Comparisons of simulated radiation, surface wind stress and SST fields over tropical pacific by the GISS CMIP6 versions of global climate models with observations

This study compares the overall performance between versions 2.1 and 3 of National Aeronautics and Space Administration (NASA) Goddard Institute for Space Studies (GISS) global climate models (referred to as GISS-E2.1 and GISS-E3, respectively), in simulating the present-day Pacific climate using the CMIP6 protocol. Model physical representations and configurations are extensively changed from GISS-E2.1 to GISS-E3, which result in greatly reduced discrepancies, including ice water path (IWP), ice water content (IWC), radiative fluxes, surface wind stress (TAU), sea surface temperature (SST), precipitation (PR) and column water vapor (PRW), relative to satellite-based observational products over south Pacific oceans. Cloud only IWP (CIWP) shows the largest change, decreasing biases from ∼400 g kg−1 in GISS-E2.1 to 10–20 g kg−1 in GISS-E3. The combination of improved CIWP and the inclusion of snow in GISS-E3 may play roles on reducing overestimated outgoing longwave radiation, overestimated reflected shortwave at the top of atmosphere, and underestimated surface downward shortwave in GISS-E2.1. Both models’ intertropical convergence zones (ITCZs) are, however, located far too north of the equator, as found in radiative fluxes, PR and PRW but not in SST relative to observations. This introduces biases in TAU, PR and PRW over north flank of the equator and north Pacific. Over south Pacific, especially the trade wind regions, the improvements of radiation fluxes, SST, PR and PRW appear to be due to improved TAU associated with inclusion of snow-radiative effects. In particular, GISS-E3 reduces a longstanding too warm SST bias over trade-wind regions, from 4 K in GISS-E2.1 to within 0.5 K, and too cold SST bias over north Pacific Ocean. Although GISS-E3 shows improved geographic patterns of the simulated fields in particular over south Pacific oceans compared to GISS-E2.1, our results suggest that the location of ITCZ needs to be further improved.


Description of GISS-E2.1 and GISS-E3 models
Here we briefly describe the differences and changes from GISS-E2.1 to GISS-E3. GISS-E2.1 model (Kelley et al 2020)-an evolution of the CMIP5 GISS model (GISS-E2, Schmidt et al 2014)-was submitted to CMIP6 already while GISS-E3 is a newer version of the GISS model with substantial structural changes (Cesana et al 2019(Cesana et al , 2021, which will be submitted to CMIP6 but will still undergo additional small changes by then. Following Cesana et al (2019), we highlight major differences in configurations and physical processes representations between GISS-E3 and GISS-E2.1 that can potentially affect tropical Pacific oceans climate are: 1. Vertical model layers in lower troposphere: GISS-E2.1 uses a 40-layer vertical grid versus 110 levels in  with the greatest refinement in the lower atmosphere important for representing planetary boundary layer (PBL) clouds, but also a better refinement in the upper atmosphere, which has been extended from 0.1 mb for E2.1 to 0.002 mb for E3.
2. PBL Turbulence. The modified Yao and Cheng (2012) scheme (Kelley et al 2020) used in GISS-E2.1 includes nonlocal transport but does not consider moist processes. In contrast, GISS-E3 uses the Bretherton and Park (2009) scheme, which considers both processes, and also the scheme of Wilson and Ballard (1999) to compute ice cloud fraction.
3. Stratiform cloud macrophysics. Both GISS-E2.1 and GISS-E3 use a diagnostic determination of cloud fraction as a function of grid-mean moisture and a condition-dependent sub-grid variance expressed as a threshold grid-mean relative humidity (RH) for cloud formation. The Sundqvist-type scheme of GISS-E2 (Del Genio et al 1996), applied identically to water and ice clouds. A triangular probability density function PDF schemes is used in GISS-E3 to compute water cloud fraction and cloud water mixing ratio (Smith 1990). Stratiform cloud microphysics in GISS-E2.1 uses a temperature-dependent autoconversion rate of supercooled liquid to ice precipitation to describe cloud phase transitions. This rate is maximum at −35°C and decreases linearly toward the warm-cloud autoconversion rate at −5°C. This scheme (Kelley et al 2020) is replaced in GISS-E3 by a two-moment microphysics scheme with prognostic precipitation ( was implemented into GISS-E3, with cold pools impacting the plume updraft properties at cloud base and having the net effect of reducing the occurrence of less-entraining convection. In addition, instead of glaciating updrafts at the melting point, updrafts now glaciate at a temperature that is treated as a tuning parameter.

