Making sense of the slowdown in global surface warming

The slowdown in global warming is a subject of intense study. Is it a real physical effect, or a few chance cool years, or something more complex? Could it have been predicted? Can we understand it in retrospect? The following lecture and commentary from the Denial101x course attempt to summarize recent work on the subject. However it is a very fast-moving field, so this summary can only cover a small fraction of the material and will quickly become out-of-date (if it is not already so).

View on YouTube

Making Sense of the Slowdown: Commentary

The term 'hiatus' is often applied to describe a slowdown in the rate of global warming since the late 1990s or the early 2000s. However there are two separate questions which are often confused in discussion of the hiatus. The first is whether there has been a change in the rate of warming, while the second concerns whether the rate of warming is in line with model projections.

When looking at the rate of warming, the year-to-year variability makes it hard to draw conclusions from short periods, especially if we are allowed to cherry-pick a start date. Separating a change in the rate of warming from a few chance cool years is hard, however careful analysis of climate models suggest that recent changes in the rate of warming can occur naturally, but are uncommon (Roberts et al 2015).

However the interesting scientific question is not whether changes in the rate of warming can be explained by chance, but rather how they affect our understanding of the climate system. Given that one way we understand the drivers of climate is by doing experiments in climate models, any difference between the models and observations is interesting.

Climate models

Climate model projections are based on predictions of what the drivers of climate will be over future years - which involves making educated guesses of not only human activity, but also solar cycles, and volcanic eruptions. The problem is that if those guesses are wrong, the climate models will give the wrong results, even if their simulation of the physics is perfect.

Climate model projections for temperature have been higher than the observations for most model runs over the last decade - see for example IPCC (2013) figure 9.8. That could mean that there is a problem with the models, but it could also be a problem with the inputs.

The problem is further complicated by internal variability - the long term climate signal is hidden under chaotic short term variations - which we can loosely describe as weather. If the same model is run several times with the same inputs and almost identical starting points, it will give different results. Any difference triggers a 'butterfly effect' leading to wildly divergent results. We can retrieve the climate signal from a climate model by running it lots of times and averaging the results.

The real world shows chaotic behaviour too, and we can't average that out because we only have one reality. But to determine whether the models are giving realistic results over periods of a decade or two, we have to first eliminate the chaotic part of the signal.

So comparing models to observations is hard. First we must correct the inputs to match what actually happened since the models were run, then we have to correct for the effects of internal variability. A number of recent papers have attempted exactly this, and we will look at some of them in detail.

Schmidt et al (2014): Reconciling warming trends

Schmidt and colleagues estimate the effect of updating the inputs to the models, including the effects of increased volcanic activity (stratospheric aerosols), a weak solar cycle, the cooling effect of pollution (tropospheric aerosols), and also higher than expected greenhouse gas emissions (although this last term is tiny). They adjust the model outputs to match the observed variability in the Pacific (the El Niño Southern Oscillation or ENSO) by estimating the contribution of El Niño to global temperatures in previous years

When they compare the adjusted model results with global temperature series from NASA GISTEMP, and from Cowtan & Way (2014) they obtain a good agreement. The largest contributor to the hiatus by this method is the increase in volcanic activity, followed by increased pollution, and then the weak solar cycle and a recent trend towards cool La Nina events.

The animated comparison of the model outputs to the data shown in the lecture is based on this paper.

Huber and Knutti (2014) Natural variability, radiative forcing and climate response in the recent hiatus

Huber and Knutti perform a similar analysis: Like Schmidt and colleagues, they adjust the model outputs to account for the difference between the predicted volcanic and solar contributions and what actually happened. Unlike Schmidt and colleagues they do not include any adjustment for an increased cooling effect due to human pollution. The biggest difference in their approach is that rather than estimating the temperature impact of El Niño cycles from the historical record, they select segments from existing climate model runs which show similar El Niño behaviour to what actually happened, and use these to estimate the impact on global surface temperatures.

In contrast to Schmidt and colleagues, Huber and Knutti find the the trend towards the cool phase of the El Niño cycle is the biggest contributor to the hiatus, with the solar and volcanic contributions each being about half the size of the El Niño contribution. However they also find good agreement with the UK Met Office temperature data once it has been extended to global coverage.

