Understanding the regional surface temperature responses
to different anthropogenic climate forcing agents, such as greenhouse gases
and aerosols, is crucial for understanding past and future regional climate
changes. In modern climate models, the regional temperature responses vary
greatly for all major forcing agents, but the causes of this variability are
poorly understood. Here, we analyze how changes in atmospheric and oceanic
energy fluxes due to perturbations in different anthropogenic climate
forcing agents lead to changes in global and regional surface temperatures.
We use climate model data on idealized perturbations in four major
anthropogenic climate forcing agents (CO
Climate change projections depend highly on future scenarios of climate mitigation actions. But in addition to uncertainty arising from different possible futures particularly in timescales of decades, the climate projection uncertainties are dominated by the climate model response uncertainty (Hawkins and Sutton, 2009; Lehner et al., 2020). This arises from structural differences between different climate models. Climate models differ in how they represent the radiative forcing of anthropogenic greenhouse gases and aerosols. But, perhaps more importantly, they respond differently to the same external radiative forcing (Nordling et al., 2019). As stated in Lehner et al. (2020), the model spread in the estimated temperature responses is affected by intermodel differences in both the forcing and in how the models respond to the forcing.
Smith et al. (2020) quantified the effective radiative forcings (ERFs) for
modern-day greenhouse gas and aerosol concentrations for a range of climate
models participating in the CMIP6 multimodel climate experiments. They showed
that since CMIP5, the spread in modeled radiative forcing has narrowed. Despite this, the response uncertainty in CMIP6 models appears to have
grown from CMIP5 models (Lehner et al., 2020; Zelinka et al., 2020).
Uncertainty in the climate response hampers efforts to robustly define
carbon emission targets to maintain global warming below specified limits,
such as below 1.5
Besides the need to better understand the impacts of different climate forcing agents on the global climate, there is an urgent need to better understand how they impact climate on a regional scale. The spatial distribution of aerosols is highly heterogenous, and much of the modern-day effective aerosol radiative forcing is concentrated over the South Asian and East Asian regions (Fiedler et al., 2019), while the radiative forcing of long-lived greenhouse gases is much more uniform (Shindell et al., 2015). Aerosols have both local and remote climate effects which depend on the emission region and type of aerosol (Merikanto et al., 2021; Nordling et al., 2019; Persad and Caldeira, 2018). Furthermore, the differences in aerosol surface temperature response between modern climate models are not dominated by the model's anthropogenic aerosol description (Nordling et al., 2019). Therefore, differences in modeled regional temperature responses for both greenhouse gases and aerosols appear to mainly depend on differences in dynamic responses of the atmosphere–ocean–sea-ice system in the models. The main focus of this paper is these differences in modeled responses to aerosol and greenhouse gas perturbations in different climate models.
The Precipitation Driver Response Model Intercomparison Project (PDRMIP)
(Myhre et al., 2017) provides a data set that allows us to investigate how
different climate forcing agents affect the Earth's climate on global and
regional scales. PDRMIP comprised idealized single-forcer scenarios for
several independent climate models. Previously, the PDRMIP data set has been
used to study, for example, how different forcing agents affect the Arctic
amplification (Stjern et al., 2019) and how they produce rapid adjustments
and ERF (Smith et al., 2018). Estimating ERF is not straightforward, and
different methods provide a variety of different results. For example, Tang
et al. (2019) used PDRMIP data to estimate ERF for different climate forcing
agents with several different methods. The model-mean estimated ERF for the
doubling of carbon dioxide concentrations varied from 3.65 to 4.70 W m
The model differences in climate response are often investigated through radiative feedback analysis (e.g., Zelinka et al., 2020). While the feedback analysis is particularly suitable for analyzing the root causes of model-to-model differences in the equilibrium climate sensitivity (the equilibrium temperature response to doubled atmospheric carbon dioxide concentrations), it is less suitable for exploring regional temperature response variance between the models due to the nonlinearity of regional feedbacks (Andrews et al., 2012). Räisänen and Ylhäisi (2015) formulated an energy balance framework to explore the impact of the top-of-atmosphere (TOA) radiative fluxes, atmospheric energy transport, and the net surface energy flux on regional surface temperatures. The method relies on the local conservation of energy and it is therefore mathematically an almost exact solution for the decomposition of energetic components of the temperature response. Its also takes into account both the horizontal energy transport and surface energy fluxes on the local energy balance. Räisänen (2017) included a more detailed shortwave radiative flux treatment according to Taylor et al. (2007), and Merikanto et al. (2021) included a cloud radiative kernel treatment for a more physical separation of longwave cloud and clear-sky radiative fluxes. In this paper, we use this energy balance framework with climate model data from PDRMIP experiments to study the origins of regional temperature response and its standard deviation in six different climate models to four different climate forcing agents (carbon dioxide, methane, sulfate, and black carbon). Evaluation of the mechanisms responsible for the model spread is key to understanding why models still exhibit a substantial spread in temperature response even when forced identically.
