1. Introduction
It is well established that human activity is influencing many aspects of the climate system (Eyring et al. 2021). A common approach to the identification of human-caused signals is to apply so-called fingerprint methods (Hasselmann 1979, 1993, 1997). These methods compare the spatial and/or temporal patterns of change in observations to the modeled patterns shown in response to individual climate forcings. Fingerprint methods have been used to detect human influence on increasing temperatures in the troposphere (Santer et al. 1996; Tett et al. 1996; Santer et al. 2013; Lott et al. 2013), at the surface (Hegerl et al. 1996; Barnett et al. 1999; Gillett et al. 2013, 2021), and in the ocean (Barnett et al. 2005; Gleckler et al. 2012; Bilbao et al. 2019). In this previous “single variable” work, each variable is typically considered individually. Relatively few studies examine whether detection of a multivariate fingerprint—in which different climate variables are combined—yields earlier detection of human influence than in “single-variable” cases (Santer et al. 1995; Pierce et al. 2012; Bonfils et al. 2020).
Because “single-variable” studies often cover different timespans of the observational record and use different fingerprint methods (or make different choices in the application of the methods), it is difficult to quantitatively compare results across multiple variables. This is particularly problematic in studies seeking to quantify the influence of anthropogenic aerosols on the climate system. For example, one investigation analyzing troposphere temperature from radiosonde data found that the Canadian Earth System Model version 2 (CanESM2) underestimates the magnitude of the aerosol fingerprint (Lott et al. 2013), while separate studies analyzing surface temperature (Gillett et al. 2013) and ocean heat content (Swart et al. 2018) suggest that the same model overestimates the response to aerosol forcing. It is not clear if these discrepancies arise from the different methods or timespans covered, or if there is a physically plausible reason (e.g., model errors that are primarily manifest in one variable) for this difference in the estimated strength of the aerosol fingerprint.
Narrowing uncertainty in aerosol forcing is important for improving understanding of past and future climate change. Cooling caused by anthropogenic aerosol forcing has partially offset greenhouse gas (GHG) warming since the preindustrial period, but to what extent is still uncertain (Eyring et al. 2021). Because the aerosol-induced cooling and GHG-driven warming show similar patterns, but with opposite sign, constraining each individually is challenging (Allen et al. 2006). Another challenge is that it is difficult to separate the response to low-frequency changes in aerosol forcing from multidecadal internal variability (Mann and Emanuel 2006; Bellomo et al. 2018; Santer et al. 2022). The uncertainty in the total warming caused by all anthropogenic forcing since preindustrial times is smaller than the uncertainty in the individual temperature changes caused by aerosol and GHG forcing (Gillett et al. 2013; Ribes et al. 2021). A stronger aerosol cooling would result in a larger role for GHG warming in observed trends and would imply stronger projected warming over the twenty-first century.
Here we apply the same regression-based fingerprint method to three different variables: the temperature of the mid-to-upper troposphere (TMT), surface temperature (ST), and upper ocean heat content from 0 to 700 m (OHC). We examine changes in TMT, ST, and OHC over a common analysis period. Our fingerprint study relies on observational data and on simulation output from the CanESM2 model large ensemble, which has been used extensively in detection and attribution investigations (Swart et al. 2018; Kirchmeier-Young et al. 2016; Santer et al. 2019, 2022) and has resulted in inconsistent estimates for the role of anthropogenic aerosol forcing (Gillett et al. 2013; Lott et al. 2013; Swart et al. 2018).
One of the main goals of this study is to determine if applying the same detection and attribution method across these three different variables leads to more consistent results for the estimated strength of the anthropogenic aerosol fingerprint. After first analyzing the temporal and spatial patterns of the model-predicted TMT, ST, and OHC fingerprints, we then check to see if we obtain quantitatively consistent estimates of the strength of the aerosol and GHG components if the same fingerprint method is applied to independent observations of TMT, ST, and OHC. We then combine the data from these three variables in a single multivariate fingerprint. Finally, we calculate detection times for each individual variable and for the multivariate fingerprint and consider whether use of a multivariate improves fingerprint detection time.
2. Methods and data
a. Observations
We investigate observations of TMT, ST, and OHC over the 1979–2018 period. For each variable, we focus primarily on one observational dataset, but we also examine additional datasets to check for sensitivity to observational uncertainty. For satellite measurements of TMT, we focus on data from Remote Sensing Systems (RSS) version 4.0 (Mears and Wentz 2017) but also examine data from the Center for Satellite Applications and Research (STAR) version 4.1 (Zou et al. 2018) and the University of Alabama at Huntsville (UAH) version 6.0 (Spencer et al. 2017). TMT is corrected using standard regression-based methods (Fu and Johanson 2005; Santer et al. 2019) to remove the contribution of the cooling of the lower stratosphere (TLS). Specifically, we use TMTc = a24TMT + (1 − a24)TLS, where a24 = 1.1 and TMTc represents the corrected TMT data.
For ST, we rely mainly on gridded, monthly data from the Met Office Hadley Centre/Climatic Research Unit global surface temperature dataset (HadCRUT5.0.0.0) (Morice et al. 2021), but we also examine data from NOAAGlobalTemp v5.0.0 (Huang et al. 2020), Berkeley Earth (Rohde and Hausfather 2020), and NASA GISTEMPv4 (Lenssen et al. 2019).
b. Model experiments and data
The model simulations examined here are from multiple 50-member large ensembles performed with CanESM2 (Arora et al. 2011; Kirchmeier-Young et al. 2016; Swart et al. 2018). This is the model version that participated in phase 5 of the Coupled Model Intercomparison Project (CMIP5). Although a newer model version exists (CanESM5), we use the older CanESM2 model version here because previous detection and attribution work has obtained inconsistent results in studies relying on CanESM2 output (Gillett et al. 2013; Lott et al. 2013; Swart et al. 2018). We seek to understand and reconcile these inconsistent detection and attribution results.
