Understanding Climate Risk

Science, policy and decision-making

Climate shifts and extremes

with one comment

This posts looks at how climate shifts affect extremes using the example of heat extremes in SE Australia. We had another burst of hot weather this week, which led to rolling power blackouts in South Australia. These are becoming more common, as our electricity bills rise to pay for network infrastructure. In every year but one since 1997, the Laverton, Victoria climate record has registered at least 1 day above 40°C. For those of you interested in how the science of detecting and projecting extremes is carried out, there is a comprehensive background on methods. For those who are just interested in results, page down to the results section.

Background

Attributing extreme climate events to climate change is controversial. Scientists tend to be cautious about doing so to avoid the mistake of attributing cause to an event that may be random. The main tool used to attribute change is a signal-to-noise model of climate indices. Anthropogenic change is assumed as an emerging trend of gradual change. Statistical  significance will only register when that trend emerges from the noise of natural climate variability.

Getting a climate baseline free of external forcing in observed climate is also difficult. This is one reason why climate model baselines run at pre-industrial levels of greenhouse gases are commonly used, although this assumes the models adequately reproduce the statistics of climate variability. In general, while doing a pretty good job, models tend to under-represent some aspects of natural climate variability (usually those on longer timescales). However, they also tend to be less sensitive to forced changes than is currently being observed. Despite that, their skill in both aspects is continually being improved.

The real world is influenced by internal variability; a small number of natural forcing mechanisms including changes in solar irradiance and volcanoes, and a number of human-induced forcings that combine with natural processes and include greenhouse gas emissions, ozone, sulphate aerosols, land cover and land use change, ice and snow cover, CFC effects on the stratospheric ozone layer and  chemical processes influencing atmospheric composition.

When that list is written out people tend to point and shout “Look!! The uncertainty monster – a big hairy thing. And climate has always changed anyway. So there.” Except this is one area where earth system science has done an excellent job over the past 50 years. Most models now include most of those forcings and for mean  climate change in a range of variables, a clear signal has emerged.

Climate extremes are tougher to assess than the mean. The least noisy and more frequent extremes are easier to assess. These include cold nights and warm days, sea surface temperatures, sea level extremes and some rainfall indices, like rainfall over a 1% threshold (the wettest 1% of days). The noiser and less frequent extremes like tropical cyclones, severe storms, hail and tornadoes are tougher. So too, are big systems that have a wide range of influences ranging from seasonal to decadal such as monsoonal systems.

These influences are reflected in confidence statements on observed changes from the recently released IPCC SREX report on extreme events and disasters. For example, hot days and cold nights are very likely to have increased and decreased respectively on a global basis since 1950. On continents with good data coverage, these changes are held to be likely. In Asia, changes are given medium confidence and in Africa and South America low to medium confidence given data coverage. When it comes to global trends in tropical cyclones, however, there is low confidence in assessing any change, as is the case for hail and tornadoes.

The projection of future changes also uses the signal-to-noise model. Many different climate models run with several  emission scenarios that include multiple runs with the same scenario (ensembles) will exhibit large variation between model results. The task of projecting extremes becomes very difficult because the signal has to be extracted not just from the noise within one simulation, but also across different simulations. This is the main reason for statements such as:

Projected changes in climate extremes under different emissions scenarios generally do not strongly diverge in the coming two to three decades, but these signals are relatively small compared to natural climate variability over this time frame. Even the sign of projected changes in some climate extremes over this time frame is uncertain. For projected changes by the end of the 21st century, either model uncertainty or uncertainties associated with emissions scenarios used becomes dominant, depending on the extreme.

Two methods can improve the situation. One develops an understanding of the physical mechanism causing a change and attributes that mechanism to a change process. The other covers fingerprinting methods that extract signals from complex data, picking out broad patterns of change. Often these can distinguish between patterns of natural variability, such as ENSO and the North Atlantic Oscillation, and patterns of change associated with greenhouse warming. However, the latter method, despite being sophisticated, linearise the results.

