2.1 A new solar reference spectrum

The CMIP6 SSI forcing dataset that was recommended by M17 is an average of two time series from two SSI reconstruction models: Naval Research Laboratory Solar Spectral Irradiance Version 2 (NRLSSI2; Coddington et al., 2016) and Spectral And Total Irradiance REconstruction (SATIRE; Yeo et al., 2014). Both models rely on a constant, so-called quiet-Sun reference spectrum on top of which comes the solar variability. NRLSSI2 uses a composite of quiet-Sun spectra, namely, the Whole Heliosphere Interval (WHI) spectrum (Woods et al., 2009) below 300 nm, the spectrum from the first Atmospheric Laboratory of Applications and Science (ATLAS-1) space shuttle mission (Thuillier et al., 1998) between 300 and 1000 nm, the spectrum from NASA's Solar Radiation and Climate Experiment (SORCE) Spectral Irradiance Monitor (SIM) instrument (Harder et al., 2005) between 1000 and 2400 nm (also used by WHI), and Kurucz's synthetic solar model atmosphere beyond that range (Kurucz, 1991). SATIRE uses the WHI spectrum in the 115–2400 nm range, extended at longer wavelengths by Kurucz's atmosphere model. Ultimately, the CMIP6 quiet-Sun spectrum is the average of both reconstruction models and, therefore, mixes two somewhat different background spectra.

Over recent years, a number of additional solar reference spectra based on observations have become available. First, there is the SOLAR-ISS reference spectrum by Meftah et al. (2018), which is based on the SOLAR/SOLSPEC observations (Thuillier et al., 2009) combined with the synthetic spectrum by Kurucz (1991). Second, there is the quiet-Sun reference spectrum using the observational SSI composite by Haberreiter et al. (2017) for the annual mean of 2008 combined with the synthetic calculations using the COde for Solar Irradiance (COSI; Haberreiter et al., 2008, 2021). Third, Coddington et al. (2021) provide a hybrid reference spectrum which is based on the latest observations from the Total and Spectral Solar Irradiance Sensor (TSIS-1) Spectral Irradiance Monitor (TSIS-SIM; Richard et al., 2020) onboard the International Space Station (ISS). According to Richard et al. (2020), the absolute uncertainty of the TSIS-1 SIM instrument is 0.2 %–0.5 %. This value is better than the absolute uncertainty of the WHI or the ATLAS-1 spectra (typically > 3 % in the visible range). Version 2 of the TSIS-1 reference spectrum, which is an incremental update, has recently been published (Coddington et al., 2023).

A major difference between the TSIS-1 spectrum and the quiet-Sun spectrum in CMIP6 is a distinct spectral shape, with the TSIS-1 spectrum showing an irradiance that is 1 %–5 % higher in the visible band and 1 %–2 % lower in the near-IR wavelength range (between 1000 and 2000 nm), after re-normalization to the same value of the TSI. This difference is illustrated in Fig. 1, which compares the irradiance with that of TSIS-1 for specific spectral bands.

Figure 1Ratio between the irradiance (in specific spectral bands) from SATIRE, NRLSSI2, CMIP6 and WHI and that of the TSIS-1 spectrum. For CMIP6, SATIRE and NRLSSI, the reference spectrum is estimated as the mean value of the SSI for three time intervals between 25 March and 16 April 2008, which are the same as the those used for estimating the WHI reference spectrum. The spectral bands are the far-UV (FUV), middle-UV (MUV), near-UV (NUV), visible (VIS), near-infrared (NIR), short-wavelength infrared (SWIR) and middle-wavelength-infrared (MWIR) bands, respectively.

Download

Such differences have direct implications on the climate response. For example, Jing et al. (2021) investigated the impact of the new TSIS-1 solar spectral irradiances, compared with earlier data, in NCAR CESM2 coupled climate model simulations. They found that the energy shifts between the visible and the near-infrared parts of the spectrum can trigger surface albedo feedbacks, resulting in significant differences between modeled high-latitude surface temperature and sea ice coverage.

Despite its different spectral shape, we consider the TSIS-1 reference spectrum (version 2) to be the most reasonable choice for future climate simulations. Indeed, the spectrum is based on the latest measurements with significantly increased accuracy compared with prior similar measurements, and it has undergone a detailed validation. In addition,the Committee on Earth Observation Satellites (CEOS) Working Group on Calibration and Validation (WGCV) recommended (in March 2022) that it be used as the new reference spectrum (https://calvalportal.ceos.org/tsis-1-hsrs, last access: 1 February 2024).