The first CanESM2 ensemble includes all major anthropogenic and natural external forcings (henceforth denoted ALL). We also use three single forcing ensembles that consist of forcing by natural factors only (NAT), anthropogenic aerosols (AER), and stratospheric ozone (OZ). All experiments relied on estimated historical forcing from 1950 to 2005 and the representative concentration pathway 8.5 (RCP8.5) forcing from 2006 to 2100. To compare with observations, we only examine the 1979–2018 period here. We use the RCP8.5 scenario because this is the only scenario for which the large ensembles were available. It is unlikely that the choice of scenario has an impact on our detection and attribution results because the differences in greenhouse gas and aerosol forcing between RCP8.5 and other commonly analyzed RCPs are relatively small over the time period analyzed here (Meinshausen et al. 2011; Lamarque et al. 2011; van Vuuren et al. 2011).
Each of the four experiments (ALL, NAT, AER, and OZ) has 50 realizations that differ in their initial conditions. The only difference between the individual realizations is in the internal variability that is superimposed on the underlying response to external forcing. Since there is no “GHG only” ensemble, we estimate the response to GHG forcing by differencing the ALL forcing experiment from the sum of the other three experiments.
To facilitate “like-with-like” comparisons between the models and observations, we calculate synthetic satellite temperatures from the models (Santer et al. 2013). For model-versus-observed ST comparisons, we blend the model sea surface temperature (SST) and surface air temperature (SAT) as outlined by Cowtan et al. (2015).
c. Detection and attribution
To estimate uncertainty in the scaling factors, we use the between-realization variability calculated from the four 50-member large ensembles. For each ensemble (ALL, NAT, AER, and OZ) we calculate and remove the ensemble mean from each individual realization of that ensemble. This results in a total of 200 individual ensemble members. Each member provides a sample of the “noise” arising from natural internal variability. Each realization is scaled by
We then use the same regression model given above, but we replace the observations with each of the 200 realizations of internal variability. This results in a set of 200 scaling factors in which the correspondence between the searched-for fingerprint and Y is solely due to internal variability. The uncertainty on the scaling factors is the 5th–95th percentiles of the 200-member range of values; this is centered on the corresponding scaling factors obtained with the observations. The underlying assumption in estimating uncertainty in this way is that external forcing has little effect on internal variability. Previous work with tropospheric temperature in the CanESM2 large ensemble found that the between-realization variability and the internal variability calculated from the CanESM2 preindustrial control simulations yielded very similar uncertainty estimates (Santer et al. 2019).
A scaling factor that is statistically greater than zero at the 5% level is considered detectable in observations. If the uncertainty range of the scaling factor is consistent with one, the magnitude of the model fingerprint pattern is consistent with its magnitude in observations. A scaling factor greater than one indicates that the fingerprint must be scaled up for the best fit to observations (i.e., the magnitude of the model fingerprint is weaker than in observations), and a scaling factor less than one means that magnitude of the fingerprint has to be scaled down.
We perform a time of detection analysis where we calculate when the fingerprints become detectable above background variability. This was done by repeating the regression using an initial 10-yr analysis period, and then repeating this with analysis periods that increase in 5-yr increments (i.e., the first 10, 15, 20 years, etc.). The time of detection is defined as the first period when the scaling factor becomes (and remains) statistically greater than zero.
Finally, the troposphere, surface, and ocean data were combined into a single multivariate fingerprint by normalizing each variable by the internal variability determined by the between realization spread (Pierce et al. 2012; Swart et al. 2018). The three variables were then concatenated, resulting in a single vector that is 3 times the size of each individual variable’s spatiotemporal vector.
3. Results
a. Time series and spatial patterns
We begin by comparing the time series of global-mean TMT, ST, and OHC over a common time period (1979–2018) in observations and model simulations (Fig. 1). In observations, there is a clear and approximately linear increase in all three variables over the 40-yr analysis period (Fig. 1a). The ALL forcing experiment shows similar behavior (Fig. 1b), but with smoother time series because averaging over individual realizations substantially reduces the size of internal variability. Relative to observations, the magnitude of the increase in TMT, ST, and OHC is stronger in ALL, consistent with previous findings that the CanESM2 model overestimates observed warming over the historical period (Gillett et al. 2013; Santer et al. 2019).
Time series of annual-mean global-mean changes in TMT (orange), ST (red), and OHC (blue). Results are for (a) observations and for model simulations with (b) ALL forcing, (c) GHG forcing, (d) AER forcing, (e) OZ forcing, and (f) NAT forcing. The time series are plotted as anomalies with respect to the 1979–2018 mean. The left axis of each panel represents the temperature anomalies for TMT and ST, and the right axis displays units for the OHC anomalies. Note the different scales on the different panels.
Citation: Journal of Climate 36, 22; 10.1175/JCLI-D-23-0068.1
The single forcing experiments are characterized by strong warming in the GHG experiments (Fig. 1c), weak cooling in the OZ experiments (Fig. 1e), and clear short-term cooling signals after the El Chichón and Mt Pinatubo eruptions in the ALL and NAT simulations (Figs. 1b,f). In all experiments except AER, the global-mean changes in TMT, ST, and OHC show consistent time evolution. For AER, however, the time evolution of global-mean OHC is noticeably different from that of TMT and ST (Fig. 1d). For TMT and ST, there is a weak cooling tendency that reverses in the 1990s and becomes a weak warming trend. This reversal coincides with the aerosol forcing peaking at this time (Bonfils et al. 2020; Deser et al. 2020). The OHC shows a stronger initial cooling trend, which weakens but does not reverse. These differences between the OHC signal and the TMT and ST signals reflect the longer time scale of the ocean’s adjustment to the decrease in anthropogenic aerosol forcing.
