Overestimating the Human Influence on the Economic Costs of Extreme Weather Events

Estimating the human influence on extreme weather events and their economic costs is relevant to many policy discussions around climate change including those that concern the social cost of carbon.

There are many different ways to estimate the economic costs of climate change that range from theoretical to empirical and apply to entire economies or the impacts of single events.

In this post, I will discuss a particular method used to estimate the economic damages attributable to climate change for a single disaster.

Attributable Costs

The method has been referred to as the “attributable costs” method. It traces its origins to a 2003 Nature commentary called “Liability for climate change” (Allen, 2003). The critical passage is the following:

“If at a given confidence level, past greenhouse-gas emissions have increased the risk of a flood tenfold, and that flood occurs, then we can attribute, at that confidence level, 90% of any damage to those past emissions”.

The statement above was hypothetical but more recently, this reasoning has been adopted in the climate impacts literature.

Notably, it was used in an assessment of the economic impact of human influence on Hurricane Harvey (Frame et al., 2020). Specifically, Frame et al. (2020) state that estimated damages from Hurricane Harvey were $90 billion and that human influence on the climate was responsible for 75% of this $90 billion or ~$67 billion. They then use this number to argue that traditional estimates of the costs of climate change that inform the social cost of carbon estimates are drastically underestimated, as the authors’ post on Carbon Brief communicates:

Cost of extreme weather due to climate change is severely underestimated.

This study received a great deal of attention. According to Altmetric, of over 20 million research outputs tracked, it was in the 99th percentile in terms of online attention. It was cited prominently (5 times) in the IPCC AR6 Working Group 2 Report. It was promoted on social media by high-profile climate communicators, endorsed by a climate scientist on Time Magazine’s 100 most influential people list, and made headlines in places like The Guardian.

Despite the attention and high-profile influence of this study and others like it, I believe it contains a fundamental flaw in reasoning that undermines its results.

The study makes use of the concept of the “Fraction of Attributable Risk (FAR)” which is a concept borrowed from epidemiology (Levin et al., 1953).

The idea is that you can quantify how a change in conditions affects the risk of some outcome. For example, it can quantify how exposure to a particular chemical affects the risk of contracting cancer over some period of time. The formula is,

where P(outcome | normal conditions) should be read as “the probability of the outcome given normal conditions”. For the chemical-cancer example, the outcome would be the contraction of cancer, normal conditions would correspond to a group not exposed to the chemical and altered conditions would correspond to a group exposed to the chemical. Probabilities of cancer for the two groups could be estimated empirically by calculating the percentage in each group that was observed to contract cancer (Attributable fraction among the exposed).

If exposure to the chemical doubles the risk of cancer, and an exposed individual contracts cancer, then half of the risk of their contraction of cancer can be attributed to exposure to the chemical. If exposure to the chemical triples the risk of cancer then 2/3rds of their risk of cancer can be attributed to exposure and so on.

Now, we can further imagine that treatment of this cancer universally costs $20,000. If the fraction of attributable risk is 3/4ths or 75%, then a person exposed to the chemical with cancer can claim that

(0.75)x($20,000) = $15,000

of the expense of cancer treatment is due to that exposure.

This is the “attributable costs” method. Expressed as a formula it is:

Implicit in the above formula is the notion that there is no cost if the outcome does not take place. In that sense, you could also think of the fraction of attributable risk as being multiplied by the difference between the cost with the outcome and the cost without the outcome where the cost without the outcome is zero.

The reasoning here makes sense because contracting cancer can more-or-less be considered a binary (either you contract it or you don’t). That means the $20,000 cost for treatment is also a binary (either $20,000 is paid for treatment or $0 for no treatment) and thus the ‘cost without outcome’ is zero and you can calculate the full financial impact by using the “attributable costs” method which simply multiplies the fraction of attributable risk by the expense of the outcome.

Analogously, the attributable costs of a weather event to human influence on the climate can be written as:

In Frame et al. (2020) they are interested in quantifying the economic damage from Hurricane Harvey that is due to human influence. But the first issue that arises is how to define the “weather event” they are studying.

The first idea that might come to mind might be that “landfalling tropical cyclones” constitute the event. Landfalling tropical cyclones are discrete phenomena that can be said to either occur or not and would thus more-or-less fit into this framework. However, many estimates of changes in tropical cyclone activity actually project a decrease in total tropical cyclone number under climate change. This would yield a negative attributable cost. However, the researchers of Hurricane Harvey weren’t necessarily interested in changes in the odds of all landfalling tropical cyclones but rather changes in the odds of tropical cyclones very much like Hurricane Harvey. Harvey was notable, in particular, because of its extreme rainfall and there is robust evidence that the most extreme rainfall should be enhanced as the world warms.

Thus, Frame et al. (2020) define the “weather event” using rainfall totals that breach the threshold of being as high or higher than what was seen during Hurricane Harvey.

Unlike landfalling tropical cyclones, however, rainfall totals are not really discrete phenomena that you can count in the same way. It’s more natural to think of rainfall totals as coming from a continuous probability distribution of possible rainfall totals.

This move from a discrete weather event, to defining an “event” by choosing a threshold from a continuum renders the “attributable costs” method invalid. When we unpack the logic into steps we can see where it breaks down:

  1. Hurricane Harvey caused $90 billion in damage
  2. The definition of the “event” representing Harvey is rainfall at or above the amount seen during Harvey
  3. Human influence on the climate is responsible for 75% of the risk of the “event”, so defined.
  4. Thus, human influence on the climate is responsible for 75% of $90 billion of damage, or $67 billion.

The above reasoning smuggles in the idea from the epidemiological example that the “event” is dichotomous: either it occurs or it does not and thus the $90 billion in damage either occurs or it does not. The reasoning requires that the ‘cost without the outcome’ is zero.

The costs from cancer only activate if you contract cancer. There are no costs for cancer if you don’t contract cancer. But the costs from rainfall begin to accumulate long before the event threshold is reached. There are costs from rainfall even when there is no ‘event’, so defined.

One could argue that the cost without the outcome would be zero if the ‘event’ was landfalling tropical cyclones (if there is no landfalling tropical cyclone then there is no damage from a landfalling tropical cyclone) but one cannot claim this for rainfall beyond a threshold. Imagine 1/100th of an inch less rain fell during Harvey. Would the $90 billion in damage disappear? Of course not. All of the rain from Harvey, including that last 1/100th of an inch, is required for the 75% ‘fraction of attributable risk’ to be valid but not all of the rain is required for there to be any damage at all.

In summary, the authors of Frame et al. (2020) are implicitly attributing all $90 billion in damages to the very last 1/100th of an inch of rainfall which is what allowed the arbitrary threshold to be crossed and  ‘eventhood’ to be achieved. This flawed reasoning also applies to the original 2003 Nature commentary which used flooding as the example.

Changes in the Magnitude of an Event

A much clearer framing of the question is obtained when human influence on changes in the magnitude of some aspect of the weather are focused on rather than the frequency of crossing an arbitrary threshold (I have discussed the relationship between these previously Brown, 2016). Some of the studies referenced in Frame et al. (2020) also estimate the human influence on the magnitude of rainfall during Harvey and they come up with the following figures:

“Human-induced climate change likely increased Hurricane Harvey’s total rainfall by at least 19%” (Risser and Wehner at al., 2017)

“We conclude that global warming made the precipitation about 15% (8%–19%) more intense” (van Oldenborgh et al., 2017)

Immediately these numbers should give one pause when they are compared to the 75% estimate from Frame et al. (2020). Does the paper claim that 75% of the damage from Harvey came from the additional 15% to 19% of precipitation? That would be theoretically possible if the additional 15%-19% of rain caused some physically meaningful threshold to be breached (like one corresponding to the height of a levee). However, the authors do not claim this and in fact, one of the authors calculates in a more recent study that the 20% increase in rain results in only a 15% increase in flood area (Wehner and Sampson, 2021).

Looking at the human influence on the magnitude rather than on the frequency of crossing an arbitrary threshold, yields an estimate of the damages from human influence of $13 billion (Wehner and Sampson, 2021). Taken at face value, this would indicate that the logical flaw in the “attributable costs” method led to an overestimation of the damages from Harvey by a factor of about 5.

However, I contend that even this type of calculation is misleading if one wishes to get a holistic sense of the economic impact of climate change in any given year. Frame et al. (2020) are indeed interested in this: in both the paper itself and in their CarbonBrief article they compare their damage estimate from Harvey to those from Integrated Assessment Models (IAMs) which correspond to mean damages for any given year. They argue that IAM-based damages must be huge underestimates because they only calculate annual mean costs of ~$20 billion per year in the US while their study attributes $67 billion to human influence for a single event (Harvey).

I argue that this too is a deeply flawed comparison because the rarity of an event like Hurricane Harvey would have to be taken into account to compare it to an annual mean estimate of damages from human influence on the climate. I illustrate what I believe is a better way to get a full sense of the human influence on economic damages below.

Expected Damages with and without Human Influence

Figure 1 illustrates a more complete way of estimating the economic damages from rainfall in Houston with and without climate change. This is intended to be an illustrative exercise and none of the exact numbers should be taken too seriously.

Figure 1 | Analysis of the expected values of various levels of damages from rainfall totals in Houston. Based on ERA-5 5-day rainfall totals from 263.875-265.375 East longitude, and 29.125-30.625 North Latitude. Note that the large region is what makes the most extreme rainfall totals more unlikely (compared to reported point totals from Harvey).