Finally, let us stress that the choice of the reference spectrum in the NRLSSI2 and SATIRE models is decoupled from the temporal variability in the spectral irradiance. The two are defined independently and are then added together.

2.2 A consistent representation of solar irradiance variability

The main challenge in making a historical solar radiative forcing dataset is the need to reconstruct SSI/TSI from proxy data for periods prior to their direct observation. Observations of the time-resolved solar spectrum from the extreme-UV to the near-IR are available for the period from 2003 onward, whereas direct TSI observations started in 1978 (Ermolli et al., 2013).

Nowadays many reconstructions of TSI and SSI coexist. For instance, SATIRE derives the SSI/TSI from synthetic intensity spectra; it uses full-disk-resolved filtergrams and magnetograms taken at visible wavelengths after 1974 as well as solar proxies such as sunspot observations before that date. NRLSSI2 is more data driven, as it uses measured spectra to adjust SSI variability via solar proxies. These are (since 1982) the University of Bremen Mg II measurement composite data and the areas and locations of sunspots as reported by the United States Air Force (USAF) Solar Observing Optical Network (SOON) sites. For sunspot region information prior to 1982, Greenwich Observatory observations are used.

Besides the NRLSSI2 and SATIRE reconstructions, there are additional SSI reconstructions available. It should be noted that all reconstruction approaches, including NRLSSI2 and SATIRE, are based on the assumption that the irradiance variations are caused by the changing magnetic features on the surface of the Sun, but they differ with respect to the implementation of those changes. Egorova et al. (2018) use the Code for the High spectral ResolutiOn recoNstructiOn of Solar irradiance (CHRONOS). CHRONOS is an update of the reconstruction approach by Shapiro et al. (2011) that includes a revised method to derive the varying contributions of the quiet Sun, faculae, sunspot umbra and sunspot penumbra, and the combined spectra. Different versions of the CHRONOS reconstruction exist that are based on different input parameters to derive the long-term evolution of the quiet-Sun irradiance; for a comparison with NRLSSI and SATIRE, the reader is referred to Yeo et al. (2020a, Fig. 1b). For discussion of further irradiance reconstruction models, we refer the reader to the reviews by authors such as Ermolli et al. (2013) and Chatzistergos et al. (2023).

In CMIP6, we selected the only two models that could reconstruct solar irradiance over the whole period considered and that had been studied in detail. These were SATIRE and NRLSSI2. Both use various solar inputs (e.g., magnetograms and sunspot number) but differ with respect to the way that these translate into SSI/TSI variability. These differences, along with the use of different versions of proxy datasets, have led to systematic discrepancies in solar cycle amplitudes and secular trends. These have been the subject of much debate, but no consensus on the most appropriate model for climate simulations has been reached by the community to date. Both models come with uncertainty estimates; however, because they are based on different metrics, these estimates cannot be meaningfully compared in a quantitative way. For these reasons, for CMIP6, it was decided that the two reconstructions should be averaged without favoring either model. In contrast, for CMIP5, only the NRLSSI (Lean, 2000) model was used.

While the averaging applied for CMIP6 was considered to be the most sensible choice, given the available information, it has also received criticism. One problem arises from the different reference spectra that are used by both models (see Sect. 2.1), the averaging of which leads to yet another spectrum of the composite. Another objection comes from the different trends that arise during the space era (after the 1980s): SATIRE produces a stronger downward trend in the SSI observed at solar minimum compared with NRLSSI2 and with the most recent measured TSI.

For CMIP7, the objective is to revisit these choices in the light of recent developments made by both model teams and to find a pragmatic solution that would provide the best solar input for climate models.

In the meantime, no community consensus has been reached regarding the relative accuracy of the models. Both are continuously being improved and the agreement between them tends to improve (Lean et al., 2020). Solar surface magnetism has been confirmed to be the main driver of SSI variations (Yeo et al., 2017); therefore, growing attention has been given to its role, which is crucial for constraining the SSI/TSI during periods of very low solar activity, such as during the Maunder Minimum (Yeo et al., 2020b; Krivova et al., 2021; Wang and Lean, 2021). New data sources are gradually becoming available, such as full-disk-resolved solar images taken in the Ca II K line since 1892, that provide new insight into the long-term evolution of surface magnetism (Chatzistergos et al., 2022). At this stage, these new data are still mostly used for validation purposes.