We next examine the spatial patterns of the trends over 1979–2018 in observations and in the model ALL simulations (Fig. 2). In all three variables, both the observations and ALL simulations show widespread warming throughout the troposphere, surface, and ocean. The simulated warming is generally larger than observed. Many of the large-scale features of the spatial patterns are similar in the observations and models. For TMT, there is a common hemispheric asymmetry in model and observed warming patterns, with larger warming over the Northern Hemisphere compared to the Southern Hemisphere (Figs. 2a,d). The ST trends display pronounced Arctic amplification and stronger warming over land than ocean, and weaker warming over the Southern Ocean (Figs. 2b,e). In OHC, observations and models reveal larger warming over most of the North Atlantic and Southern Ocean (Figs. 2c,f). The similarity in the patterns between observations and models for each variable can be more easily seen when the global-mean trend is removed (Figs. 2g–l).
Linear trends over 1979–2018 in annual-mean (a),(d) TMT, (b),(e) ST, and (c),(f) OHC. Trends are shown for (a)–(c) observations and model (d)–(f) ALL forcing experiments. (g)–(l) As in (a)–(f), but with the global-mean trend removed.
Citation: Journal of Climate 36, 22; 10.1175/JCLI-D-23-0068.1
One of the most noticeable differences between simulated and observed OHC trends is in the eastern Pacific, where observations indicate cooling of the upper ocean over 1979–2018. The observed cooling is likely due to the phasing of multidecadal Pacific internal variability during this specific period (Watanabe et al. 2021; Po-Chedley et al. 2021). While individual model ensemble members can (by chance) capture similar phasing of the single realization of observed internal variability, the ALL ensemble average (which is shown in Figs. 2d–f) damps internal variability and more clearly reveals the underlying warming response to anthropogenic forcing. Nevertheless, the model does show weaker warming over the eastern Pacific relative to the global mean. This can be seen more clearly when the global-mean trends are removed (Figs. 2k,l).
The spatial patterns of TMT, ST, and OHC trends due to anthropogenic GHG forcing (Figs. 3a–c) are similar to trend patterns in the ALL experiment (Figs. 2d–f), indicating that GHG forcing is the primary driver of the ALL trends. In response to changes in anthropogenic aerosol emissions over 1979–2018, both TMT and ST yield Arctic warming and negligible changes outside the Arctic (Figs. 3d,e). This is consistent with previous findings that a reduction in aerosol emissions since 1990 over Europe has contributed to Arctic surface warming (Acosta Navarro et al. 2016). The OHC response to aerosols differs from the temperature response at the surface and in the troposphere, with weak cooling throughout most ocean basins (Fig. 3f). There is minimal response to stratospheric ozone forcings (Figs. 3g–i). Natural forcing produces weak warming (Figs. 3j–l) due to the recovery from the volcanically induced cooling in the first half of the 40-yr time period considered.
Linear trends over 1979–2018 in annual-mean (a),(d),(g),(j) TMT, (b),(e),(h),(k) ST, and (c),(f),(i),(l) OHC. Trends are shown for the model (a)–(c) GHG forcing, (d)–(f) AER, (g)–(i) OZ, and (j)–(l) NAT experiments.
Citation: Journal of Climate 36, 22; 10.1175/JCLI-D-23-0068.1
b. Pattern-based detection and attribution
Figure 4 shows the scaling factors obtained using the model and observed datasets in Figs. 1 and 2. We begin with a one-signal analysis where the observations are regressed onto the ALL fingerprint. The fingerprint is clearly detectable in each of the three temperature variables, with positive scaling factors and scaling factor uncertainties that do not overlap with zero. All three scaling factors (for TMT, ST, and OHC) are smaller than one and have uncertainty ranges that do not overlap with one. This indicates that the magnitude of simulated warming is overestimated relative to observations, consistent with previous regression analysis of global-mean surface temperature with this model (Bindoff et al. 2014).
Scaling factors for TMT (orange), ST (red), OHC (blue), and combined (black) fingerprints calculated from a one-signal (ALL) and four-signal (GHG, AER, OZ, NAT) detection and attribution analysis. The error bars represent the 5%–95% uncertainty range in the scaling factor.
Citation: Journal of Climate 36, 22; 10.1175/JCLI-D-23-0068.1
We also performed a four-signal analysis with the GHG, AER, OZ, and NAT fingerprints. The strength of the GHG fingerprint is similar to that obtained with the ALL fingerprint in the one-signal analysis, with scaling factors of approximately 0.5 for each of the three variables considered. This indicates that the model GHG fingerprint needs to be scaled down by a factor of 2 to best fit the observations. It also suggests that the ALL fingerprint results are dominated by the GHG component of ALL, as was visually obvious from Figs. 2 and 3. The GHG scaling factors show greater consistency across the three variables than the ALL scaling factors; for GHG, the scaling factors for TMT, ST, and OHC are all within the uncertainty range of each other. This was not the case for ALL.
The AER fingerprint is detectable in TMT and ST with the scaling factors significantly above zero. The uncertainty ranges of the TMT and ST AER fingerprints encompass one, indicating that the strength of the AER fingerprint is consistent in observations and the model. In contrast, the OHC scaling factor for aerosols is lower than one and is not detectable (the lower bound of the uncertainty range overlaps with zero). There is still some disagreement between the three variables, similar to what has been shown in detection studies that did not compare results across variables using the same fingerprint method and a common analysis period (Lott et al. 2013; Gillett et al. 2021; Swart et al. 2018). We note, however, that we find smaller disagreement here for the scaling factors of the AER fingerprint: the TMT and ST scaling factors are within the uncertainty range of the OHC scaling factor.