Risbey et al (2014) Well-estimated global surface warming in climate projections selected for ENSO phase

In an approach similar to that of Huber and Knutti, Risbey and colleagues select segments from climate model runs which show a similar trend in El Niño over overlapping 15 year periods. They do not however update the volcanic or solar contributions. When they compare these selected model segments to the observed trends from global temperature series from NASA GISTEMP and from Cowtan & Way (2014) they obtain a good agreement. This supports the suggestion of Huber and Knutti that the El Niño contribution is the most significant.

Saffioti et al (2015) Contributions of Atmospheric Circulation Variability and Data Coverage Bias to the Warming Hiatus

All of the preceding papers have examined the contribution of the tropical Pacific, and in particular the El Niño cycle to the slowdown in warming. However this presents a conundrum: The slowdown in warming is not a global phenomena. In fact, most of the planet has continued to warm, however this warming has been masked by a much larger localised cooling of the northern mid-latitudes, particularly over Asia. Furthermore, the cooling is primarily a winter phenomena (Cohen et al, 2012).

Saffioti et al examine air pressure data for the northern hemisphere excluding the tropics, and find two modes of internal variability (one being the North Atlantic oscillation, also investigated by Trenberth and Fasullo 2013) which have contributed to a cooling over the hiatus period, with a similar geographical pattern to the observed cooling. These variations, along with the omission of the Arctic from observational datasets, explain the majority of the slowdown in warming.

Evidence for and against the different contributions to the hiatus

In the lecture, the different contributions to the hiatus were summarised in this graphic, illustrating how the greenhouse warming has been offset by cooling effects from some combination of the El Niño cycle, volcanic activity, the solar cycle, pollution, and incomplete coverage of the observations.


Unpicking the precise impact of the different contributions is clearly not a solved problem. We can however look at the evidence for the existence of each effect in turn.

El Niño

The influence of the El Niño cycle has long been known, and can be seen by simply comparing the El Niño cycle against the global temperature record over various time spans, demonstrated by papers including Lean and Rind (2009) and Foster and Rahmstorf (2011). However the fact that the cycles agree does not necessarily mean that one causes another - additional evidence is required.

Kosaka and Xie (2013) (blog post) used an ensemble of climate model runs and constrained the ocean surface temperatures over a small region of the Pacific to match observations, and found that the resulting global temperatures tracked observations well, suggesting that El Niño does indeed play a driving role in global temperature, and that the El Niño trend plays a dominant role in the hiatus.

England et al (2014) (blog post) find a possible mechanism for the El Niño trend in a strengthening of easterly winds in the tropical Pacific, driving heat down into the western pacific ocean, but with impacts in the Indian ocean as well. When replicated in models this mechanism explains a large part, but not all of the hiatus.

The El Niño cycle is not the only natural cycle in the climate system. Some have questioned whether other cycles might be involved in the hiatus. Steinman et al (2015) (blog post) address this question by examining the effect of both Atlantic and Pacific oscillations. They develop a method to isolate internal climate variability by subtracting out the human contribution from climate model runs, then look for the same signature in the observations to determine the impact of internal variability in the real temperature record. They find a small contribution from the Atlantic to the current warmth, but a major role for the Pacific in the slowdown in warming.

These studies provide a range of evidence for a role of internal variability, particularly in the tropical Pacific, in the slowdown in global warming. The results typically explain between half and most of the slowdown.


Climate projections assume an 'average' frequency of volcanic eruptions, and while the 2000's have not featured any massive eruptions like the 1991 Pinatubo eruption, there have been an increasing number of modest eruptions. Our graphic shows all of the volcanic eruptions with a Volcanic Explosivity Index (VEI) of 4 or greater, however in reality the type and location of the eruption are also important.

One of the primary sources of evidence comes from satellite observations of the transparency of the earth's atmosphere: Vernier et al (2011) detected the impact of volcanic particles using three different satellite instruments. Santer et al (2015) also detect the volcanic influence in surface observations, including  incoming solar radiation, air temperatures, tropical and global sea surface temperatures and water vapour observations.