5.
Convective cloud microphysics. For ice, particle size distributions (PSDs) and size-fall-speed relationships used in GISS-E2.1 (Del Genio et al 2005) have been replaced for GISS-E3 with field-experiment-based normalized gamma PSDs and fall speeds for ice described by Elsaesser et al (2017) for a large-particle mode. A second small particle cloud mode was also implemented into GISS-E3 (a gamma PSD), with a volume-mean cloud ice particle radius introduced and used as a tuning parameter for determining the partitioning of ice mass between the two PSDs. For liquid, a similar bi-modal representation is also implemented into GISS-E3, with the rain-droplet size distribution (DSD) following Thompson et al (2008), but with a different shape parameter (following Shipway and Hill (2012)), and a similar tuning parameter introduced and used for liquid partitioning across the two DSDs modes.
6. Other relevant changes. E2.1's treatment of ice radiative properties as spheroids has been replaced by a method based on the MODIS collection 6 treatment, with the degree of fluffiness for stratiform cloud ice and snow being treated as tuning parameter. Finally, a number of changes have been made in the radiation scheme emerging from a Radiative Forcing Model Intercomparison Project (RFMIP) exercise.

CloudSat-CALIPSO 2C-ICE hydrometeors
For hydrometeors reference for ice water content (IWC) and path (IWP), we use the CloudSat-CALIPSO radarlidar based 2C-ICE cloud product (Deng et al 2013). Readers desiring a more in-depth description of the 2C-ICE algorithm should refer to Deng et al (2010) for details. Based on a flag-based partitioning method (Waliser et al 2009, Li et al 2012, we separate total ice water content (TIWC) and total ice water path (TIWP) into larger particle mass, which is considered to be precipitating (either stratiform or convective form) including snow, graupel and hail, and smaller particle mass, which is considered to be quasi-suspended and nonprecipitating cloud ice (either stratiform or convective form). Here, we briefly describe the FLAG method with more details referred to Li et al (2012) and Waliser et al (2009). For stratiform cloud only (suspended) cloud ice (CIWC), we exclude all the retrievals in 2C-ICE level-2 granules, in any profile that are flagged as precipitating at the surface and exclude any retrieval within the profile whose cloud type is classified as 'deep convection' or 'cumulus' (from CloudSat 2B-CLDCLASS data). By excluding these portions of the ice mass, we obtain an estimate of the cloud- There are caveats and uncertainties associated with the above-mentioned observational estimates, such as strong radar attenuation under intensive convective activities as well as lidar attenuation leading to underestimate total IWP/IWC in particular over such as storm tracks, ITCZ/SPCZ and Tropical Western Pacific regions. In addition, there are differences between observation and model representations of the precipitating and cloudy ice components. The data used is limited to 2007-2010 as these are the only complete years of data available for all products. The limiting dataset is 2C-ICE, which lost nighttime data following a CloudSat battery anomaly in 2011. All satellite datasets are regridded onto a 1°×1°latitude-longitude grid.

Radiation datasets
We use Edition 4.0 Clouds and the Earth's Radiant Energy System-Energy Balanced and Filled (CERES-EBAF) product (Kato et al 2011, 2012a, 2012b, Loeb et al 2009, 2018 as references shown in figure 3(c) for annual mean top of atmosphere (TOA) upward thermal emission radiation (RLUT, aka, outgoing longwave radiation, OLR) and figure 3(f), for TOA reflected shortwave (RSUT). Figure 3(i) shows reference for downward shortwave fluxes at surface (RSDS) computed with a radiative transfer model but constrained by the TOA fluxes from Edition 4.0 CERES-EBAF data product (Kato et al 2011(Kato et al , 2012a(Kato et al , 2012b. The data period of 2001-2014 is used.