Illustration of the local atmospheric energy budget in a
single atmospheric column from the surface to the top of model atmosphere
(TOA). We attribute the change in local surface temperature to changes in
different terms of the local energy budget.
We attribute local surface air temperature response to different net
energetic components, namely, to changes in local longwave fluxes associated
with changes in clear-sky and cloud emissivity (
The rate of energy change within an atmospheric column is given by the
energy balance equation:
The change in
Using Eq. (2) with Eq. (5), the local change in surface temperature can be
decomposed to different energetic components as
Also
First the left-hand side in Eq. (8) is obtained by substituting the all-sky
LW flux to Eq. (2). Second, the first (clear-sky) right-hand side term in Eq. (8) is
obtained by substituting the clear-sky flux into Eq. (2). The CRE component is
obtained as a residual. However,
where
All results have been calculated using three different kernels, ECHAM (Block and Mauritsen, 2013), GFDL (Pendergrass et al., 2018), and HadGEM2 (Smith, 2018), to obtain a better estimate of the overall cloud effect. The correction factor of Eq. (9) has been calculated as an average of the three kernels.
Finally, the local surface temperature responses are decomposed as
In the above equation, the temperature responses related to the first five
components build up from a sum of the instant radiative forcing (if any),
rapid adjustments associated with the component, and a temperature-dependent
feedback which adjusts its magnitude as the surface temperature changes,
normalized by
Decomposing the temperature responses
We use climate model data from (PDRMIP) (Myhre et al., 2017). In PDRMIP,
several independent climate models were used to simulate various idealized
climate perturbations. The models used in this study are listed in Table 1.
According to Knutti (2013), all these models belong to different model
families and hence are largely independent of each other. Our study uses
data from experiments of instant doubling of CO
PDRMIP models used in this study, ocean and aerosol configuration of the model, and which aerosol–cloud interactions are included.
Description of PDRMIP experiments.
All simulations consisted of 100-year baseline and perturbed runs, and the
last 50 years of these runs are used for the temperature response analysis
carried out here. The PDRMIP experiments also included additional fixed
sea surface temperature runs, which we use for the calculation of the
effective radiative forcing (ERF
In this paper, we focus on decomposed local and global temperature responses
normalized by the global effective radiative forcing (ERF
Figure 2 shows the calculated effective radiative forcings and the global
mean temperature responses (the difference in perturbed climate for the
years 50–100 and the corresponding years from the base case) in the analyzed
PDRMIP experiments. The effective radiative forcing is calculated from both
fixed-sea-surface-temperature simulations (ERF
The global average temperature responses for each
experiment and each model (
One of the models (NCAR-CESM1-CAM) was ran using a slab ocean configuration,
while the rest of the models contained fully interactive ocean
configurations. Since the equilibrium is reached in a few decades with slab
ocean configurations but for the fully interactive ocean configuration it
takes centuries, the perturbed experiments with models besides
NCAR-CESM1-CAM are still in a transient state. As a multimodel mean over
the years 50–100 of the perturbed runs, the doubling of CO
The multimodel-mean ERF
Figure 2 shows that only a weak relationship between the model-to-model values
in ERF
In the following sections, we present decomposed effective temperature
responses for each analyzed experiment and model-to-model spread of these
decompositions. The effective surface temperature responses and their
decompositions are calculated for each atmospheric column separately from
the average differences in perturbed climates for the years 50–100 after a
sudden perturbation and the corresponding years from the baseline
simulations without perturbations. The local temperature responses are
normalized by the globally averaged ERF
The local temperature responses related to longwave and shortwave TOA components build up from a combination of the local instantaneous top-of-atmosphere radiative forcing and rapid adjustments associated with each term, as well as a temperature-dependent feedback which adjusts its magnitude as the surface temperature changes, as described in the end of Sect. 2.1. Therefore, temperature responses related to these components are functions of a forcing (if any), rapid adjustments, and a time-dependent term (the impact of feedback as surface temperature changes).