The stratospheric ozone fingerprint is not detectable in TMT and ST but is identifiable in OHC. This is likely due to changes in OHC in the Southern Ocean that have been previously detected and attributed to changes in stratospheric ozone (Swart et al. 2018). The NAT fingerprint is not detectable in TMT but is detectable in ST and OHC with scaling factors less than one. This result implies that the model may overestimate the magnitude of the response to natural (volcanic and solar) forcing. An alternative interpretation of the NAT scaling factors is that in the real world, the warming caused by El Niño events partly obscured the volcanically induced cooling signal associated with the eruptions of El Chichón in 1982 and Pinatubo in 1991—a problem that does not impact the ensemble averages used to estimate the NAT fingerprint (Santer et al. 2014).
Previous work has shown that combining physically connected variables into a single fingerprint can help to reduce uncertainty in detection and attribution studies and may yield earlier fingerprint detection times (Santer et al. 1995; Pierce et al. 2012; Bonfils et al. 2020). We combine the TMT, ST and OHC information into one multivariate fingerprint by normalizing each variable by its internal variability. The scaling factors for the combined fingerprint are denoted by black in Fig. 4. They yield similar values to the scaling factors obtained in the individual analysis of the three variables. Importantly, the uncertainty in the multivariate fingerprint is reduced relative to the “single-variable” fingerprints. For the GHG fingerprint, the reduction is small, with the scaling factor uncertainty in the multivariate case only about 10%–35% lower than the “single-variable” scaling factor uncertainties. For the AER fingerprint, however, combining the three variables leads to a substantially larger reduction in the aerosol scaling factor uncertainty, with a 40%–50% reduction compared to the “single-variable” cases.
We next examine the sensitivity of the scaling factors to the choice of observational dataset for each of the three variables (Fig. 5). We seek to determine whether fingerprint detection and attribution (i.e., whether a scaling factor is consistent with one) is sensitive to the choice of observations. For TMT (Fig. 5a), there is good agreement in scaling factor values across the three different observational datasets, but the ALL and GHG scaling factors are higher in STAR version 4.1 and lower for UAH. These differences are in accord with the larger warming in STAR and weaker warming in UAH (Santer et al. 2019). The key point is that the ALL, GHG, and AER fingerprints are robustly detectable in all three satellite TMT datasets, but only the AER fingerprint has scaling factors consistent with one. Similarly, the OZ and NAT fingerprints are not detectable in any satellite dataset.
Scaling factors calculated for each of the different observational datasets for (a) TMT, (b) ST, and (c) OHC. The error bars represent the 5%–95% uncertainty range in the scaling factor.
Citation: Journal of Climate 36, 22; 10.1175/JCLI-D-23-0068.1
The scaling factors obtained with the four different observational ST datasets yield similarly close agreement. This holds for each of the five fingerprints (Fig. 5b). In the case of OHC (Fig. 5c), scaling factors also show close correspondence across different observations. The sole exception is the OZ fingerprint, which is detectable in the two EN4 datasets, but not in the NCEI/NOAA dataset. The individual datasets we chose to focus on throughout this study (RSS, HadCRUT5, and EN4 L09) are in the middle of the range seen across all available datasets.
c. Time of detection analysis
We next examine the time of detection of the GHG and AER fingerprints. This analysis was performed separately for the individual TMT, ST, and OHC fingerprints, as well as for their combined multivariate fingerprint. Figure 6a shows the scaling factors for the GHG fingerprint as a function of end year. The TMT, ST, and combined fingerprint all become detectable above background internal variability in 1998, after 20 years of monitoring. The OHC fingerprint becomes detectable 5 years later. For the AER fingerprint (Fig. 6c), the combined fingerprint becomes detectable in 1998 (i.e., after only 20 years), which is much earlier than in the AER “single-fingerprint” cases for TMT (35 years), ST (30 years), or OHC (not detectable after 40 years).
(a),(c) Scaling factors as a function of end year of the analysis period for TMT (orange), ST (red), OHC (blue), and combined (black) fingerprints calculated for the (a) GHG fingerprint and (c) AER fingerprint. (b),(d) Scaling factor uncertainty as a function of end year. The uncertainty represents the 5%–95% uncertainty range in the scaling factor. The colored crosses in (a) and (c) represent the detection times for the individual fingerprints and for the multivariate fingerprint.
Citation: Journal of Climate 36, 22; 10.1175/JCLI-D-23-0068.1
The differences in the scaling factor uncertainties due to internal variability can be more easily seen in Figs. 6b and 6d. For GHG (Fig. 6b), the largest uncertainties are for the TMT fingerprint, particularly for shorter time scales (<20 years). The GHG fingerprint for the multivariate case has the smallest uncertainties, but they are only marginally smaller than for any of the individual variables.
The AER fingerprint yields different results. The three individual variables show similar uncertainties in the behavior of scaling factors with increasing analysis period, with the exception of the slightly larger uncertainties in OHC at longer time scales. In the case of the combined AER fingerprint, however, the scaling factor uncertainty is substantially lower than in any of the three individual variables (40%–60% smaller, depending on time span and variable).
We next investigate the range of time of detection in the ALL model ensemble and how it compares to the observed detection time. This is done by repeating the time of detection calculation, but replacing the observations with each of the 50 ensemble members. The histograms of the time of the detection for the GHG and AER fingerprints are shown in Fig. 7. The GHG fingerprint is detectable in most ensemble members of the model between 10 and 20 years after 1979 (i.e., between 1988 and 1998). This holds across all individual variables and for the combined fingerprint. For TMT, the fingerprint detection times in both the model ensemble and observations are similar to those found previously using a different detection method (Santer et al. 2019). The distribution of detection times for the combined GHG fingerprint is only shifted slightly earlier relative to the “single-variable” cases, consistent with the slight reduction in scaling factor uncertainty described above. The observed detection times are within the ranges simulated by the model, but are on average later, consistent with scaling factors that are less than one for the GHG fingerprint (see Fig. 4).