The impact of volcanic emissions is also demonstrated through a number of modelling studies, for example Fyfe et al (2013) who find that volcanic emissions could have cooled the earth by 0.07°C compared to projections. Other studies include Solomon et al (2011), Ridley et al (2014), and Haywood et al (2014) with the last study finding a smaller effect.

Solar cycle

The influence of the solar cycle on recent climate is in some respects the simplest to understand, since the intensity of incoming solar radiation has been directly measured by satellites since 1978 (e.g. Fröhlich 2000). The last solar cycle has featured a prolonged minimum followed by a rather weak maximum, with a corresponding cooling effect on the climate. However while the amount of incoming radiation is measured, there are possible secondary effects from the impact of variations in ultraviolet radiation on atmospheric chemistry which could amplify the direct impact of the weak solar cycle (e.g. Shindell et al 1999).


The Schmidt et al study found a cooling contribution to the hiatus from industrial pollution particles (aerosols) emitted into the atmosphere, with the principal change being rapid industrialisation in Asia. This was also suggested by Kaufmann et al in 2011. In our graphic the factory symbols track the growth of the Chinese economy. This is probably the most contentious contribution to the hiatus, with recent studies like Regayre et al (2014) and Gettelman et al (2015) finding little or no impact from pollution aerosols over the hiatus period. However Schmidt et al do include the effect of nitrates, whereas the other studies are restricted to sulphate pollution.


The widely used HadCRUT4 temperature data from the UK Met Office only covers about five sixths of the planet, with significant areas of missing coverage over the Arctic, Antarctica and central Africa. Cowtan and Way (2014) examined the effect of missing coverage, and found that if the observations are extended to cover the whole globe either by infilling from nearby weather stations, or using satellite data to fill the gaps, the temperature trend over the hiatus period substantially increased.

About half of the increase comes from the Arctic, which although comprising a rather small region has been warming many times faster than the rest of the planet. The rapid warming of the Arctic is well established from observations (e.g. Berkeley Earth), weather models (Simmons and Poli 2014, Saffioti et al 2015), and infrared observations from satellites (Comiso and Hall 2014), and is of course implicated in the decline in Arctic sea ice.

The other half of the coverage bias is split between Antarctica and the rest of the world. For the rest of the world, coverage is reasonable and the agreement between Berkeley Earth, NASA GISTEMP and Cowtan and Way are good. Antarctica is more difficult with very few stations covering a massive area. GISTEMP and Cowtan and Way show similar trends, however Berkeley Earth shows slower warming and better agreement with the weather models. Therefore while most of the coverage bias is well founded, the Antarctic contribution is more uncertain.


There is substantial evidence that pacific variability, volcanic cooling, the solar cycle, and incomplete data coverage have all contributed to offsetting greenhouse warming since the late 1990s. Pollution may also have played a role, but is more contentious. In size, the effects of El Niño, volcanoes, and coverage are all substantial. The impact of the weak solar cycle is smaller unless it has been significantly amplified through effects on atmospheric chemistry. If all the effects reviewed here are as large as the authors suggest then the hiatus probably over explained, however this assumes they are independent - some of the effects could be different ways of looking at the same phenomenon.

So we have a good idea what factors are causing the hiatus. The size of each of the effects is less certain - together they may fall a little short of explaining the slowdown in warming, or they may overshoot. Estimates for the different cooling influences over recent years will continue to be improved, giving us a clearer picture of how climate changes over short as well as longer timescales.

The hiatus has not changed our understanding of the fundamentals of climate science in terms of the greenhouse effect and amplifying feedbacks. However the desire to understand short term trends, rather than the long term changes on which climate science has traditionally focussed, has driven significant new work into internal variability, volcanic impacts and observational biases. In other words it has forced us to improve our understanding of the climate system, but so far those changes have not affected our expectations concerning future climate change.


Posted by Kevin C on Tuesday, 26 May, 2015

Creative Commons License The Skeptical Science website by Skeptical Science is licensed under a Creative Commons Attribution 3.0 Unported License.