Surface wind stress
We use surface wind stress estimates from QuikSCAT scatterometer from September 1999 to October 2009 (Risien and Chelton 2008) available in the Scatterometer Climatology of Ocean Winds (SCOW: http://cioss. coas.oregonstate.edu/scow/). Even the period of estimates are short with ten years, Lee et al (2013) stated that the multidecadal variability in CMIP models is much smaller than the biases in climatology. This means that the QuikSCAT climatology is a reasonable reference data set to evaluate GISS-E2.1 and GISS-E3 models' biases averaged over a multidecadal period.

Sea surface temperature
The Extended Reconstructed Sea Surface Temperature (ERSST) version 5 dataset for the period of 1970-2014 is used (data source: https://www1.ncdc.noaa.gov/pub/data/cmb/ersst/v5/). The data period is the same as that during which GISS-E2-1 and GISS-E3 output is analyzed. ERSSTv5 is a global monthly dataset derived from the International Comprehensive Ocean-Atmosphere Dataset release 3.0 SST on a 2°× 2°grid with spatial completeness enhanced using statistical methods (Huang et al 2017(Huang et al , 2018a(Huang et al , 2018b.

Surface precipitation dataset
The Global Precipitation Climatology Project (GPCP) Version 2.3 data set combines precipitation observations over land and ocean from low-orbit satellite microwave data, geosynchronous-orbit satellite infrared data and, over land, surface rain gauge observations to provide gridded monthly precipitation from January 1979 to May 2019 (Adler et al 2003(Adler et al , 2018. The data for January 1980 to December 2014 are used in this study and we take the mean of the monthly time series for each grid cell to represent our observational estimate of precipitation climatology. same as (a) but for stratiform falling ice hydrometeors (snow + graupel + hail) (c) same as (b) but for combined convective and stratiform estimated; (d) same as (c) but for total ice water content; (e) same as (a) but simulated from CMIP6 GISS-E3 averaged over 1980-2014; (f) same as (b) but for GISS-E3 snow water content (SWC); (g) same as (c) but for E3 Stratiform + Convective (snow + cloud ice); (h) same as (d) but for GISS-E3 stratiform + convective form snow + cloud ice. (i) same as (e) but for GISS-E2.1 averaged over 1980-2014.

Radiation budget components
To see the improvements from GISS-E2.1 to GISS-E3, figure 4 shows the differences in annual mean radiation budget components (RLUT, RSUT and RSDS) between GISS-E2.1 and GISS-E3 with 99% confidence level (upper row in figure 3), the biases of GISS-E2.1 (2nd row) and GISS-E3 (3rd row) relative to CERES and the differences in absolute biases (4th row). With 99% confidence level there are more RLUT (figure 4(a)) and RSDS (figure 4(i)) with less RSUT (figure 4(e)) in GISS-E2.1 than in GISS-E3 over north Pacific from 10 N to 30 N (models' ITCZs) and southwest Pacific around SPCZ, which might be contributed by adding radiatively active stratiform and convective snow ice mass suggested by Li et al (2015Li et al ( , 2016 and Waliser et al (2011), as discussed in section 1.
In general, GISS-E3 has substantially reduced absolute biases relative to GISS-E2.1 in all components over south Pacific trade wind regions and stratocumulus regions: typically reduced from 20-40 W m −2 shown in figure 4(d) for RLUT, figure 4(h) for RSUT and RSDS in figure 4(l). The more northward shifted ITCZ in GISS-E3 relative to GISS-E2.1 might be the reason for excessive RLUT and RSDS and underestimated RSUT over CERES data for all radiation components in GISS-E3 (figures 4(b), (c), (f), (g), (j), (k)). Thus, GISS-E3's bias has actually worsened compared to GISS-E2.1 over the north Pacific except for the stratocumulus region off the coast of California (from 30-40 W m −2 in GISS-E2.1 to less than 10-20 W m −2 in GISS-E3).
The seasonal-mean radiation component biases for December-January-February (DJF) (figure A1) and June-July-August (JJA) (figure A2) are, in general, very similar to the annual mean biases (figure 4) except that RLUT is improved in GISS-E3 over both north and south Pacific in JJA (figure A2(d)), due to the northward marching of observed ITCZ. GISS-E3 radiation improved in both seasons with the absolute mean biases substantially reduced relative to GISS-E2.1 over south Pacific trade wind regions: reducing biases up to 40 W m −2 but still with non-trivial biases about 20 W m −2 shown in figure A1(d)/A2(d) (DJF/JJA) for RLUT, (A1h/A2h) for RSUT and A1l/A2l) for RSDS. It seems that RSUT (figure A2(h)) and RSDS (figure A2(l)) are inconsistent, which should be quite symmetric but not in several areas in GISS-E2.1, for example, near 30 N, and that GISS-E2.1 is worse in RSUT but better in RSDS. Degradation for GISS-E3 is seen over the north Pacific and except for coastal stratocumulus regions in DJF because of simulated ITCZ does not move southwards as much as the observed ITCZ.

