Modtran), because of their high spectral resolution. General Circulation Models, however, cannot afford to include line-by-line, or narrow-band, radiation codes because of their computational requirements and thus employ spectrally-integrated (broad-band) parameterizations. In the past many of these were based on the ground-based spectral data of Labs and Neckel [1970]. More recently the spectrum of Kurucz [1995] has been employed, for example in the UK Met Office Unified Model HadGEM1 [Davies et al., 2005] and the Australian GCM (Z. Sun, personal communication, 2008). In this note, we investigate the influence of the choice of solar spectrum on spectral heating rates and assess the impact on temperatures simulated in a vertically-extended version of the Met Office GCM.

2. Data and Models

2.1. Spectral Data

[5] We compare the spectra of Kurucz [1995], Lean [2000] and (for the broad-band calculations only) Labs and Neckel [1970].

[6] The Kurucz spectrum is generated from a theoretical spectral line model added to a solar continuum model. Here we use a version with 1 cm−1 spectral resolution. The Lean [2000] spectra, derived from satellite observations since the 1991 together with an empirical model, are based on SOLSTICE (Solar Stellar Irradiance Comparison Experiment) on UARS (the Upper Atmosphere Research Satellite) spectra to 400 nm, ATLAS (ATmospheric Laboratory for Applications and Science missions) spectra to 800 nm and the Kurucz spectrum beyond. ATLAS spectra [Thuillier et al., 2004] are based on the satellite measurements of SOLSPEC (SOLar SPECtrum instrument) and SOSP (SOlar SPectrum instruments). We use the Lean [2000] data between 0.12 and 10 μm, at a resolution of 1 nm, averaged over the solar cycle between 1975 and 1986 inclusive. The Labs and Neckel [1970] spectrum was compiled from many datasets obtained during the 1960s and 1950s including ground-based measurements of the solar disk centre and rocket-borne observations, carefully inter-calibrated based on a model of the solar continuum. It extends from 0.2 to 1 μm with a spectral resolution of 0.01 μm.

2.2. Radiative Transfer Models

[7] The radiative transfer models used to calculate fluxes and heating rates are (i) a high resolution line-by-line (LBL) model (the modified version of GENL2 [Zhong et al., 2001; Edwards, 1992]), and (ii) a broad-band model [Edwards and Slingo, 1996] (herein after referred to as ES) with several different spectral parameter files. The LBL model provides benchmark results against which various parameterizations can be validated. The ES broad-band model is the radiation parameterizations used in the UK Met Office Unified Model. A spectral resolution of 0.01 cm−1 is used for LBL in this work. ES has six broad bands in the solar spectral region: three in the UV and visible regions and three in the near infrared (see Table 2) which generally provides adequate accuracy for flux and heating rate calculations in GCMs.

[8] The line-by-line calculations are performed for a mid-latitude summer atmospheric model which has 155 layers from the surface to 0.07 mb. Four absorber gases (O3, O2, H2O and CO2) and Rayleigh scattering are included and the Hitran2000 spectral parameters are used. The zenith angle is 53° and the surface albedo is 0.1 for all these calculations. The two different solar irradiance spectra (Kurucz and Lean) are used as input respectively. A solar flux of 1360 W/m2 is specified for all the line-by-line calculations which cover the region of 0.2 to 5 μm (and thus does not include the remaining approximately 6.3 W/m2 of the total solar irradiance which is emitted at longer wavelengths).

[9] The ES broad-band model is used both within the GCM and for off-line calculations. The basic version, subsequently referred to as BB1, has the ozone absorption parameters of Cusack et al. [1999]. For this work another version, BB2 with recalculated ozone absorption parameters, has been developed. The revision was made in order to improve ozone heating rate calculations and better incorporate solar variability. Briefly, the first UV band is divided into six relatively narrow sub-bands, each of which has only one ozone absorption coefficient so that, although the total number of bands is increased, the computational demands are similar to the previous k-distribution method for the UV band. Each new sub-band has physically realistic band limits and the ozone absorption coefficients are obtained from mean transmission functions calculated with high resolution (1 cm−1) and using a fitting procedures similar to that described by Chou and Lee [1996]. Ozone cross-sections used in the calculations are a combination of Hitran2004 (0.24–0.34 μm), Molina and Molina [1986] (0.2–0.24 μm) and Voigt et al. [2001] (above 0.34 μm). This new broad-band model has greater accuracy due to the higher number of bands within the UV. It also provides an easier vehicle for experiments in which variations in the solar irradiance spectrum may be imposed because the k-distribution of lines is restricted to the narrower bands.