Another aspect to be considered for CMIP7 is the consistency of the recommended solar forcing with that to be used for paleoclimatic reconstructions in Phase 5 of the Paleoclimate Modelling Intercomparison Project (PMIP5). At the time of writing, the latter is not yet known. The recommended solar forcing for PMIP4 was based on SATIRE-M (Jungclaus et al., 2017), which was not fully consistent with the forcing used for CMIP6.

Another issue is the production of the SSI/TSI dataset for future scenarios. As for CMIP6, we are planning to provide a set of forcing scenarios with daily values up to 2300. These will be produced from one single solar input (the sunspot number or the group number), similarly to the way historical reconstructions of the SSI/TSI are made for the period before solar images or magnetograms became available. For internal consistency, it would be preferable to use the same solar proxies as well as the same version in both models. Unfortunately, different versions of the sunspot number record coexist (Clette et al., 2023), which has led to small but significant differences in historical solar forcing (Kopp et al., 2016). Finally, there are practical considerations, such as the process for building historical and future forcings, which should be flexible enough to allow for operationalization and regular updates, with a short latency.

What is the best solar irradiance forcing dataset for CMIP7? Considering the absence of a community consensus on the models and the lack of comparable uncertainty estimates, the most reasonable choice would be to again average the latest versions of the two SSI/TSI models, namely, SATIRE (including its recent improvements) and NRLSSI3 (or possibly NRLSSI4, which is in preparation). However, to avoid some of the problems that were encountered in CMIP6, a meaningful average requires that both models use the same reference spectrum (see the suggestion made in Sect. 2.1) and are driven by the same solar proxies, namely, either sunspot number or group number. Furthermore, a consistent treatment of TSI/SSI variability in the CMIP7 historical and PMIP time periods should be considered.

3.1 Improved estimates of the top-of-the-atmosphere MEE fluxes

Mid-energy electron precipitation fluxes are derived from the Medium Energy Proton and Electron Detector (MEPED)/Polar Orbiting Environmental Satellites (POES) instruments, which provide observations in three energy bins (≥ 30, ≥ 100 and ≥ 300 keV) and at two perpendicular viewing angles (Evans and Greer, 2000). For CMIP6, electron fluxes were extracted using data from only the 0^∘ telescope (van de Kamp et al., 2016). The low bias of the fluxes used in CMIP6 has been primarily attributed to an underestimation of the loss cone when using only the 0^∘ telescope. Datasets based on an estimate of the loss cone combining the 0 and 90^∘ telescopes and using daily observations provide higher fluxes (Nesse Tyssøy et al., 2019; Nesse Tyssøy et al., 2022) and lead to a stronger atmospheric response (Sinnhuber et al., 2022; Pettit et al., 2021). Therefore, for CMIP7, improvement of the estimates of precipitating fluxes by using data from both telescopes is proposed, e.g., based on the approach of Nesse Tyssøy et al. (2016) on the new homogeneous composite developed by Asikainen and Ruopsa (2019) and Asikainen (2019), which will enable the estimate of precipitating fluxes over the full observation period from 1979 to present day in the energy range from 30 to 1000 keV. The long observation period covering multiple solar cycles allows for a better foundation and validation of the MEE parameterization.

3.2 Refined reconstruction of MEE fluxes

As a response to the underestimation of the MEE fluxes in M17, updated particle flux observations as outlined in the previous section should be used to construct an updated precipitation model. Consistent with M17, we propose following the theoretical framework of van de Kamp et al. (2016) for parameterizing the fluxes on L-shells in terms of geomagnetic index, but we recommend doing this based on estimated electron fluxes using data from both MEPED/POES telescopes (see above). At this stage, no need for including the magnetic local time (MLT) dependency in the fluxes (van de Kamp et al., 2018) has been identified (Verronen et al., 2020).

Further developments to overcome the current deficiencies in the atmospheric impact could come from the following: (1) using an alternative geomagnetic index to Ap (e.g., the aa index could be used directly to reconstruct the long-term dataset); (2) incorporating a lagged or an accumulated response to better represent the temporal evolution of geomagnetic storms; and (3) using a piecewise energy spectra power law for extracting spectra in the range from 30 to 1000 keV, rather than the single power law approach in M17. The motivation of the latter arises from seeking improvements for the fluxes in the high-energy tail of the spectrum, which were likely underestimated in M17. Separation of the spectral fits by energy range could then further allow for a delayed impact (∼ 2 d delay) of high-energy electrons, in a manner consistent with what is seen in observations (Nesse Tyssøy et al., 2021; Salice et al., 2023). Finally, if the validation reveals that the dependent variable (aa or Ap) has a wide range of possible flux responses where an average representation would dampen the overall precipitating flux variability, implementing a stochastic solar-cycle-dependent element should be considered.