Surface wind stress, sea surface temperature, precipitation and total column water vapor
It is expected that the abovementioned spatial changes in radiation components with biases reduction/ enhancement in the magnitude of discrepancies relative to CERES from GISS-E2.1 to GISS-E3 would lead to changes in surface wind stress and sea surface temperature. In our previous studies summarized in Supplementary Information (SI), we found that without FIREs inclusion, there are excessive RLUT over 'modeled ITCZ' convective zones, which would lead to weaken surface wind stress (with westerly wind stress and divergent meridional TAUV away from modeled ITCZ) and associated with warm SST, excessive precipitation and PRW in particular over south Pacific trade-winds regions. In this section, we examine the changes of surface wind stress, precipitation, sea surface temperature (SST) and total column water vapor (PRW) between GISS-E2.1 and GISS-E3. Figure 5 shows comparisons of zonal surface wind stress (TAUU), meridional surface wind stress (TAUV) biases (panels a, b, d, e) and differences in absolute biases (panels h, i) for GISS-E2.1 and GISS-E3 relative to QuikSCAT estimates, as well as differences in vectors (TAU), TAUU and TAUV between GISS-E2.1 and GISS-E3 (panels c, f, h). Figure 5(a) shows that GISS-E2.1 has strong westerly biases over the tropics (15 S-15 N) and strong easterly biases in subtropics/mid-latitudes north of 15 N and south of 15 S while meridional wind stress (TAUV) biases diverge away from the 'modeled' ITCZ north of the equator ( figure 5(d)). The TAUU biases imply for anomalous low-level cyclonic circulations over the subsidence regions north of equator. In GISS-E3 ( figure 5(b)), the westerly TAUU bias is reduced mainly over south flanks of equator (15 S-15 N) but still with enhancing westerly over north of the equator accompanied by equatorial easterly bias. There are also stronger easterly biases over north Pacific oceans (north of 15 N) in GISS-E3 (figures 5(b), (h)). The changes in TAUV from GISS-2 to GISS-3 are different between the two hemispheres. There are stronger divergence anomalies over the modeled ITCZ but smaller biases over south Pacific Ocean and off coast of Peru relative to GISS-E2.1 (figures 5(e), (i)).
Shown in figure 5(g), the TAU differences between GISS-E2.1 and GISS-E3 exhibits clear south-easterly TAU vectors differences south of modeled ITCZ, indicating weaker TAU in GISS-E2.1 relative GISS-E3 against the prevailing trade wind directions. The differences in its components are all statistically significant at 99% level (figures 5(c), (f)).
Next, figure 6 shows SST, PR and PRW maps of GISS-E2.1 and GISS-E3 along with their respective observations while figure 7 shows their biases. Although there is the far north displacement of the modeled ITCZ found in CIWP and radiation fields, the modeled maximum SST locations in GISS-E2.1 (figure 6(a)) and GISS-E3 (figure 6(b)) are better matched but still in GISS-E3 with the slightly north and too wide compared to maximum SST shown in ERSST estimates (figure 6(c)). The above-mentioned improved TAU biases over south Pacific trade wind regions found in GISS-E3 might lead to improved SST and PR against GISS-E2.1 (Li et al 2020b(Li et al , 2022a. Weaker TAU in GISS-E2.1 (figure 6(g)) implies much warmer SST bias (figures 7(a) and (b)) than in GISS-E3 ( figure 7(c)). In general, the reduction of SST biases is up to 4-7 K over south Pacific and 1-4 K over north of 30 N ( figure 7(d)).
The warm SST, excessive precipitation (figures 6(d), 7(f)) and PRW (figures 6(g), 7(g)) are found south of the equator in GISS-E2.1 against GPCP (figure 6(f)) and RSS PRW (figure 6(i)). Discrepancies of PR versus GPCP in general in most areas from GISS-E2.1 to GISS-E3 (figure 7(g)) the maximum precipitation is not reduced in particular over a band north of equator. Note that the maximum PR location, i.e., modeled ITCZ, is at the same location indicated by CIWP and radiation fields, suggesting that it might be caused by the atmospheric process rather than controlled by better simulated maximum SST (figures 6(a), (b)). GISS-E3 shows smaller PR/PRW biases than GISS-E2.1 in most areas, except for a small band in the eastern ITCZ (figures 6(d), (g)). Excessive PR is also found extended southeastward from SPCZ relative to GPCP estimate (figure 6(f)). The hydrometeor-radiation-circulation interactions can be explained with the above-mentioned differences in TAUU and TAUV between GISS-E2.1 and GISS-E3 with the weakened TAU (figure 6(g)) producing warmer SST (figure 7(a)), higher precipitation (figure 7(e)) and PRW (figure 7(i)) in GISS-E2.1 relative to GISS-E3. The SST biases are reduced drastically from GISS-E2.1 (figure 7(b)) to GISS-E3 (figure 7(c)) with improved TAUU and TAUV over north Pacific and southeast Pacific indicated in figure 7(d). The PR and PRW biases are also reduced from GISS-E2.1 (figures 7(f) and (j)) to GISS-E3 (figures 7(g) and (k)) through improved SST but mainly over south Pacific. The drastically improved PR (figure 7(h)) and PRW (figure 7(l)) from GISS-E2.1 to GISS-E3 are found but limited in south Pacific regions associated with the improved TAU.