[10] BB1 is used with the Labs and Neckel data and the Kurucz data; BB2 is used with the Lean spectrum. All other absorption and scattering parameters are identical for the three spectral files. Table 1 summarizes the solar spectra and the corresponding broadband models used in this paper. Table 2 shows the normalized solar fluxes for the six broad bands obtained from the three solar irradiance spectra. It can be seen that the Kurucz spectrum contains more energy in band 1 than estimated by Labs and Neckel with the Lean value lying between the two. For the off-line broadband calculations the same mid-latitude atmospheric profile and other parameters are used as for the line-by-line model.

Table 1. Solar Spectra Discussed in This Paper
Solar SpectrumSpectral ResolutionSourceGCM Broadband Model
Kurucz [1995]1 cm−1theoreticalBB1 (Kurucz)
Lean [2000]1 nmsatellite obs.BB2 (Lean)
Labs and Neckel [1970]5 nmground-based obs.BB1 (Lab and Neckel)
Table 2. Normalized Solar Fluxes in the Spectral Bands
Band NumberWavelength Range (μm)BB1 (Kurucz)BB1 (Labs and Neckel)BB2 (Lean)

2.3. GCM

[11] To examine the impact on GCM simulations of changes in shortwave spectral irradiance a version of the Met Office Unified Model, based on the UK Met Office Unified Model version HadGAM2-A L60, has been used. The model comprises a completely new dynamical core, running with semi-implicit, semi-Lagrangian timestepping solving a non-hydrostatic set of primitive equations. The model is finite difference based comprising 60 height levels on a Charney-Phillips grid spanning the surface through to an extended lid at 84 km; of these ∼35 are in the stratosphere. A horizontal resolution of 3.75° and 2.5° was maintained. The new version of GCM includes a number of improvements over previous versions including cloud radiation, the hydrological cycle and has a better tropopause structure [Ringer et al., 2006; Martin et al., 2006]. The version used here does not include an interactive photochemistry scheme, the ozone concentration field is prescribed and thus there are no feedbacks in the experiments presented here between radiation, ozone and temperature. The basic model incorporates the BB1 (Kurucz) radiation scheme.

3. Results

3.1. Heating Rates

[12] Solar heating rates are largest in the upper stratosphere/lower mesosphere due to absorption of UV radiation by ozone. Figure 1 (top) presents the line-by-line calculated spectral heating rate for the Lean [2000] solar spectrum as a function of height. The main heating region in the upper and middle stratosphere is due to the ozone Hartley band (200–320 nm) and part of the Huggins bands (300–390 nm). Heating in the lower stratosphere is mainly due to the ozone Chappuis band (380–800 nm) and part of the Huggins bands. A smaller contribution to the lower stratospheric heating around 700 nm is due to the ozone Wulf bands. Heating rates in the Chappuis and Wulf bands are by more than an order of magnitude smaller than those in the Hartley band. Therefore we focus on the spectral Band 1 (200–320 nm, see Table 2) for our analysis.

Figure 1.

Line-by-line model solar spectral heating rates for a mid-latitude summer atmosphere at solar zenith angle of 53°: (top) as a function of wavelength and altitude (Lean's [2000] spectrum); (middle) For Kurucz [1995] and Lean [2000] spectra and their difference at 50 km; (bottom) the same as Figure 2 (middle) but at 34 km.

[13] Figure 2 shows a comparison of heating rate profiles in Band 1 estimated using line-by-line and broadband models with the different solar spectra. The two line-by-line results (thicker curves) show that large differences (maximum 1.1 K/day) between using the Kurucz and Lean spectra occur in the upper stratosphere. Diagnosis of the reasons for the difference in heating rates is provided in Figure 1 (middle) and 1c which compare the spectral heating rates calculated at two different altitudes. The largest differences at ∼50 km exceed 20% with a maximum of ∼0.08 K/day/nm in the centre of the Hartley band and at ∼34 km about 40% (maximum ∼0.04 K/day/nm) shifted to the edge of the Hartley band and Huggins bands.

Figure 2.

Heating rates in the 200–320 nm region calculated for a mid-latitude summer atmosphere: thick curves represent line-by-line model results and thin ones are broadband model results.

[14] The broadband calculation with the Kurucz spectrum overestimates the LBL values around the stratopause with a maximum difference of 0.6 K/day. However, the broadband calculation using the Lean spectrum is in very good agreement with the corresponding line-by-line results, indicating that the new broad band parameterization (BB2) provides an improvement. The combined effects of the changes in spectrum and ozone parameterization mean that the maximum difference between Lean and Kurucz broadband calculations reaches around 1.7 K/day at ∼0.3 mb, with the difference in input spectrum accounting for about two-thirds of this. The broad band calculation with the Labs and Neckel spectrum is closer to the Lean broadband calculation (generally lower, by up to 0.7 K/day at ∼1 mb) although this actually reflects some cancellation between the effects of the spectrum and ozone parameterization.