Figure 1a shows histograms of 5-day rainfall totals for the 30 year period from 1950 to 1980 (grey) and for the 30 year period from 1990 to 2020 (red) averaged over the broader Houston area. Superimposed are gamma distributions fit to the raw data. Gamma distributions are often used to model rainfall (Wilks, 2011) and although the fit is imperfect, these distributions exemplify the most relevant properties of how rainfall probabilities are thought to shift under warming. In particular, the distribution shifts in the most recent period such that the probability of low rainfall weeks is decreased and the probability of high rainfall weeks is increased (c.f., red and black in Figure 1a and 1b). Return periods and shifts in return periods of the most extreme events are also roughly in line with those calculated from more sophisticated methods.

In this example, the most extreme rainfall bin (above ~8.5 inches) shifts its probability from a once-in-4,300 year event in the 1950 to 1980 period to a once-in-600 year event in the 1990 to 2020 period (Figure 1b). Consistent with more comprehensive studies and basic theory, the more extreme the rainfall, the more the probability shifts between the two distributions. Figure 1c shows the ratio between the probabilities (or equivalently, return periods) between the two distributions, and Figure 1d shows the fraction of attributable risk to human influence on the climate (assuming that 100% of the shift in the rainfall distribution can be attributed to human influence).

Now let’s suppose a “damage function” that estimates the economic damage as a function of rainfall totals. Small rainfall totals cause almost no damage but the largest rainfall totals can cause damage close to 100 billion (as seen in Harvey). A function that mimics this behavior (shown in Figure 1e) is the exponential function with a=1×108 and b=0.8,

Multiplying the economic damage function by the ‘fraction of attributable risk’ gives us the “attributable cost” function (Figure 1f) which emphasizes how extreme the costs attributable to human influence get using this method: If we define an event to be an occurrence of the most extreme rainfall (thick circles in all plots), and that event occurs, then the costs attributable to human influence would be calculated to approach $100 billion.

However, as noted above, even in the most recent time period, this most extreme rainfall total is still a once-in-600 year event. Thus, if you want to estimate the annual average damage from this level of rainfall, you’d have to divide it by 600 to account for its rarity.

Thus, damages estimated from this single event (even if they are magnitude-based rather than frequency-based) could be off by two to three orders if they are interpreted as corresponding to an annual mean value.

More formally, the holistic way to compare the two climate states is to take the difference of the expected values of the damages between the current climate and the preindustrial climate. In doing this, the economic damage function is weighted by the corresponding probabilities and then summed across all values.

Figure 1g shows that as rainfall becomes more extreme, the rarity of the event reduces its impact on the expected value. We see a change in expected value from $276 million per year in the 30 year period from 1950 to 1980 to $393 million per year in the 30 year period from 1990 to 2100 or a total difference of $117 million per year attributable to climate change.

The IAM figure that Frame et al. (2020) claim must be a huge underestimate was $20 billion annual damage in the US from climate change. Comparing $117 million per year to $20 billion per year, is it possible that 0.6% of total us climate damages attributable to human activity comes from rainfall shifts in one of the largest metro areas on the gulf coast? I don’t know, but that calculation does not obviously invalidate the IAM estimate as Frame et al. (2020) claim to do.

I should note that the more the damage function is magnified at higher rainfall totals, the more the ‘attributable cost’ method will overestimate the costs from climate change relative to the more holistic ‘expected value’ method. Thus, a lot of effort should be placed on defining this damage function.

In conclusion, the ‘attributable cost’ method is not appropriate to apply when the ‘event’ must be defined with an arbitrary threshold on a variable that imposes costs on a continuum (rather than flipping costs on or off). It should only be applied if the ‘event’ is a real physical phenomenon that can be said to either ‘occur or not’ or if the threshold is physically meaningful (i.e., enough rain to overwhelm a levee). Applying the attributable costs method likely overestimated the economic damages from Harvey by a factor of about 5 (if the estimates of Wehner, M., Sampson, C. (2021) are taken at face value). Furthermore, if one wishes to compare the numbers produced from these types of analyses to existing annual mean damage estimates (e.g., from IAMs), the damages from extremes have to be weighed by their low probability of occurrence in any given year. In this case, not doing so could cause an overestimate of the annual mean costs of human influence on the climate by two to three orders of magnitude.

This is a very important topic and it’s crucial that the methods used to address it are as accurate as possible. I hope this post serves as constructive criticism and possibly helps inform future research in this area.


Since writing this, I see that several of the authors of the Frame et al. (2020) study have published a paper (Perkins et al., 2022) walking back some of their claims and/or urging caution in interpreting the results of the ‘attributable cost’ method. It looks like the authors of this study would agree with several of the points I have raised here. However, I don’t see an explicit repudiation of the reported results of Frame et al. (2020) or any effort to correct any media coverage of the results.