The temperature response decomposition applied here relies on a local conservation of energy in each atmospheric column; hence, the sums of individual temperature response components generate the local total surface temperature responses with high accuracy. Below, Sect. 3.1 presents the globally averaged results. Section 3.2 then presents the regional distributions of the decomposed surface temperature responses and their zonal averages. Section 3.3 presents the regional and latitudinal distributions of the model-to-model standard deviations of the effective temperature components and the contributions of each of the decomposed surface temperature response components to the total standard deviations of the responses.
Figure 3 shows the globally averaged effective surface temperature responses
and their decomposed components for each model and perturbation experiment,
calculated by using the temperature decomposition method described in
Sect. 2.1. The components of the effective surface temperature responses
describe the combined global contributions of the TOA forcing (in the case of
clear-sky
The global mean effective temperature response and its
decomposition, calculated as the difference between means over the last 50 years of the perturbed and the baseline experiments.
The total effective temperature responses (temperature response divided by
the ERF
The differences in effective temperature responses associated with
The multimodel-mean effective temperature responses related to
The global effective temperature response from the changes in surface albedo
is similar across each experiment. The mean effective temperature response
due to albedo change varies from
The model-mean spatial distributions of effective temperature responses and
their decomposed components are shown in Fig. 4. The zonal means of
different components are shown in Fig. 5, where we have summed up the
contributions of surface and atmospheric energy transport components
(
The multimodel-mean effective temperature response (row 1) for four different climate forcers, i.e., carbon dioxide (column 1),
methane (column 2), sulfate (column 3), and black carbon (column 4), and its
decomposition into different energy balance terms (long- and shortwave
clear-sky (
The spatial distribution of the total effective temperature response is
largely similar for each forcer, although the total response to aerosols is
stronger over the continental northern midlatitudes, compared to total
responses to greenhouse gases, and weaker over the Southern Hemisphere
oceans. Regionally, local maximum effective temperature responses are found
in the Barents Sea for all forcers, with maximum values of 2.38, 2.04, 2.96, and 2.53 K W
For greenhouse gases, the regional effective temperature responses are
mostly associated with the response to
For aerosols, most of the effective temperature response is due to
Zonal-average multimodel-mean effective temperature
response (thick blue lines) and its decomposition into different energetic
terms (thin colored lines) for different climate forcers. Panel
There is significant variation in the regional effective temperature
contributions due to clouds between regions and forcing agents. In the
greenhouse gas experiments (co2x2 and ch4x3), the regional effective
temperature responses due to
With aerosols, the net effect of clouds is more complicated. Both the sulx5
and bcx10 experiments show a similar negative effective temperature response
due to
The effective temperature response to surface albedo change originates from
the change in sea-ice and snow cover and is always positive. Changes in
surface albedo have a modest effect on the global effective temperature
response with all forcers (0.07, 0.06, 0.08, and 0.06 K W
Over the oceans,
Similarly to the effective temperature response itself, also its model-to-model spread (standard deviation) can be decomposed into components that sum up to the total spread in the effective surface temperature response (Sect. 2.2). Figure 6 shows the decomposed model-to-model standard deviations of the total effective temperature responses (first row) for each perturbation experiment and the decomposed contributions of each component to the spread in total responses. The latitudinal distributions of the different components are shown in Fig. 7.