Summary and discussion
GISS-E3 is the first ever -to our knowledge-GCM including complete hydrometeors-radiation interactions, which represents convective and stratiform types of both floating and precipitating ice (snow) and liquid (rain) and considers their radiative effects, while GISS-E2.1 represents only the radiative effects of both floating ice and liquid. We examined and compared the differences between simulated subtropical and tropical Pacific climate between GISS-E2.1 and GISS-E3 following CMIP6 'historical' scenario protocol. Used for comparisons are the annual mean frozen hydrometeors content and path (CIWP/CIWC/SWP), radiation budget (RLUT, RSUT and RSDS), surface wind stress (TAU), sea surface temperature (SST), precipitation (PR) and total column water vapor path (PRW) of satellite-based observational products from 2C-ICE, CERES, QuikSCAT, ERSST, GPCP and RSS, respectively.
By comparing IWP/IWC, radiation, surface wind stress, SST, precipitation and PRW changes together, we have provided a comprehensive understanding of the model performance, which is summarized in the following: Regions of the ITCZ, SPCZ and TWP exhibit comparable CIWP (figure 1) and CIWC (figure 2) in GISS-E3 relative to 2C-ICE in terms of magnitudes while GISS-E2.1 has large overestimates. The modeled ITCZ, defined with the maximum CIWP, in the west Pacific to middle sections (160E to 130 W) is found too north relative to observed location both in GISS-E2.1 and GISS-E3, coincided with modeled minimum RLUT, maximum RSUT but not the case for minimum RSDS (figures 3 and 4) and maximum SST. This result leads to a 'misleading' modeled ITCZ biases commonly seen in other studies, which lead to overestimated RLUT and RSDS and underestimated RSUT, when compared to the observed ITCZ location from CERES data.
In general, the radiation components are improved but limited to the south Pacific trade-wind regions between the ITCZ and SPCZ, with drastic improvements in CIWP, radiation and surface wind stress in GISS-E3 against GISS-E2.1 (figures 4(d), (h), (l)). Both RSUT and RSDS are significantly improved over stratocumulus/stratus cloud decks off the coasts of California and Peru, implying improved low cloud albedo simulated in GISS-E3 versus GISS-E2. Improvements in radiation components from GISS-E2.1 to GISS-E3 are expected to have an effect on enhancing surface wind stress (Li et al 2015(Li et al , 2022b in GISS-E3, where weaker surface winds with biases of westerly and meridional low-level divergences anomalies of TAU in GISS-E2.1 are reduced in GISS-E3 ( figure 5).
Overall, CIWP, TAU, all radiative fluxes, SST, PR and PRW are closer to the references in GISS-E3 relative to GISS-E2.1. We found that the improved CIWP, TAU biases from GISS-E2.1 to GISS-E3 are critical to alleviate long-standing biases of too warm SST (figures 6 and 7), excessive precipitation and PRW found in this study, but only limited in the region between ITCZ and SPCZ and off the coasts of California and Peru regions. The latter may be related to improved model vertical resolution and boundary-layer cloud parameterization.