Allen M (2003) Liability for climate change. Nature 421:891–892

Brown, P. T. (2016) Reporting on global warming: A study in headlines, Physics Today, doi:10.1063/PT.3.3310

Frame, D.J., Wehner, M.F., Noy, I. et al. (2020) The economic costs of Hurricane Harvey attributable to climate change. Climatic Change 160, 271–281. https://doi.org/10.1007/s10584-020-02692-8

Levin, M. (1953) The occurrence of lung cancer in man. Acta Unio Int. Contra Cancrum., 9, 531–541.

Perkins-Kirkpatrick, S. E. et al. (2022) Environ. Res. Lett. 17 024009. https://doi.org/10.1088/1748-9326/ac44c8

Risser, M. D., & Wehner, M. F. (2017). Attributable human-induced changes in the likelihood and magnitude of the observed extreme precipitation during Hurricane Harvey. Geophysical Research Letters, 44, 12,457–12,464. https://doi.org/10.1002/2017GL075888

Oldenborgh GJV, Wiel KVD, Sebastian A, Singh R, Arrighi J, Otto F, Haustein K, Li S, Vecchi G, Cullen H (2017) Attribution of extreme rainfall from Hurricane Harvey, august 2017. Environ Res Lett 12:124009

Wehner, M., Sampson, C. (2021) Attributable human-induced changes in the magnitude of flooding in the Houston, Texas region during Hurricane Harvey. Climatic Change 166, 20. https://doi.org/10.1007/s10584-021-03114-z

Wilks, D. (2011) Statistical Methods in the Atmospheric Sciences. 3rd addition, Elsevier.

Posted in Climate Change | Tagged , , | Leave a comment

Meteorology and Climatology of Wind and Solar Droughts

We have a paper out on a field of increasing importance in meteorology: The analysis of synoptic-scale extreme reductions in wind and solar power energy resources (i.e., wind and solar “droughts”).

If you like talks more than papers here is a talk I gave to Woods Hole Oceanographic Institution on the research.

More details are below but first, some background: Two broad trends make it so that our society’s energy consumption is being impacted progressively more by the weather: 1) Increasingly more and more of society’s energy consumption is coming via electricity (e.g., the electrification of transportation and water/space heating). 2) Increasingly more and more of that electricity is being generated from weather-dependent resources like wind and solar power.

Thus, as wind and solar power resources become more indispensable for society, a full scientific understanding of their droughts may approach the importance of understanding droughts in other resources like precipitation.

This concern has inspired a lot of work on the general topic of the (co)variability of wind and solar power resources but less work has been done on the atmospheric circulation patterns responsible for generating synoptic-scale reductions in wind and solar power resources, both independent of each other and in conjunction.

These extreme weather events have been occurring ‘under the radar’ since the dawn of meteorology but they have not received significant scientific attention because of their historical lack of impact. However, this situation is now rapidly changing as market forces and legal mandates increase the penetration of wind and solar power generation in electricity grids around the world.

This is exemplified in the recent situation in Europe where electricity prices spiked in part because of reduced wind power supply.

In our study, we focus on the meteorological causes of these extreme events over a case study region of western North America with an emphasis on synoptic-scale atmospheric circulations.

We found that perhaps not surprisingly, wind droughts are typically associated with high-pressure systems and solar droughts are typically associated with low-pressure systems and thus wind and solar resources should be anti-correlated at synoptic weather scales.

This along with the fact that wind and solar resources are anticorrelated at the seasonal timescale supports the notion that including both wind and solar power in an energy resource portfolio reduces risk in the same way that diversifying a stock portfolio does.

More specifically, wind drought events over the western US were associated with a thermal mid-level ridge (warm-core high) centered over British Columbia & solar drought events were associated with a thermal mid-level trough (cold core low) off of the west coast of North America.

Furthermore, we found that the synoptic-dynamic meteorology of these phenomena was interpretable through classic Quasi-Geostrophic Theory. This is evidence that these droughts are manifestations of well-known and well-understood weather patterns and are not the result of some unexpected or exotic mechanism. That hopefully means that wind and solar droughts should be as forecastable as any synoptic weather phenomena on daily to seasonal timescales and it suggests that they should be represented reasonably well by courser resolution climate models.

We also find some indication that both wind droughts and solar droughts are slightly associated with positive El Niño and/or positive PDO situations, especially in their most extreme manifestations (Supp Fig. 17 j,k,l).

Finally, this research also supports the notion that there is value in pooling renewable energy resources over areas large enough to encompass the full wavelength of typical Rossby Waves which implies that, in the US, there are significant benefits in moving towards a continental scale “super grid” connected by high voltage transmission.