The model-to-model standard deviation of the effective temperature response to different climate perturbations (row 1) and its decomposed different energetic components (rows 2–8). Each column shows results for four different climate forcers, i.e., carbon dioxide (column 1), methane (column 2), sulfate (column 3) and black carbon (column 4). The global mean values are shown at the bottom right corner of each panel.
Zonal mean of the total standard deviation of the
effective temperature response (thick blue line) and the contributions of
the different energy balance terms to it (thin lines; see the legend in
The globally averaged magnitude of the model-to-model spread is similar
between co2x2, ch4x3, and sulx5 experiments (0.19, 0.18, and 0.18 K W
In the aerosol experiments (sulx5 and bcx10) the build-up of the
model-to-model spread is more complicated than for the greenhouse gas
experiments, despite similarities in the latitudinal distribution of the
total spread of the effective temperature response. The contributions of
On the other hand, in the aerosol experiments (sulx5 and bcx10) the
atmospheric heat transport (
Previously, the model-to-model spread in global climate sensitivity
(equilibrium response to doubled CO
In this work, we have conducted an energy balance decomposition of the
near-surface temperature response resulting from doubling CO
The temperature decomposition method provides a tool for understanding
regional and global temperature changes. However, the original method is
somewhat simplistic in its treatment of LW cloud processes
(Räisänen, 2017). In Merikanto et al. (2021), we implemented a
radiative kernel correction to make the LW treatment of clouds more
realistic. However, despite this correction, we still have a negative
effective LW temperature response from clouds when the CO
In our study, clouds play a minor role in the global mean temperature response, as the LW cloud and SW cloud terms tend to cancel each other out. However, regionally the temperature response originating from the clouds is a significant contributor. For all forcers, the temperature response in the Antarctic sea-ice region and in the Southern Ocean is dampened by clouds. With BC, clouds dampen the regional temperature response in Asia, North America, Africa, and Europe and enhance the warming in the Amazon. In contrast to clouds, with all forcing agents surface albedo changes enhance the temperature responses in high latitudes. For greenhouse gases, the mild polar amplification in the south is associated with a negative contribution from the ocean heat exchange over the Southern Ocean, negative total cloud contribution and a mild LW clear-sky component.
We also decompose the model-to-model spread into the contributions of energy
balance terms. The model-to-model spread is the largest in the same regions
as the average temperature response, i.e., at high latitudes, where the
spread is driven by differences in the lapse-rate and water vapor feedbacks
(
The aerosol configuration is important in the generation of the effective
temperature response and its model-to-model spread. In the aerosol
experiments, part of the model-to-model spread originates from the
difference between aerosol setups, with the emission-driven models
generating a higher effective temperature response than the
concentration-driven models. For sulx5, the concentration-driven models'
mean effective temperature response is 0.49 K W
We have demonstrated that the mechanisms behind model uncertainty vary between different regions and forcing agents. Understanding the atmosphere's dynamical response to different forcers is key to understanding future climate changes at the regional level. This is especially important in the case of aerosols, which are predicted to decline in the near future due to climate change and air pollution mitigation actions.
Data and scripts used for data analysis can be obtained by contacting
the corresponding author. The temperature decomposition script is available from
The supplement related to this article is available online at:
The article was written by KN and JM, with contributions from all authors. KN and JM performed the analysis with the help of JR. BHS provided the PDRMIP data.
The contact author has declared that neither they nor their co-authors have any competing interests.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We gratefully acknowledge the efforts of the PDRMIP community and the modelers who have kindly made their simulation results publicly available. Storage and availability of PDRMIP data were provided by UNINETT Sigma2 – the National Infrastructure for High Performance Computing and Data Storage in Norway. Bjørn H. Samset acknowledges funding by the Research Council of Norway (project no. 244141, NetBC). Jouni Räisänen acknowledges funding by the Academy of Finland Flagship funding (grant. no 337549). Kalle Nordling acknowledges funding by the Academy of Finland (grant no. 340791). The authors would like to thank the and two reviewers for reviewing this paper.
This research has been supported by the European Research Council, H2020 European Research Council (grant no. ECLAIR (646857)), and the Academy of Finland (grant nos. 287440, 308365, and 331764).
This paper was edited by Tanja Schuck and reviewed by William Collins and one anonymous referee.