Time of detection of the (a),(c),(e),(g) GHG and (b),(d),(f),(h) AER fingerprints for (a),(b) TMT, (c),(d) ST, (e),(f) OHC, and (g),(h) the combined multivariable case. The vertical purple lines indicate the time of detection in the three observational datasets of primary interest and the histograms represent the distribution of detection times in the 50 individual realizations of the ALL model experiment. If the fingerprint is not detectable before the end of the 1979–2018 period, the time of detection is listed as N/A.
Citation: Journal of Climate 36, 22; 10.1175/JCLI-D-23-0068.1
For the three individual variables, the range of the time of detection for the AER fingerprint is much larger than for the GHG fingerprint. In all variables, some ensemble members are detectable after only 10 years, while others are not detectable after 40 years. Ocean heat content appears to display a bimodal distribution, with 35 ensemble members detectable prior to 2008, but 15 of the ensemble members are not detectable by 2018. For the combined AER fingerprint, there is a clear shift to earlier detection times relative to detection times for the “single-variable” fingerprints, with most ensemble members detectable by 1998 or earlier (i.e., within 20 years). The AER multivariate fingerprint is detectable in all individual AER ensemble members by 2013 (within 35 years).
d. Evaluation of the model internal variability
An important assumption in the methods used here is that the model estimates of internal variability are consistent with observations (Allen and Tett 1999). In particular, it is important that the model does not systematically underestimate the magnitude of internal variability relative to observations. To test this, we first calculate the residual between the best fit from the four-signal analysis and the observations for each of the three variables. This results in anomalies of 5-yr averages, which should represent internal variability in observations. We then compare these observed residuals to the range of anomalies from internal variability from CanESM2. As was done for estimating the scaling factor uncertainty in the previous analysis, the model internal variability is calculated by subtracting the respective ensemble means from each realization. For all three variables, each time step of the global mean of the observed residuals lies within the model ensemble spread from internal variability (Figs. 8a,c,e). In addition, the standard deviations of the eight timesteps of the global-mean observed residuals are well within the distribution of standard deviations from the 200 model ensemble members (Figs. 8b,d,f). These results suggest that for global-mean TMT, ST, and OHC, the internal variability in CanESM2 is consistent with observed internal variability.
(a),(c),(e) Time series of the global mean of the observed residuals (black line) and the 2.5%–97.5% range of the global-mean anomalies of the model internal variability (shading) for (a) TMT, (c) ST, and (e) OHC. Each point represents a 5-yr average identified by the end year of the 5-yr average. (b),(d),(f) The standard deviation of the observed global-mean residuals (black line) for (b) TMT, (d) ST, and (e) OHC. The histograms show the standard deviations of global-mean anomalies for the 200 realizations of internal variability from CanESM2.
Citation: Journal of Climate 36, 22; 10.1175/JCLI-D-23-0068.1
In addition to the global-mean analysis, we also compared simulated and observed spatial patterns of internal variability. To check whether there are regions where the model underestimates observed internal variability, Fig. 9 shows the spatial patterns of the standard deviations of the observed residuals. The stippling indicates where the observed standard deviations are greater than the standard deviations in 95% of the 200 realizations of internal variability estimated from the model. There are only a few small regions where the magnitude of the observed variability is greater than most ensemble members. Furthermore, previous analysis has found no evidence that the CanESM2 model systematically underestimates the magnitude of internal modes of variability on multidecadal time scales (Santer et al. 2022). Overall, we find no evidence that the model used for our detection and attribution study is systematically underestimating the internal variability in any of the three variables investigated in this study, either at the global-mean level or in terms of spatial patterns.
The standard deviation of the observed residuals for (a) TMT, (b) ST, and (c) OHC. The × symbols indicate where the observed standard deviation is greater than 95% of the standard deviations from the 200 realizations of internal variability estimated from CanESM2.
Citation: Journal of Climate 36, 22; 10.1175/JCLI-D-23-0068.1
4. Discussion
When the same regression-based detection and attribution method is applied to temperature changes estimated separately for the troposphere, surface, and ocean, we obtain consistent scaling factors for the magnitude of a model-predicted greenhouse gas fingerprint in observations. These observations of different parts of the climate system are completely independent. The internal consistency of results from independently monitored TMT, ST, and OHC increases our confidence in the identification of a human-caused GHG fingerprint in observations. For the role of anthropogenic aerosols, we find consistent results in the troposphere and at the surface, but the scaling factor for the OHC was found to be lower and not yet detectable in observations. The uncertainty range of the OHC aerosol scaling factor brackets the scaling factors for the surface and troposphere, so it is possible that these “between-variable” differences are due to internal variability. It is also possible that they reflect differences in the ability of the model to accurately simulate the responses in the different components of the climate system, or to residual observational errors (and inconsistences between observed TMT, ST, and OHC data). We note, however, that the scaling factors we find are more internally consistent compared to those obtained in previous studies that did not examine a consistent analysis period for TMT, ST, and OHC and did not make the same methodological choices in applying fingerprinting to each of these variables (Lott et al. 2013; Gillett et al. 2013; Swart et al. 2018).