For more on this research topic see here: https://weatherclimatehumansystems.org/research If you are interested in collaborating on research along these lines please feel free to contact me!

Posted in Uncategorized | 1 Comment

Claims in “How Climate Migration Will Reshape America” vs. observational data

A few people asked me about the accuracy of a recent NY Times Magazine / NY Times Daily Podcast story “How Climate Migration Will Reshape America”. It contains plenty of interesting discussion on e.g., property insurance under a changing climate but it is very loose with the facts on current climate change in the US and it continuously errs on the side of hyperbole and exaggeration.

Below I take a look at observational data corresponding to some of the claims in the article. Overall, I think the data gives a very different impression of the state of climate change in the US than does the rhetoric in the piece.

This is important because this kind of exaggeration undermines the credibility of the NY Times / NY Times Daily Podcast at a time when people really need journalism they can trust.

Furthermore, the issue of human migration under climate change is important and thus it needs to be covered in a serious and fact-based way. We cannot afford for this kind of sloppiness to give people an excuse to dismiss the entire premise.

Most of this piece’s claims are about the future. It is difficult to fact-check claims about the future so I will just look at a few claims that can be easily put in context against historical observational data.

Claim: “August besieged California with a heat unseen in generations.”

Context on CA heat: August 2020 was the fourth hottest month on record in California. It’s average temperature was 79°F. Not as hot as July 2018 (79.6°F), July 2006 (79.3°F), or July 1931 (79.5°F). So it was hot but not “unseen in generations”.

Claim: “I am far from the only American facing such questions. This summer has seen more fires, more heat, more storms — all of it making life increasingly untenable in larger areas of the nation”

Context on fires: The nation has not seen more fires through August but when September numbers are tallied this will indeed go up and likely will set a record. Though it is important to note that there is not an obvious nationwide long-term trend in wildfires.

More context on fires: It is perhaps worth mentioning that global annual area-burned has decreased over the past 20 years. This illustrates that, thus far, climate is not a first-order influence on area-burned and its influence can be totally overwhelmed by other factors. https://www.weatherclimatehumansystems.org/faq-on-fires-humans-and-global-change

Context on US heat: It is true that this summer was hot in the US. It was the 4th warmest on record behind 2011, 2012 and 1936.

More context on heat: Rather than looking at summer means, we can look at indices that measure attributes of heatwaves like the “Warm Spell Duration Index”. Here’s what that looks like for the US. The 1920s and 1930s had higher values that what we have seen recently.

Context on storms: It is not clear what is meant by storms. But maximum 1-day precipitation might be a pretty good proxy for what most people think of as storms. There have been increases in this metric over recent decades but we are not at a historical maxima:

More context on storms: The author could also be referring to tropical storms. Below is a measure of tropical storms making landfall in the United States. We can see some increase in recent decades.

Claim: “Already, droughts regularly threaten food crops across the West…”

Context on droughts: The brown bars are the area in the US that are much more arid than normal. The US West has indeed seen bad droughts in the 2000s and 2010s. More on crops below.

More context on droughts: It is worth noting that we do not see long-term trends in droughts over the US overall.

Claim: “while destructive floods inundate towns and fields from the Dakotas to Maryland”

Context on floods: Here is the long-term change in an index of maximum 5-day precipitation. While not a direct measure of floods, this can be thought of as a rough proxy. By this measure, we are not currently at a historical maxima.

Claim: “Rising seas and increasingly violent hurricanes are making thousands of miles of American shoreline nearly uninhabitable.”

Context on rising seas: Global sea levels have risen about 3 inches since the mid 1990s. This will be a major problem as it continues but does 3 inches plus the change in tropical cyclones shown above amount to “thousands of miles of shoreline nearly uninhabitable”? Sorry, it does not.


Claim: “Let’s start with some basics. Across the country, it’s going to get hot. Buffalo may feel in a few decades like Tempe, Ariz., does today”

Context: This is absolutely false. The average daily high temperature in July in Buffalo NY is near 79°F. The average daily high temperature in July in Tempe Arizona is near 106°F. That’s a 27°F difference.

Buffalo is projected to warm by roughly 2.3°F under a medium emissions scenario by 2050 (and by 5°F by 2100). So that would mean that the author claimed a warming of something like 27°F “in a few decades” when our best estimate is something closer to 10% of that.

Claim: “The Great Plains states today provide nearly half of the nation’s wheat, sorghum and cattle and much of its corn; the farmers and ranchers there export that food to Africa, South America and Asia. Crop yields, though, will drop sharply with every degree of warming.”