If the human-induced component of climate is not linear as I’ve been maintaining, but changes  in a step-wise fashion, extremes may also change rapidly. So what does happen?

Results

Temperature in SE Australia shifted upward in a step-wise fashion in 1997, as did temperature in most regions of the world. Maximum temperature rose by 0.8°C 1997-2010 compared to 1910-1996. Taking daily max temperatures from Laverton RAAF, which is the closest high quality designated rural station near Melbourne, I tallied days above 35°C and above 40°C since 1943, when the station opened. Climate is stationary until at least 1968, there was an underlying anthropogenic warming in max temp since at least 1973, suppressed somewhat by above average rainfall in many years. Higher max temp since 1997 has been partly enhanced by below average rainfall.

When the resulting time series for days >35°C for Laverton are subject to statistical tests for shifts, the change in 1997 is significant using the bivariate test  to the 1% level, and to the 5% level using the Rodionov STARS regime shift test and the z test. The average goes from 8 days before the shift to 12 days after, a 50% increase. The shift in days >40°C goes from 0.8 to 2.6 per year. Roughly one-third of that change is the effect of rainfall reduction (which may be anthropogenic) the other two-thirds is due to direct anthropogenic warming.

Step and trend analysis for Tmax for south-eastern Australia 1910–2010

Days above 35°C and above 40°C for Laverton Victoria 1943-2011

This station was used to produce the linear projections of changes in extremes presented in the recently released The Critical Decade: Climate Change and Health report (and featured in the original Garnaut report). The baseline period was 1974-2003 and average days >35°C averaged 9 per year. The projected increase for 2030 was 12 days per year, exactly what we have experienced in the period 1997-2011.  The effect of the oceans on atmospheric warming first suppresses warming because basically most of the energy is going into the ocean, then brings it all on at once in a series of step changes occurring every few decades. This produces significant effects on extreme temperatures.

In recent years these extreme temperatures have led to health effects due to heat stress, damaged crops in warmer areas, disrupted transport (especially trains), increased peak power demand, led to rolling blackouts and contributed to a significantly higher fire risk. The current regional average is likely to persist for a while before another shift any time between now and 2030. If we get a better idea of the exact mechanism that makes the ocean “work”, so that the atmosphere warms rapidly relative to the ocean, these events may become predictable or at least identifiable shortly after the event. Otherwise about 15 years of statistics is needed to positively identify any shift in conditions.

This can be contrasted with Rahmstorf’s recent summary at RealClimate on changing extremes that describes the non-linear effects on variability plus climate change. This can be seen in temperature records from Texas as well as Moscow, where cool conditions existed from 1956-1977, probably due to the North Atlantic Oscillation. Texas warms from 1990 in a step change reminiscent of the one that affected southern Australia in the late 1960s. I’m still working on that data and hope to do a post that looks at last summer’s heat wave. I’ve also had a quick look at the Russian data for 2010, but it’s not in great shape.

Tmax anomaly for 49 climate stations in Texas, not spatially weighted

Tmin anomaly for 49 climate stations in Texas, not spatially weighted

How the climate changes is important information for adaptation. If we are confident that abrupt shifts in extremes are not going to be temporary, then investment in comprehensive adaptation measures is likely to be cost effective. Anticipating a timely response to further abrupt changes in advance also makes sense.

The signal to noise model, which extracts a linearised signal from climate model ensembles is not so useful in this context. More useful questions are “Are shifts more frequent and larger under high emissions scenarios?”  and “If models show step changes in extremes (they do), what are diagnosable signs that we can look for in observations?”

If climate moves through a series of shifts, or jerks, it requires re-think how we analyse and use climate information from both the observations and climate models. On the other hand, if we better understand how the signal changes, the decades of delay in recognising a linear response may no longer apply – at least for some of the more frequent extremes.

Advertisements

One Response

Subscribe to comments with RSS.

  1. […] of his Ph D was to take the bivariate test of Maronna and Yohai, which was being run manually to detect step changes, and put together an objective rule-based program that would run multiple time series. That took […]


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: