Search

Four time-dependent greenhouse warming experiments were performed with the same global coupled atmosphere-ocean model, but with each simulation using initial conditions from different ''snapshots'' of the control run climate. The radiative forcing - the increase in equivalent CO2 concentrations from 19852035 specified in the Intergovernmental Panel on Climate Change (IPCC) scenario A - was identical in all four 50-year integrations. This approach to climate change experiments is called the Monte Carlo technique and is analogous to a similar experimental set-up used in the field of extended range weather forecasting. Despite the limitation of a very small sample size, this approach enables the estimation of both a mean response and the ''between-experiment'' variability, information which is not available from a single integration. The use of multiple realizations provides insights into the stability of the response, both spatially, seasonally and in terms of different climate variables. The results indicate that the time evolution of the global mean warming signal is strongly dependent on the initial state of the climate system. While the individual members of the ensemble show considerable variation in the pattern and amplitude of near-surface temperature change after 50 years, the ensemble mean climate change pattern closely resembles that obtained in a 100-year integration performed with the same model. In global mean terms, the climate change signals for near surface temperature, the hydrological. cycle and sea level significantly exceed the variability among the members of the ensemble. Due to the high internal variability of the modelled climate system, the estimated detection time of the global mean temperature change signal is uncertain by at least one decade. While the ensemble mean surface temperature and sea level fields show regionally significant responses to greenhouse-gas forcing, it is not possible to identify a significant response in the precipitation and soil moisture fields, variables which are spatially noisy and characterized by large variability between the individual integrations.

Observed and model simulated warming is particularly large in high latitudes, and hence the Arctic is often seen as the posterchild of vulnerability to global warming. However, Mahlstein et al (2011) point out that the signal of climate change is emerging locally from that of climate variability earliest in regions of low climate variability, based on climate model data, and in agreement with observations. This is because high latitude regions are not only regions of strong feedbacks that enhance the global warming signal, but also regions of substantial climate variability, driven by strong dynamics and enhanced by feedbacks (Hall 2004). Hence the spatial pattern of both observed warming and simulated warming for the 20th century shows strong warming in high latitudes, but this warming occurs against a backdrop of strong variability. Thus, the ratio of the warming to internal variability is not necessarily highest in the regions that warm fastest—and Mahlstein et al illustrate that it is actually the low-variability regions where the signal of local warming emerges first from that of climate variability. Thus, regions with strongest warming are neither the most important to diagnose that forcing changes climate, nor are they the regions which will necessarily experience the strongest impact. The importance of the signal-to-noise ratio has been known to the detection and attribution community, but has been buried in technical 'optimal fingerprinting' literature (e.g., Hasselmann 1979, Allen and Tett 1999), where it was used for an earlier detection of climate change by emphasizing aspects of the fingerprint of global warming associated with low variability in estimates of the observed warming. What, however, was not discussed was that the local signal-to-noise ratio is of interest also for local climate change: where temperatures emerge from the range visited by internal climate variability, it is reasonable to assume that changes in climate will also cause more impacts than temperatures that have occurred frequently due to internal climate variability. Determining when exactly temperatures enter unusual ranges may be done in many different ways (and the paper shows several, and more could be imagined), but the main result of first local emergence in low latitudes remains robust. A worrying factor is that the regions where the signal is expected to emerge first, or is already emerging are largely regions in Africa, parts of South and Central America, and the Maritime Continent; regions that are vulnerable to climate change for a variety of regions (see IPCC 2007), and regions which contribute generally little to global greenhouse gas emissions. In contrast, strong emissions of greenhouse gases occur in regions of low warming-to-variability ratio. To get even closer to the relevance of this finding for impacts, it would be interesting to place the emergence of highly unusual summer temperatures in the context not of internal variability, but in the context of variability experienced by the climate system prior to the 20th century, as, e.g. documented in palaeoclimatic reconstructions and simulated in simulations of the last millennium (see Jansen et al 2007). External forcing has moved the temperature range around more strongly for some regions and in some seasons than others. For example, while reconstructions of summer temperatures in Europe appear to show small long-term variations, winter shows deep drops in temperature in the little Ice Age and a long-term increase since then (Luterbacher et al 2004), which was at least partly caused by external forcing (Hegerl et al 2011a) and therefore 'natural variability' may be different from internal variability. A further interesting question in attempts to provide a climate-based proxy for impacts of climate change is: to what extent does the rapidity of change matter, and how does it compare to trends due to natural variability? It is reasonable to assume that fast changes impact ecosystems and society more than slow, gradual ones. Also, is it really the mean seasonal temperature that counts, or should the focus change to extremes (see Hegerl et al 2011b)? Is seasonal mean exceedance of the prior temperature envelope a good and robust measure that also reflects these other, more complex diagnostics? Lots of food for thought and research! References Allen M R and Tett S F B 1999 Checking for model consistency in optimal finger printing Clim. Dyn. 15 419–34 Hall A 2004 The role of surface albedo feedback in climate J. Clim. 17 1550–68 Hasselmann K 1979 On the signal-to-noise problem in atmospheric response studies Meteorology of Tropical Oceans ed D B Shaw (Bracknell: Royal Meteorological Society) pp 251–9 Hegerl G C, Luterbacher J, Gonzalez-Ruoco F, Tett S F B and Xoplaki E 2011a Influence of human and natural forcing on European seasonal temperatures Nature Geoscience 4 99–103 Hegerl G, Hanlon H and Beierkuhnlein C 2011b Climate science: elusive extremes Nature Geoscience 4 142–3 IPCC 2007 Climate Change 2007: Impacts, Adaption and Vulnerability. Contribution of Working Group II to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change ed M L Parry, O F Canziani, J P Palutikof, P J van der Linden and C E Hanson (Cambridge: Cambridge University Press) Jansen E et al 2007 Palaeoclimate Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change ed S Solomon et al (Cambridge: Cambridge University Press) Luterbacher J et al 2004 European seasonal and annual temperature variability, trends, and extremes since 1500 Science 303 1499–503 Mahlstein I, Knutti R, Solomon S and Portmann R W 2011 Early onset of significant local warming in low latitude countries Environ. Res. Lett. 6 034009