Context on crop yields: Historically it has warmed and crop yields have only increased. For one thing, there has been little detrimental climate stress on US crops so far, as measured by indices like the Crop Moisture Stress Index:

Further Context on crop yields: Yields of corn and other crops have only increased globally because changes in technology and agricultural practices have vastly outweighed any negative impact from climate. https://ourworldindata.org/crop-yields

Claim: “It was the kind of thing that might never have been possible if California’s autumn winds weren’t getting fiercer and drier every year”.

Context on CA autumn winds: The most infamous fire-enhancing autumn winds in CA are the “Santa Ana winds”. They are not increasing every year and projections suggest that their occurrence will be less frequent not more frequent under climate change.


Claim: “The 2018 National Climate Assessment also warns that the U.S. economy overall could contract by 10 percent.”

Context: This is a 10% contraction relative to a no-climate change scenario not 10% contraction relative to today. That distinction makes a huge difference. It means that the projection says that if GDP were to increase by 100% without climate change over the next 80 years (very conservative estimate) then climate change would cause GDP to “only” increase by 90%.

Claim: “Once you accept that climate change is fast making large parts of the United States nearly uninhabitable…”

Context: I am sorry but this is just not reconcilable with the data above.

I could go on with other claims made in this piece but I’ll stop here for now. You should get the idea. It paints a picture of current climate change in the US that is very different than the story that is told from looking at the actual observational data and all the errors are in the direction of overstating the negative impact on the US today.

The editors at NY Times Magazine / The Daily Podcast must think that being cavalier with the facts is OK because ‘sending the right message’ on climate change is more important than accuracy. I could not disagree more.

At a time when trust in institutions like the New York Times is faltering, and the right calls them fake news – they cannot afford to confirm that narrative. It makes it too easy for those who want to be dismissive of climate change to feel vindicated in that belief.

Plots above are mostly from: https://ncdc.noaa.gov/cag/ and https://climdex.org/access/

Posted in Uncategorized | 2 Comments

Tipping Points in the Climate System

Lecture on tipping points in the climate system from my frosh, general education level Global Warming course.

Posted in Uncategorized | 1 Comment

Seasonal Prediction of Particularly-Impactful Hot Days

Screen Shot 2020-01-03 at 11.45.26 AM

My talk on this research at the American Meteorological Society’s Annual Meeting.


AGU E-Poster

Posted in Uncategorized | Leave a comment

Net Economic Impact of UN Global Warming Mitigation Targets




Posted in Uncategorized | Leave a comment

This video is a visual explanation of meteorological Skew-T, Log-P sounding diagrams (aka thermodynamic diagrams)

Posted in Uncategorized | Leave a comment

Why is concern about global warming so politically polarized?

As a climate scientist, I often hear it bemoaned that the public discussion of human-caused global warming is so politically polarized (Pew Research, 2019). The argument goes that global warming is simply a matter of pure science and thus there should be no divisions of opinion along political lines. Since it tends to be the political Right that opposes policies designed to address global warming, the reason for the political division is often placed solely on the ideological stubbornness of the Right.

This is a common theme in research on political divides regarding scientific questions. These divides are often studied from the perspective of researchers on the Left who, rather self-servingly, frame the research question as something like “Our side came to its conclusions from pure reason, so what exactly makes the people who disagree with us so biased and ideologically motivated?” I would put works like The Republican Brain: The Science of Why They Deny Science — and Reality in this category.

Works like The Republican Brain correctly point out that those most dismissive of global warming tend to be on the Right, but they incorrectly assume that the Left’s position is therefore informed by dispassionate logic. If the Left was motivated by pure reason then the Left would not be just as likely as the Right to deny science on the safety of vaccines and genetically modified foods. Additionally, as Mooney has argued elsewhere, the Left is more eager than the Right to deny mainstream science when it doesn’t support a blank-slate view of human nature. This suggests that fidelity to science and logic are not what motivates the Left’s concern about global warming.

Rather than thinking of the political divide on global warming as being the result of logic vs. dogma, a much better explanation is that people tend to accept conclusions, be they scientific or otherwise, that support themes, ideologies, and narratives that are a preexisting component of their worldview (e.g., Washburn and Skitka, 2017). It just so happens that the themes, ideologies, and narratives associated with human-caused global warming and its proposed solutions align well with archetypal worldviews of the Left and create great tension with archetypal worldviews of the Right.

The definitional distinction between the political Right and the political Left originates from the French Revolution and is most fundamentally about the desirability and perceived validity of social hierarchies. Definitionally, those on the Right see hierarchies as natural, meritocratic and justified while those on the Left see hierarchies more as a product of luck and exploitation. A secondary distinction, at least contemporarily in the West, is that those on the Right tend to emphasize individualism at the expense of collectivism and those on the Left prefer the reverse.

There are several aspects of the contemporary human-caused global warming narrative that align well with an anti-hierarchy, collectivist worldview. This makes the issue gratifying to the sensibilities of the Left and offending to the sensibilities of the Right.