Combining the model-predicted fingerprints from the troposphere, surface, and ocean reduced uncertainty in the scaling factors. While this reduction was small for the greenhouse gas fingerprint, it was substantial for the multivariate anthropogenic aerosol fingerprint, leading to earlier detection times in observations (relative to detection times computed with the “single-variable” AER fingerprints). The reason for this reduction in uncertainty is likely that each variable contributes overlapping but nonidentical information about forced signals and unforced variability. Our results suggest that these “between-variable” differences in the spatiotemporal structure of signals and noise are useful in disentangling forced and unforced temperature changes.
For example, over the period analyzed here (1979–2018), the net effect of aerosols on the surface and the troposphere is warming over the Arctic, while the ocean shows widespread cooling in response to anthropogenic aerosol forcing. These fingerprint differences likely result in internal variability projecting onto the fingerprints in ways that are not strongly correlated, thereby reducing uncertainty in the estimated scaling factors. The smaller impact of the multivariate GHG fingerprint on scaling factor uncertainty (relative to the “single-variable” GHG fingerprints) probably arises because each of the TMT, ST, and OHC GHG fingerprints shows strong, spatially coherent warming, and are therefore more similar than “single-variable” fingerprints in the AER case.
The model we used in our analysis yields warming over the historical period that is significantly larger than the observed warming. One potential explanation for this result could be that the model underestimates the net cooling effect of anthropogenic aerosols over this period. The results presented here suggest that this explanation is unlikely for several reasons. First, over the time period considered, the model simulates weak warming in the troposphere and surface in response to aerosol forcing. An underestimation of the impact of aerosols would therefore result in weaker warming than observed. Second, the scaling factors we obtain for greenhouse gases are substantially less than one, while the aerosol scaling factors are consistent with one. Our results support the interpretation that a model overestimate of the response to greenhouse gas forcing is more likely than an underestimate of the response to aerosol forcing. Our confidence in this conclusion is reinforced by the fact that it is based on data from the different parts of the climate system and different observational datasets.
We performed our analysis using a large initial condition ensemble generated with a single model only. It would be useful to repeat the analysis presented here with more model large ensembles spanning a wider phase space of aerosol forcing uncertainty and climate sensitivity. Such multimodel large ensemble analyses may help to more reliably constrain the temperature changes attributed to anthropogenic greenhouse gases and aerosols. Additionally, a number of recent studies have relied on observed surface warming to constrain projected warming (Ribes et al. 2021; Tokarska et al. 2020). Information on the spatiotemporal patterns of warming in the ocean and troposphere could help to further reduce uncertainties in projected warming.
Acknowledgments.
We thank the many centers for producing and making available the observational datasets used in this study, and we acknowledge the Canadian Centre for Climate Modelling and Analysis for running the CanESM2 large ensemble and the Canadian Sea Ice and Snow Evolution Network for proposing the simulations. Neil Swart and Nathan Gillett provided constructive comments on the manuscript. B. D. S was supported by the Francis E. Fowler IV Center for Ocean and Climate at Woods Hole Oceanographic Institution.
Data availability statement.
All data used in this study are publicly available. The CanESM2 ocean and surface temperature data are available from https://open.canada.ca/data/en/dataset/aa7b6823-fd1e-49ff-a6fb-68076a4a477c. The CanESM2 synthetic satellite temperature data and satellite observations are available from https://pcmdi.llnl.gov/research/DandA/PNAS_2019/index.html. The surface temperature are available from HadCRUT5 (https://www.metoffice.gov.uk/hadobs/hadcrut5/), Berkeley Earth (https://berkeleyearth.org/data/) NOAA GlobalTemp (https://www.ncei.noaa.gov/products/land-based-station/noaa-global-temp), and NASA GISTEMP (https://data.giss.nasa.gov/gistemp/). Ocean heat content was calculated using ocean temperature data from EN4 (https://www.metoffice.gov.uk/hadobs/en4/) and NCEI/NOAA (https://www.ncei.noaa.gov/products/climate-data-records/global-ocean-heat-content).
REFERENCES
Acosta Navarro, J. C., and Coauthors, 2016: Amplification of Arctic warming by past air pollution reductions in Europe. Nat. Geosci., 9, 277–281, https://doi.org/10.1038/ngeo2673.
Allen, M. R., and S. F. B. Tett, 1999: Checking for model consistency in optimal fingerprinting. Climate Dyn., 15, 419–434, https://doi.org/10.1007/s003820050291.
Allen, M. R., and Coauthors, 2006: Quantifying anthropogenic influence on recent near-surface temperature change. Surv. Geophys., 27, 491–544, https://doi.org/10.1007/s10712-006-9011-6.
Arora, V. K., and Coauthors, 2011: Carbon emission limits required to satisfy future representative concentration pathways of greenhouse gases. Geophys. Res. Lett., 38, L05805, https://doi.org/10.1029/2010GL046270.
Barnett, T. P., and Coauthors, 1999: Detection and attribution of recent climate change: A status report. Bull. Amer. Meteor. Soc., 80, 2631–2660, https://doi.org/10.1175/1520-0477(1999)080<2631:DAAORC>2.0.CO;2.
Barnett, T. P., D. W. Pierce, K. M. AchutaRao, P. J. Gleckler, B. D. Santer, J. M. Gregory, and W. M. Washington, 2005: Penetration of human-induced warming into the world’s oceans. Science, 309, 284–287, https://doi.org/10.1126/science.1112418.
Bellomo, K., L. N. Murphy, M. A. Cane, A. C. Clement, and L. M. Polvani, 2018: Historical forcings as main drivers of the Atlantic multidecadal variability in the CESM large ensemble. Climate Dyn., 50, 3687–3698, https://doi.org/10.1007/s00382-017-3834-3.