[15] These differences in heating rates, caused mainly by the use of different solar irradiance spectra, will influence temperatures simulated in GCMs and potentially influence atmospheric dynamics and circulation. In the next section we investigate the impact on temperature.

3.2. GCM Experiments

[16] Previous experiments with the 60 level version of the Hadley Centre model found a warm bias in stratospheric temperatures (A. Bushell, personal communication, 2006). This is demonstrated in Figure 3a which shows globally averaged annual-mean temperature profiles from the model and from observations (UK Met Office Assimilations 1992–2001). Model temperatures exceed the observations by around 10 K near the stratopause. This annual mean bias has little variation with latitude (not shown) and is thus unlikely to be related to the model's gravity wave drag parameterization. Potential causes for the error include the prescribed ozone climatology or some spectral parameter in the radiation scheme (either longwave or shortwave).

Figure 3.

(a) Globally averaged annual-mean temperatures (1992–1999) from HadGAM2-A L60 runs and from the UK Met Office assimilated data. (b) The difference in global annual-mean temperature between two GCM runs using Kurucz [1995] and Lean [2000] solar spectra.

[17] Two 25-year simulations of the model were carried out with initial conditions specified for September 1978. The run-length was determined so as to clearly identify the stratospheric temperature bias and to better assess the significance of the changes to SW irradiance. Contemporaneous HadISST sea-surface temperature and sea-ice were applied, while seasonally varying, (though inter-annually stationary) zonal-mean ozone were prescribed (K. Rosenlof and M. Dall'Amico, personal communication, 2008). A linear trend in CO2 and methane mass-mixing ratio was employed, ranging from (5.1–5.6) × 10−4 and (8.2–9.7) × 10−7, respectively. Although the model has explicit schemes for aerosols, carbon and sulphur cycles, these were not included in these integrations. The radiation scheme is called every 3 hours.

[18] Using the Kurucz [1995] spectral data we find a similar warm bias to that of Bushell (personal communication, 2006) in comparison with the UK Met Office Assimilations. Using the Lean [2000] spectrum, however, we find a significant reduction in this bias. Figure 3b shows the difference introduced in replacing the Kurucz spectrum by that of Lean. Use of the latter brings down the globally averaged annual-mean temperature by nearly 4.3 K above the stratopause.

4. Conclusions

[19] From the high spectral resolution calculations for a mid-latitude summer atmosphere we have shown that the shortwave heating rates in the middle and upper stratosphere are very sensitive to the specification of the solar irradiance spectrum incident at the top of the atmosphere. The maximum differences in heating rate between using the Kurucz [1995] and Lean [2000] spectra are around 1.1 K/day in the upper stratosphere with the greatest difference in heating rate occurring in the central region of the ozone Hartley band. The maximum differences in heating rate between calculations using the Kurucz [1995] spectrum and the ground-based measurements of Labs and Neckel [1970] reach 1.8 K/day in the stratosphere for a mid-latitude atmosphere.

[20] Previous GCM simulations with the Met Office Unified Model have shown a warm bias in the middle atmosphere compared with observations. To diagnose whether this is associated with the absorption of solar radiation we have conducted a series of experiments using the model's broadband radiation scheme. With the same input solar spectrum we find that the broadband scheme overestimates heating rates compared with line-by-line calculations by a small amount. We have implemented a new parameterization which has a revised band structure and parameters for the ozone bands and provides a better match to the line-by-line calculations. It has the additional advantage, for future work, in that it can straightforwardly and accurately incorporate variability in solar spectral irradiance.

[21] Experiments with the GCM in which the radiation scheme improvements are implemented and the Kurucz [1995] spectrum is replaced by the Lean [2000] spectrum show significant reductions in the modelled global average annual-mean temperature, by 2–4 K in the upper stratosphere and nearly 4.3 K above the stratopause, ∼40% of the reduction can be associated to the improvement of the broad-band model but it is use of the Lean spectrum in the UV region that makes the dominant contribution. Further work on the reason for the bias could examine the use of different ozone datasets.

[22] Our results lead us currently to recommend the use of observational irradiance spectra, such as those of Thuillier et al. [2004] or Lean et al. [2005] in GCM radiation schemes.


[23] This work was supported by the UK Natural Environment Research Council consortium project SOLCLI. We thank Andrew Bushell, Tim Hinton and Gill Martin for providing their GCM results, and Keith Shine for helpful discussions.