The most fundamental of these themes is the degree to which humanity itself can be placed at the top of the hierarchy of life on the planet. Those on the Right would be more likely to articulate that it is justified to privilege the interests of humanity over the interests of other species or the “interests” of the planet as a whole (to the degree that there is such a thing). On the other hand, those on the Left would be more likely to emphasize across-species egalitarianism and advocate for reduced impact on the environment, even if it is against the interest of humans.

Within humanity, there are also at least two levels for which narratives about hierarchies influence thinking on global warming. One is the issue of developed vs. developing countries. The blame for global warming falls disproportionately on developed countries (in terms of historical greenhouse gas emissions) and thus proposed solutions often call on developed countries to bear the brunt of the cost of reducing emissions going forward. (Additionally, it is argued that developed countries have the luxury of being able to afford the associated increases in the cost of energy.) Overall, the solutions proposed for global warming imply that wealthy countries owe a debt to the rest of humanity that should come due sooner rather than later.

Those on the Right are more likely to see the wealth of developed countries as being rightfully earned through their own industriousness while those on the Left are more likely to view the disproportionate wealth of different countries as being fundamentally unjust and likely originating from exploitation. Thus, the story that wealthy countries are to blame for the global warming problem and that the solution is to penalize wealthy countries and subsidize poor countries is one that aligns well with preexisting narratives on the Left but not those on the Right. An accentuating factor is the tendency of the Right to be more in favor of national autonomy and thus opposed to global governance and especially international redistribution.

The third level for which hierarchy narratives couple with political divides on global warming relates to the wealth of corporations and individuals. On the Right, the story of oil and gas companies (as well as electric utilities that utilize fossil fuels) is one of innovation and wealth creation: The smartest and most deserving people and organizations found the most efficient ways to transform idle fossil fuel resources into the power that runs society and greatly enhances human wellbeing. Under such a narrative, it is fundamentally unjust to point a finger of blame at those entities (both corporations and individuals) that have done so much for human progress. The counter-narrative from the Left is that greedy corporations and individuals exploited natural resources for their own gain at the expense of the planet and the general public. Under this narrative, policies that blame and punish those in the fossil fuel industry are seen as bringing about a cosmic justice that is necessary for them to atone for their sins.

The other major overlapping theme that defines the divide between the Left and the Right on global warming is the degree to which collectivism is emphasized compared to individualism. Global warming is fundamentally a tragedy of the commons problem in which logical agents act in such a way that ends up being in the worst interest for everyone in the long term. These types of ‘collective-action problems’ almost necessarily call for top-down government intervention and thus they are inevitably associated with collectivism at the expense of individualism. Also, global warming’s long term nature calls for the embracement of collectivism across generations. Again, this natural alignment of the global warming problem with collectivist themes makes the issue much more palatable for the Left than for the Right.

In addition to these fundamental ideological issues, there are a number of more circumstantial characteristics that’s I believe have contributed to polarization regarding global warming.

One is that, in the U.S. at least, Al Gore was the primary actor that brought global warming into the national consciousness. If one wanted the issue to be “non-political” one couldn’t have conceived of a worse person than a former vice president and presidential nominee to be the main flagbearer for the movement.

Also, there is the longstanding claim by those on the Right that the global warming issue is just a Trojan Horse intended as an excuse to bring about all the desired policies of the Left. Books like This Changes Everything: Capitalism vs. The Climate and plans like the Green New Deal do little to dispel this narrative. For example, the Green New Deal Resolution contained the following proposals:

“Providing all people of the United States with— (i) high-quality health care; (ii) affordable, safe, and adequate housing; (iii) economic security; and (iv) access to clean water, clean air, healthy and affordable food, and nature.”

“Guaranteeing a job with a family-sustaining wage, adequate family and medical leave, paid vacations, and retirement security to all people of the United States.”

“Providing resources, training, and high-quality education, including higher education, to all people of the United States, with a focus on frontline and vulnerable communities, so those communities may be full and equal participants in the Green New Deal mobilization”.

These are objectives that clearly seek to satisfy goals of the Left but it is much less clear how directly related these objectives are to global warming.

So, it should really not be particularly mysterious that opinions on global warming tend to divide along political lines. It is not because one side embraces pure reason while the other remains obstinately wedded to political dogmatism. It is simply that the problem and its proposed solutions align more comfortably with the dogma of one side than the other. That does not mean, however, that the Left is equally out-of-step with the science of global warming as the Right. It really is the case that the Right is more likely to deny the most well-established aspects of the science. But, if skeptical conservatives are to be convinced, the Left must learn to reframe the issue in a way that is more palatable to their worldview.

Posted in Uncategorized | 27 Comments

Daily, Seasonal, Annual and Decadal Temperature Variability on a Single Graph

New York Daily Seasonal Annual and Decadal Temperature

The graph above is a record of temperature from 1950-2017 for New York City.

