Signal and Noise
Can we keep climate science separate from misinformation?
Climate change is real and important. Mitigation and adaptation policies make excellent sense. Scientific integrity is important also. Climate advocates have been extremely successful in promoting representations of the science of extreme weather that depart significantly from the actual research of the scientific community and assessed by the Intergovernmental Panel on Climate Change (IPCC).
This post is inspired by the successful efforts last week of climate activists — including three widely-cited scientists — to enforce misinformation by the legacy media. In a nutshell, ABC News wrote an accurate story about how climate was not a major or even significant factor in the Lahaina, Maui fire and disaster. After being mobbed by the enforcers, the story was changed to emphasize the role of climate. These sort of activist scientists who seek to enforce preferred public narratives have been called the “science police.”
Today’s post pushes back against this narrative enforcement with some actual science. Have a look at the panel below. It shows three versions of a climate time series for annual counts of North Atlantic major hurricanes from 1995 to 2050. Two of the graphs include a large change in climate, one of them does not.
Please take a moment to consider which of the three panels you think includes climate change, which does not, and why you come to that judgment.
Here’s the answer.
The top panel takes annual counts of major hurricanes in the North Atlantic to create a probability distribution based on actual data from 1945 to 2022. For 2023 to 2050, the annual counts are randomly selected from this distribution.
For the middle panel I took this probability distribution based on observations and shifted it — starting instantaneously in 2023 — one count to the right (so, for example, the probability of having 3 major hurricanes becomes the probability of having 4, and so on). This large shift is much larger than changes in counts projected to 2100, but this is just an exercise.
The bottom panel shows a time series based on a shift of 2 counts to the right, representing a much larger change in probabilities.
You can see the three probability distributions in the figure below, with colors to match the panel above. This sort of figure has become common in climate discussions, and you can see an example at the top of this post.
Let me suggest two important lessons from this exercise.
First, neither of the climate change series in the top panel come anywhere close to meeting the IPCC criteria for the detection of climate change.
But wait a second there! We know that in the time series the climate has in fact changed — because I changed it — as shown in the shifted probabilities.
So, what gives?
It is not just possible, but often likely, that for specific climate variables the underlying climate changes and yet the consequences of those changes are not detectable in short (climatically speaking) time series. I chose the period 1995 to 2050 because over that time frame my kids will enter their 50s, and be about my age today. It’s the same as me looking at a time series from the 1960s to today. In practice, a lot of climate research on detection and attribution uses much shorter time series.
It is standard today for climate advocates to link every extreme weather event and disaster to climate change. My suspicion is that when they do so they are not actually saying anything about detection or attribution of changes but rather, they are simply witnessing to the reality of climate change and its deep importance to them. Part of this witnessing may be to preemptively defend against concerns that if climate changes are not perceived to be large enough, people won’t have enough fear — no one wants to be accused of minimizing climate change.
Lesson: Just because the signal of climate change for particular variables cannot (yet) be detected in the context of historical variability does not mean that climate change is not real or important, and in many if not most cases a lack of signal is to be expected.
In fact, that is exactly what the IPCC concludes with respect to most extreme weather events, as shown in the table below and discussed here. The white cells show where signal detection is not expected to be unequivocally achieved at present, by 2050 or by 2100 (respectively in the three right-most columns).
Advocates who promote every extreme event as being caused by — linked to, made worse by, fueled by — climate change are promoting misinformation in almost all cases. It is an expression of faith not science. This complicates public discussions of detection and attribution. Some people are expressing a deeply-held, religious-like sentiment and others are talking about data and evidence. No wonder people talk past each other.
Second, you’ll note in the top figure in the panel above — the one with the black linear trend line — that there is a large decrease in major hurricane activity in the North Atlantic from 1965 to 2050. There were about 4 per year at the start of the time series and only 2 at the end, a drop of 50%.
That decrease is real in the sense that there are fewer major hurricanes towards the end of this record as compared to the beginning, but it has nothing to do with a change in the climate, as the annual probabilities do not change over the entire data range.
Compare the figure below, which shows the entire dataset from which the top figure in the panel above was taken. The below figure shows 1945 to 2080 and you can see an increase of 50% from the start to the end — but this trend also is not the result of a change in climate, as the annual probabilities are constant.
So, what gives?
Many climate variables exhibit large variability — since 1945 we’ve observed anywhere from 0 to 7 major hurricanes yearly in the North Atlantic, that’s a huge variation.
The three probability distributions graphed above represent annual counts of major hurricanes of 2.6 based on observations (black), and about 3.6 (red) and 4.6 (green) in the two shifted probabilities. The shifted probabilities represent increases in major hurricane annual counts of ~40% and ~80% — much larger that almost all model projections assessed by Knutson et al. for 2100 (and many actually project a decrease).
Many projected changes in the behavior of extreme events due to climate change are in fact small in the context of historical variability. That is just a fact — have another look at the IPCC table above. Further, historical variability is sufficiently large that we can look at various time series and fool ourselves into thinking we see change when all we are observing is cherry-picked variability (see the two time series above with the black linear trends).
Lesson: Natural variability is real and significant. It does not mean that climate change is not real or important, but that detecting signals is often difficult even when climate is changing and there is always a risk of erroneously detecting signals where none is present.
There is a good reason that climate change has been detected and attributed by the IPCC for temperature and precipitation at large spatial scales — this is where we have the most measurements and through aggregation, variability becomes smaller, making it easier to separate signal from noise. Consider that there are millions of measurements of temperature and precipitation around the world, but there were only 78 major hurricanes in the North Atlantic from 1945 to 2022 — numbers matter!
But even for temperature, at local levels detection and attribution are challenging, as climate scientists Ed Hawkins explains (emphasis added):
The signal of temperature change varies spatially, as does the size and timing of the natural fluctuations of temperature. Together, these two aspects combine to produce the local experience of how the climate is changing. For the public, this is critical – it is extremely difficult to detect trends over decadal timescales for individual locations, especially in regions of high variability . . .
The challenges of detection and attribution should tell us that both adaptation and mitigation policies must be built upon a foundation that involves justifications for action that are much broader than climate change alone.
So far, climate advocates have sought to shape perceptions of science to support a climate-change-is-everything agenda. We will have a lot more success if we instead shape policy to align with what science actually says.
For a deeper and more mathematically focused exploration of this issue, see:
Crompton, R. P., Pielke, R. A., & McAneney, K. J. (2011). Emergence timescales for detection of anthropogenic climate change in US tropical cyclone loss data. Environmental Research Letters, 6(1), 014003.
I welcome your comments and discussion. Please like, share and send to your favorite climate journalist. Thanks for reading and subscribing!
The elephant in the room is the magnitude of natural climate change, that is the climate change not caused by human generated CO2. Otherwise known as the climate change humans can do nothing about. It has become fairly clear to me, at least, that the magnitude of natural climate change is at least as high as the human caused climate change, quite possibly much larger. And this is the reason that the IPCC and others have such trouble identifying the signal of climate change caused by CO2. The natural consequence of that observation is that humans attempting to stop the use of fossil fuel in order to stop 'climate change' becomes an exercise in futility. I also think it is clear that much of the climate change community has realized this, and that is the reason that they have become more and more strident. They are desperate to claim relevancy in a discussion in which human caused climate change is just not significant.
It occurs to me that your point about the challenge of distinguishing an imposed signal from random noise could be made simply by comparing the uncertainties of the slopes of the regression lines.