Despite Schneider's claim1, the method we use to estimate equilibrium climate sensitivity from multiple proxy-based reconstructions of the temperature in the Northern Hemisphere2 does account for uncertainty in reconstructions, including that associated with non-temperature and sampling error in the reconstruction. We arrive at a tighter constraint on climate sensitivity not by neglecting uncertainties, but by combining our wide-tailed proxy-based estimate with an independent estimate of climate sensitivity based on twentieth-century warming.

Earth's climate is getting warmer — why? The difficulty for climatologists is discriminating between human-induced changes, principally through the release of 'greenhouse gases' such as carbon dioxide, and natural climate fluctuations. A new approach to this issue employs a variety of indicators to provide annual temperature anomaly patterns over much of the world back to the year AD 1400. For all the uncertainties of such a reconstruction, when these results are compared with external influences on climate it does seem that the main effect on global temperature in the twentieth century has been from increases in greenhouse gases.

Political decisions, adaptation planning, and impact assessments need reliable estimates of future climate change and related uncertainties. To provide these estimates, different approaches to constrain, filter, or weight climate model projections into probabilistic distributions have been proposed. However, an assessment of multiple such methods to, for example, expose cases of agreement or disagreement, is often hindered by a lack of coordination, with methods focusing on a variety of variables, time periods, regions, or model pools. Here, a consistent framework is developed to allow a quantitative comparison of eight different methods; focus is given to summer temperature and precipitation change in three spatial regimes in Europe in 2041–60 relative to 1995–2014. The analysis draws on projections from several large ensembles, the CMIP5 multimodel ensemble, and perturbed physics ensembles, all using the high-emission scenario RCP8.5. The methods’ key features are summarized, assumptions are discussed, and resulting constrained distributions are presented. Method agreement is found to be dependent on the investigated region but is generally higher for median changes than for the uncertainty ranges. This study, therefore, highlights the importance of providing clear context about how different methods affect the assessed uncertainty—in particular, the upper and lower percentiles that are of interest to risk-averse stakeholders. The comparison also exposes cases in which diverse lines of evidence lead to diverging constraints; additional work is needed to understand how the underlying differences between methods lead to such disagreements and to provide clear guidance to users.

This paper focuses primarily on extremes in the historical instrumental period. We consider a range of phenomena, including temperature and precipitation extremes, tropical and extra-tropical storms, hydrological extremes, and transient extreme sea-level events. We also discuss the extent to which detection and attribution research has been able to link observed changes to external forcing of the climate system. Robust results are available that detect and often attribute changes in frequency and intensity of temperature extremes to external forcing. There is also some evidence that on a global scale, precipitation extremes have intensified due to forcing. However, robustly detecting and attributing forced changes in other important extremes, such as tropical and extratropical storms or drought remains challenging.

Human influences have likely already impacted the large-scale water cycle but natural variability and observational uncertainty are substantial. It is essential to maintain and improve observational capabilities to better characterize changes. Understanding observed changes to the global water cycle is key to predicting future climate changes and their impacts. While many datasets document crucial variables such as precipitation, ocean salinity, runoff, and humidity, most are uncertain for determining long-term changes. In situ networks provide long time-series over land but are sparse in many regions, particularly the tropics. Satellite and reanalysis datasets provide global coverage, but their long-term stability is lacking. However, comparisons of changes among related variables can give insights into the robustness of observed changes. For example, ocean salinity, interpreted with an understanding of ocean processes, can help cross-validate precipitation. Observational evidence for human influences on the water cycle is emerging, but uncertainties resulting from internal variability and observational errors are too large to determine whether the observed and simulated changes are consistent. Improvements to the in situ and satellite observing networks that monitor the changing water cycle are required, yet continued data coverage is threatened by funding reductions. Uncertainty both in the role of anthropogenic aerosols, and due to large climate variability presently limits confidence in attribution of observed changes.

This paper addresses the question of where we now stand with respect to detection and attribution of an anthropo- genic climate signal. Our ability to estimate natural climate variability, against which claims of anthropogenic signal detection must be made, is reviewed. The current situation suggests control runs of global climate models may give the best estimates of natural variability on a global basis, estimates that appear to be accurate to within a factor of 2 or 3 at multidecadal timescales used in detection work. Present uncertainties in both observations and model-simulated anthropogenic signals in near-surface air tem- perature are estimated. The uncertainty in model simulated signals is, in places, as large as the signal to be detected. Two different, but complementary, approaches to detection and attribution are discussed in the context of these uncertainties. Applying one of the detection strategies, it is found that the change in near-surface, June through August air tem- perature field over the last 50 years is generally different at a significance level of 5% from that expected from model- based estimates of natural variability. Greenhouse gases alone cannot explain the observed change. Two of four climate models forced by greenhouse gases and direct sulfate aerosols produce results consistent with the current climate change observations, while the consistency of the other two depends on which model's anthropogenic fingerprints are used. A recent integration with additional anthropogenic forcings (the indirect effects of sulfate aerosols and tropospheric ozone) and more complete tropospheric chemistry produced results whose signal amplitude and pattern were consis- tent with current observations, provided the model's fingerprint is used and detection carried out over only the last 30 years of annually averaged data. This single integration currently cannot be corroborated and provides no opportunity to estimate the uncertainties inherent in the results, uncertainties that are thought to be large and poorly known. These results illustrate the current large uncertainty in the magnitude and spatial pattern of the direct and indirect sulfate forcing and climate response. They also show detection statements depend on model-specific fingerprints, time pe- riod, and seasonal character of the signal, dependencies that have not been well explored. Most, but not all, results suggest that recent changes in global climate inferred from surface air temperature are likely not due solely to natural causes. At present it is not possible to make a very confident statement about the relative con- tributions of specific natural and anthropogenic forcings to observed climate change. One of the main reasons is that fully realistic simulations of climate change due to the combined effects of all anthropogenic and natural forcings mecha- nisms have yet to be computed. A list of recommendations for reducing some of the uncertainties that currently hamper detection and attribution studies is presented.

Assessments of the veracity of model-generated variability are difficult because variability in observed climate data during the 20th century is composed of natural and anthropogenic (greenhouse gas and sulfate aerosol) factors in addition to internal climate fluctuations. Comparisons should be improved if some of these factors could either be extracted from real world data or additions be made to model-generated variability. Both options involve several assumptions, the most important of which is that the climate system is linear to a first approximation. We discuss this and other assumptions and present results from two different methods of removing variability related to anthropogenic factors using energy balance models (EBMs) and atmosphere/ocean general circulation models (A/OGCMs). At a global scale, the pattern of the trend of surface temperature over the 1966–1995 period, after removing the anthropogenic effect, shows some strong similarities between the two methods, with the strongest residual warming evident over much of northern Asia and northern North America.

This paper reviews recent research that assesses evidence for the detection of anthropogenic and natural external influences on the climate. Externally driven climate change has been detected by a number of investigators in independent data covering many parts of the climate system, including surface temperature on global and large regional scales, ocean heat content, atmospheric circulation, and variables of the free atmosphere, such as atmospheric temperature and tropopause height. The influence of external forcing is also clearly discernible in reconstructions of hemispheric-scale temperature of the last millennium. These observed climate changes are very unlikely to be due only to natural internal climate variability, and they are consistent with the responses to anthropogenic and natural external forcing of the climate system that are simulated with climate models. The evidence indicates that natural drivers such as solar variability and volcanic activity are at most partially responsible for the large-scale temperature changes observed over the past century, and that a large fraction of the warming over the last 50 yr can be attributed to greenhouse gas increases. Thus, the recent research supports and strengthens the IPCC Third Assessment Report conclusion that “most of the global warming over the past 50 years is likely due to the increase in greenhouse gases.”

Climate models predict a strengthening contrast between wet and dry regions in the tropics and subtropics (30 °S–30 °N), and data from the latest model intercomparison project (CMIP6) support this expectation. Rainfall in ascending regions increases, and in descending regions decreases in climate models, reanalyses, and observational data. This strengthening contrast can be captured by tracking the rainfall change each month in the wettest and driest third of the tropics and subtropics combined. Since wet and dry regions are selected individually every month for each model ensemble member, and the observations, this analysis is largely unaffected by biases in location of precipitation features. Blended satellite and in situ data from 1988–2019 support the CMIP6-model-simulated tendency of sharpening contrasts between wet and dry regions, with rainfall in wet regions increasing substantially opposed by a slight decrease in dry regions. We detect the effect of external forcings on tropical and subtropical observed precipitation in wet and dry regions combined, and attribute this change for the first time to anthropogenic and natural forcings separately. Our results show that most of the observed change has been caused by increasing greenhouse gases. Natural forcings also contribute, following the drop in wet-region precipitation after the 1991 eruption of Mount Pinatubo, while anthropogenic aerosol effects show only weak trends in tropic-wide wet and dry regions consistent with flat global aerosol forcing over the analysis period. The observed response to external forcing is significantly larger (p > 0.95) than the multi-model mean simulated response. As expected from climate models, the observed signal strengthens further when focusing on the wet tail of spatial distributions in both models and data.