What is unique about this graph is that it shows daily, seasonal, annual and decadal temperature variability on a single Y-axis, revealing how their magnitudes compare.

The daily temperature cycle is represented by the three colored lines in each panel, where red, black and blue represent the daily maximum, daily average and daily minimum for each season and year. For example, the red dots in the far left panel represent the average of all the daily maximum temperatures for the spring of each year.

We can see that in New York, the daily minimum temperature tends to be around 13 degrees C (23 degrees F) lower than the daily maximum temperature.

The annual temperature cycle is illustrated by the variation across the four panels with each panel representing one of the four canonical seasons.

We can see that in New York, the summer tends to be about 22 degrees C (40 degrees F) warmer than the winter.

Interannual temperature variability is illustrated by the year-to-year wiggles in each line.

We can see that in New York, there can be year-to-year swings in temperature (for a given season) of several degrees C. For example, the summer of 1999 had a daily average temperature of 24 C (75 F) and the summer of 2000 had a daily average temperature of 21 C (70 F). It is also notable that year-to-year variability in winter temperature is substantially larger than year-to-year variability in summer temperature.

Decadal temperature changes are represented by the linear trend lines. We can see long term warming which is primarily driven by increases in greenhouse gasses (i.e., this is the local manifestation of global warming). The long term warming is generally more prominent in the daily minimum temperature compared to the daily maximum temperature and more prominent in the winter compared to the summer. In other words, global warming is shrinking both the daily and seasonal temperature cycles.

In terms of absolute magnitude, the seasonal cycle is the dominant mode of variability, followed by the daily cycle, year-to-year variability and finally, long term warming.

Thus, while Global Warming is very pronounced on global spatial scales and centennial and greater timescales, we can see that, thus far, it has had a modest influence on the temperature in New York relative to the typical variability at the daily, seasonal and annual timescales.

Data here is from Berkeley Earth.

Posted in Uncategorized | 1 Comment

El Nino’s influence the upcoming season’s global land temperatures

The El-Niño Southern Oscillation (ENSO) is the preeminent mode of global climate variability on timescales of months to several years. El Niño events cause temporary elevations in global average temperatures, and in the context of background global warming from increasing greenhouse gas concentrations, El Niño events are often associated with setting new global temperature records. El Niños cause warmer than typical global average temperatures because they are associated with a great amount of heat release from the equatorial Pacific to the atmosphere which is then distributed globally. This release of heat also imprints on the structure of the atmosphere and shifts the tendencies of typical atmospheric circulations. In certain locations, advection from climatologically colder locations (e.g., flow from the north in the Northern Hemisphere) becomes more prominent than normal during El Niño events which can cause a local tendency for temperatures to cool during El Niños, despite elevated temperatures globally. The large scale atmospheric circulation is also influenced by the state of ENSO differently depending on the time of the year.

This all means that if you want to translate the state of ENSO into a seasonal forecast (e.g., a forecast for 3-month average temperatures) at a particular location, you have to be careful to examine both the specific relationship between ENSO and climate variability at the location you are interested in as well as how that relationship depends on the time of the year. This is the purpose of the Simple ENSO Regression Forecast (SERF).

The SERF is based on an ensemble of dynamical and statistical model forecasts that predict the future state of ENSO, combined with the historical relationships between the state of ENSO and concurrent local surface air temperature departures from average (as a function of location and time of the year).

At ClimateAi, we are developing considerably more sophisticated machine learning techniques for application to seasonal forecasting that are able to achieve enhanced skill over this simple method. Nevertheless, this simple method is transparent and serves as a useful benchmark for more sophisticated methods to be compared to.

Below is the Simple ENSO Regression Forecast (SERF) for the 2019 Northern Hemisphere summer and Southern Hemisphere winter (June-July-August 2019). A weak El-Niño like state is expected to persist throughout the upcoming season. This translates into an expectation for below normal temperatures over northern/central Canada, the US upper Midwest and much of Russia. Above average temperatures are expected over the US Pacific Northwest, Mexico, much of South America, Africa, India, the Middle East and Europe (see Figure 1 and Figure 2 below). One reason that the tropics shows more consistent warming is that the background global warming has a higher signal-to-noise ratio there which means it is more likely that any given season will be above its 1971-2000 average, regardless of the state of ENSO.


Figure 1. Top) SERF forecast of the average temperature for June-July-August 2019 relative to the long term average (from 1971-2000) for each location. Bottom) Chance that the average temperature over June-July-August will be above the long term average (from 1971-2000) for June-July-August at that location.


Figure 2. Same as the bottom of figure 1 but zoomed in to particular regions.

Posted in Uncategorized | Leave a comment