Although the net effect of changes in GISS-E3 relative to GISS-E2.1 results in smaller discrepancies in the fields examined, the improvements are found to be limited to south Pacific Oceans. Non-trivial regional biases such as over north Pacific remain in GISS-E3 with generally too much CIWP, warm SST, shortwave reflectance, excessive PR/PRW over north of the equator, due to northward shifting of modeled ITCZ. This may be a higher priority in further improving GISS-E3 in the near future.
There are many physical processes that have been changed from GISS-E2.1 to GISS-E3 such as clouds, convection and radiation, including the addition of the radiative effects of convective and stratiform precipitating ice and liquid hydrometeors. Despite the fact that the magnitude of CIWP (same as TIWP in GISS-E2) is reduced to that of 2C-ICE estimates in GISS-E3, it remains overestimated in the trade wind regions in both hemispheres relative to CloudSat 2C-ICE estimates. Over the ITCZ/SPCZ and TWP regions, it is, however, underestimated. In addition to suspended cloud ice, CloudSat 2C-ICE falling ice potentially includes snow, graupel and hails while GISS-E3 represents suspend cloud ice and snow processes but no graupel and hails. Therefore, snow water content and its impact of FIREs on the present climate may be underestimated. However, there are uncertainties of CloudSat 2C-ICE IWP/IWC estimates in frozen hydrometeors over convective areas as the attenuation of radar signals might result in underestimate total frozen hydrometeor path.
GISS-E3 represents a huge advance in representing complete hydrometeors-radiation interactions by implementing more realistic prognostic radiatively active hydrometeors involved from stratiform + convective snow and stratiform + convective forms of rain prognostically. In particular, convective snow and stratiform + convective forms of rain are not included in our previous experiences (Li et al 2015(Li et al , 2016. We argue that including rain radiative effects might not have large impacts on the Pacific climates, but adding stratiform snow can have substantially changes the simulation of Pacific climate found by, for example, Li et al (2015Li et al ( , 2016Li et al ( , 2020bLi et al ( , 2022aLi et al ( , 2022b and Michibata et al (2019). It might be worth to separate the roles played by the stratiform and convective snow radiative effects and their impacts on the circulation-radiation coupling in GISS-E3. However, these are out of the scope of the present study.
In conclusion, we found numerous improvements in GISS-E3 against its previous GISS-E2.1 for the CMIP6 historical runs. GISS-E3 reduces discrepancies versus observation-based products across the Pacific. This includes substantial improvements in atmospheric ice content, radiative fluxes, surface wind stress, sea surface temperatures, precipitation and column total column water vapor but limited to the south Pacific regions.