Bilbao, R. A. F., J. M. Gregory, N. Bouttes, M. D. Palmer, and P. Stott, 2019: Attribution of ocean temperature change to anthropogenic and natural forcings using the temporal, vertical and geographical structure. Climate Dyn., 53, 5389–5413, https://doi.org/10.1007/s00382-019-04910-1.
Bindoff, N. L., and Coauthors, 2014: Detection and attribution of climate change: From global to regional. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 867–952.
Bonfils, C. J. W., B. D. Santer, J. C. Fyfe, K. Marvel, T. J. Phillips, and S. R. H. Zimmerman, 2020: Human influence on joint changes in temperature, rainfall and continental aridity. Nat. Climate Change, 10, 726–731, https://doi.org/10.1038/s41558-020-0821-1.
Cowtan, K., and Coauthors, 2015: Robust comparison of climate models with observations using blended land air and ocean sea surface temperatures. Geophys. Res. Lett., 42, 6526–6534, https://doi.org/10.1002/2015GL064888.
Deser, C., and Coauthors, 2020: Isolating the evolving contributions of anthropogenic aerosols and greenhouse gases: A new CESM1 large ensemble community resource. J. Climate, 33, 7835–7858, https://doi.org/10.1175/JCLI-D-20-0123.1.
Eyring, V., and Coauthors, 2021: Human influence on the climate system. Climate Change 2021: The Physical Science Basis, V. Masson-Delmotte et al., Eds., Cambridge University Press, 423–552.
Fu, Q., and C. M. Johanson, 2005: Satellite-derived vertical dependence of tropical tropospheric temperature trends. Geophys. Res. Lett., 32, L10703, https://doi.org/10.1029/2004GL022266.
Gillett, N. P., V. K. Arora, D. Matthews, and M. R. Allen, 2013: Constraining the ratio of global warming to cumulative CO2 emissions using CMIP5 simulations. J. Climate, 26, 6844–6858, https://doi.org/10.1175/JCLI-D-12-00476.1.
Gillett, N. P., and Coauthors, 2021: Constraining human contributions to observed warming since the pre-industrial period. Nat. Climate Change, 11, 207–212, https://doi.org/10.1038/s41558-020-00965-9.
Gleckler, P. J., and Coauthors, 2012: Human-induced global ocean warming on multidecadal timescales. Nat. Climate Change, 2, 524–529, https://doi.org/10.1038/nclimate1553.
Good, S. A., M. J. Martin, and N. A. Rayner, 2013: EN4: Quality controlled ocean temperature and salinity profiles and monthly objective analyses with uncertainty estimates. J. Geophys. Res. Oceans, 118, 6704–6716, https://doi.org/10.1002/2013JC009067.
Gouretski, V., and F. Reseghetti, 2010: On depth and temperature biases in bathythermograph data: Development of a new correction scheme based on analysis of a global ocean database. Deep-Sea Res. I, 57, 812–833, https://doi.org/10.1016/j.dsr.2010.03.011.
Hasselmann, K., 1979: On the Signal-to-Noise Problem in Atmospheric Response Studies. Royal Meteorological Society, 251–259.
Hasselmann, K., 1993: Optimal fingerprints for the detection of time-dependent climate change. J. Climate, 6, 1957–1971, https://doi.org/10.1175/1520-0442(1993)006<1957:OFFTDO>2.0.CO;2.
Hasselmann, K., 1997: Multi-pattern fingerprint method for detection and attribution of climate change. Climate Dyn., 13, 601–611, https://doi.org/10.1007/s003820050185.
Hegerl, G. C., H. von Storch, K. Hasselmann, B. D. Santer, U. Cubasch, and P. D. Jones, 1996: Detecting greenhouse-gas-induced climate change with an optimal fingerprint method. J. Climate, 9, 2281–2306, https://doi.org/10.1175/1520-0442(1996)009<2281:DGGICC>2.0.CO;2.
Huang, B., and Coauthors, 2020: Uncertainty estimates for sea surface temperature and land surface air temperature in NOAAGlobalTemp version 5. J. Climate, 33, 1351–1379, https://doi.org/10.1175/JCLI-D-19-0395.1.
Kirchmeier-Young, M. C., F. W. Zwiers, and N. P. Gillett, 2016: Attribution of extreme events in Arctic sea ice extent. J. Climate, 30, 553–571, https://doi.org/10.1175/JCLI-D-16-0412.1.
Lamarque, J.-F., G. P. Kyle, M. Meinshausen, K. Riahi, S. J. Smith, D. P. van Vuuren, A. J. Conley, and F. Vitt, 2011: Global and regional evolution of short-lived radiatively-active gases and aerosols in the Representative Concentration Pathways. Climatic Change, 109, 191–212, https://doi.org/10.1007/s10584-011-0155-0.
Lenssen, N. J. L., G. A. Schmidt, J. E. Hansen, M. J. Menne, A. Persin, R. Ruedy, and D. Zyss, 2019: Improvements in the GISTEMP uncertainty model. J. Geophys. Res. Atmos., 124, 6307–6326, https://doi.org/10.1029/2018JD029522.
Levitus, S., J. I. Antonov, T. P. Boyer, R. A. Locarnini, H. E. Garcia, and A. V. Mishonov, 2009: Global ocean heat content 1955–2008 in light of recently revealed instrumentation problems. Geophys. Res. Lett., 36, L07608, https://doi.org/10.1029/2008GL037155.
Levitus, S., and Coauthors, 2012: World ocean heat content and thermosteric sea level change (0–2000 m), 1955–2010. Geophys. Res. Lett., 39, L10603, https://doi.org/10.1029/2012GL051106.
Lott, F. C., P. A. Stott, D. M. Mitchell, N. Christidis, N. P. Gillett, L. Haimberger, J. Perlwitz, and P. W. Thorne, 2013: Models versus radiosondes in the free atmosphere: A new detection and attribution analysis of temperature. J. Geophys. Res. Atmos., 118, 2609–2619, https://doi.org/10.1002/jgrd.50255.