3.3 Further minor updates

We also propose the following minor updates:

• The atmospheric ionization rates from MEE precipitation in M17 were calculated using the formulation of Fang et al. (2010), which is accurate over the energies up to 1 MeV but does not consider the secondary bremsstrahlung peak at lower altitudes. A new parameterization for this calculation has been formulated by Xu et al. (2021) for the high-energy tail from 100 to 1000 keV considering bremsstrahlung and could be used to replace (or extend) the Fang et al. (2010) parameterizations. However, the impact of including this on the atmospheric composition is likely small.

• Since M17, we are aware of no studies that have highlighted significant deficiencies in the specification of solar energetic particle (SEP) fluxes. SEP fluxes are derived from Geostationary Operational Environmental Satellite (GOES) observations in the energy range from a few megaelectronvolts (MeV) to 100 MeV, and they are extrapolated to 300 MeV. This yields good results in the upper stratosphere and mesosphere above 35–40 km, the altitude region most affected by solar proton events (e.g., Jackman et al., 2001; Funke et al., 2011). However, in rare events with a harder spectrum, this can lead to an underestimation of the SEP impact below this altitude (Jia et al., 2020). Therefore, the energy range should be extended from 100 MeV (used in M17) to 400 MeV using a recent recalibration of the GOES detectors (Raukunen et al., 2022). This would account for the previously missing contribution of SEPs to stratospheric chemistry in the vertical range of ∼ 20–40 km. Ionization rates associated with the lower-energy part of the energetic particle spectrum can still be calculated as in M17 using the analytical approach by Jackman et al. (1980). This approach, however, cannot be applied to high-energy (> 100 MeV) protons which initiate the atmospheric nucleonic cascade and can penetrate to the lower atmosphere. For that, we propose an approach based on the ionization yield functions precomputed with a physics-based model, based on Monte Carlo simulations of the atmospheric cascade, CRAC:CRII (Usoskin and Kovaltsov, 2006; Usoskin et al., 2010, 2011; Väisänen et al., 2023). The long-term dataset covering the historical period from 1850 to 1962 can be reconstructed in the same stochastic manner as in M17.

• We propose treating galactic cosmic rays (GCRs) in a similar way to M17, i.e., by means of the force-field approximation parameterized via the modulation potential Φ. However, Φ should be obtained from the ground-based neutron monitor dataset that begins in 1951 (Usoskin et al., 2005, 2017) and is continuously updated at https://cosmicrays.oulu.fi/phi/phi.html (last access: 1 February 2024). For the historical period before 1951, the Φ time series can be based upon the solar open-flux model of Krivova et al. (2021).

• Geomagnetic shielding affects the spatial distribution of atmospheric ionization by GCRs, SEPs and magnetospheric electrons. For CMIP7, the approach implemented by M17 using the International Geomagnetic Reference Field (IGRF) model truncated to the eccentric tilted dipole component (the first eight Gaussian coefficients) is proposed, which is known to adequately represent the realistic field for the cosmic-ray shielding at the global scale (Nevalainen et al., 2013). The newest version of the IGRF, the 13th-generation model (Alken et al., 2021), is recommended for use.

• The main impact of energetic particle precipitation on the composition, independent of the particle source, is the formation of NO_x (N, NO) and HO_x (H, OH) by atmospheric ionization. This is implemented in CCMs using simple parameterizations that were first outlined by Porter et al. (1976) and Solomon et al. (1981) or by including the complex D-region ion chemistry (e.g., Verronen et al., 2016). The simple parameterization approach has been recommended for CMIP6 (M17). It yields overly low NO_x formation in the lower thermosphere (Nieder et al., 2014) but has been shown to perform well throughout the middle atmosphere below ∼ 80 km altitude in many studies. An altitude parameterization of the NO_x formation similar to that for HO_x could be constructed based on Nieder et al. (2014), although this would only have significant implications for models extending higher than 1 Pa.

• For those CCMs with an upper lid in the mesosphere, an odd-nitrogen upper-boundary condition (UBC) is required, accounting for EPP production higher up. M17 recommended the use of the UBC model described in Funke et al. (2016), which is based on Michelson Interferometer for Passive Atmospheric Sounding (MIPAS) observations taken during the 2002–2012 period. It is planned to maintain the same approach for CMIP7; however, an extended validation of the UBC model with more recent NO observations (and a possible update, if required) should be considered. In addition, the use of the aa index instead of the Ap index (similar to that for the MEE reconstruction) to drive the UBC model should be explored.