Mann, M. E., and K. A. Emanuel, 2006: Atlantic hurricane trends linked to climate change. Eos, Trans. Amer. Geophys. Union, 87, 233–241, https://doi.org/10.1029/2006EO240001.
Mears, C. A., and F. J. Wentz, 2017: A satellite-derived lower-tropospheric atmospheric temperature dataset using an optimized adjustment for diurnal effects. J. Climate, 30, 7695–7718, https://doi.org/10.1175/JCLI-D-16-0768.1.
Meinshausen, M., and Coauthors, 2011: The RCP greenhouse gas concentrations and their extensions from 1765 to 2300. Climatic Change, 109, 213–241, https://doi.org/10.1007/s10584-011-0156-z.
Morice, C. P., and Coauthors, 2021: An updated assessment of near-surface temperature change from 1850: The HadCRUT5 data set. J. Geophys. Res. Atmos., 126, e2019JD032361, https://doi.org/10.1029/2019JD032361.
Pierce, D. W., P. J. Gleckler, T. P. Barnett, B. D. Santer, and P. J. Durack, 2012: The fingerprint of human-induced changes in the ocean’s salinity and temperature fields. Geophys. Res. Lett., 39, L21704, https://doi.org/10.1029/2012GL053389.
Po-Chedley, S., B. D. Santer, S. Fueglistaler, M. D. Zelinka, P. J. Cameron-Smith, J. F. Painter, and Q. Fu, 2021: Natural variability contributes to model–satellite differences in tropical tropospheric warming. Proc. Natl. Acad. Sci. USA, 118, e2020962118, https://doi.org/10.1073/pnas.2020962118.
Ribes, A., S. Qasmi, and N. P. Gillett, 2021: Making climate projections conditional on historical observations. Sci. Adv., 7, eabc0671, https://doi.org/10.1126/sciadv.abc0671.
Rohde, R. A., and Z. Hausfather, 2020: The Berkeley Earth land/ocean temperature record. Earth Syst. Sci. Data, 12, 3469–3479, https://doi.org/10.5194/essd-12-3469-2020.
Santer, B. D., U. Mikolajewicz, W. Brüggemann, U. Cubasch, K. Hasselmann, H. Höck, E. Maier-Reimer, and T. M. L. Wigley, 1995: Ocean variability and its influence on the detectability of greenhouse warming signals. J. Geophys. Res., 100, 10 693–10 725, https://doi.org/10.1029/95JC00683.
Santer, B. D., and Coauthors, 1996: A search for human influences on the thermal structure of the atmosphere. Nature, 382, 39–46, https://doi.org/10.1038/382039a0.
Santer, B. D., and Coauthors, 2013: Identifying human influences on atmospheric temperature. Proc. Natl. Acad. Sci. USA, 110, 26–33, https://doi.org/10.1073/pnas.1210514109.
Santer, B. D., and Coauthors, 2014: Volcanic contribution to decadal changes in tropospheric temperature. Nat. Geosci., 7, 185–189, https://doi.org/10.1038/ngeo2098.
Santer, B. D., J. C. Fyfe, S. Solomon, J. F. Painter, C. Bonfils, G. Pallotta, and M. D. Zelinka, 2019: Quantifying stochastic uncertainty in detection time of human-caused climate signals. Proc. Natl. Acad. Sci. USA, 116, 19 821–19 827, https://doi.org/10.1073/pnas.1904586116.
Santer, B. D., and Coauthors, 2022: Robust anthropogenic signal identified in the seasonal cycle of tropospheric temperature. J. Climate, 35, 6075–6100, https://doi.org/10.1175/JCLI-D-21-0766.1.
Spencer, R. W., J. R. Christy, and W. D. Braswell, 2017: UAH version 6 global satellite temperature products: Methodology and results. Asia-Pac. J. Atmos. Sci., 53, 121–130, https://doi.org/10.1007/s13143-017-0010-y.
Swart, N. C., S. T. Gille, J. C. Fyfe, and N. P. Gillett, 2018: Recent Southern Ocean warming and freshening driven by greenhouse gas emissions and ozone depletion. Nat. Geosci., 11, 836–841, https://doi.org/10.1038/s41561-018-0226-1.
Tett, S. F. B., J. F. B. Mitchell, D. E. Parker, and M. R. Allen, 1996: Human influence on the atmospheric vertical temperature structure: Detection and observations. Science, 274, 1170–1173, https://doi.org/10.1126/science.274.5290.1170.
Tokarska, K. B., M. B. Stolpe, S. Sippel, E. M. Fischer, C. J. Smith, F. Lehner, and R. Knutti, 2020: Past warming trend constrains future warming in CMIP6 models. Sci. Adv., 6, eaaz9549, https://doi.org/10.1126/sciadv.aaz9549.
van Vuuren, D. P., and Coauthors, 2011: The representative concentration pathways: An overview. Climatic Change, 109, 5–31, https://doi.org/10.1007/s10584-011-0148-z.
Watanabe, M., J.-L. Dufresne, Y. Kosaka, T. Mauritsen, and H. Tatebe, 2021: Enhanced warming constrained by past trends in equatorial Pacific sea surface temperature gradient. Nat. Climate Change, 11, 33–37, https://doi.org/10.1038/s41558-020-00933-3.
Zou, C.-Z., M. D. Goldberg, and X. Hao, 2018: New generation of U.S. satellite microwave sounder achieves high radiometric stability performance for reliable climate change detection. Sci. Adv., 4, eaau0049, https://doi.org/10.1126/sciadv.aau0049.
