Assessment of climate change risk to the insurance sector

By Judith Curry

The insurance sector is abuzz with a new report from AIR Worldwide on the insurance risk from the impact of climate change on hurricanes.  Insurance industry clients of my company, Climate Forecast Applications Network (CFAN), have requested a critique of this report.

AIR Worldwide, a respected catastrophe risk modeling and consulting company, has recently published a report Quantifying the Impact from Climate Change on Hurricane Risk.  AIR’s assessment has three components:

  • Hazard component (relates to the frequency and intensity of events)
  • Engineering component (relates to physical assets at risk)
  • Financial component (relates to monetary losses)

The AIR Report purports to “capture the full range of plausible events that could impact an area.”

My critique focuses solely on the hazard component. A summary of my analysis is provided below:

  1. The driver for AIR’s assessment is warming associated with the emissions/concentration scenario RCP8.5, which AIR refers to as a ‘business as usual scenario.’  In fact, RCP8.5 is increasingly being judged as implausible by energy economists, and is not recommended for use in policy planning.
  2. The hurricane risk from climate change focuses on the number and intensity of U.S. landfalls in a changing climate.  Their scenario of the number of major hurricanes striking the U.S. by 2050 is judged to be implausible for medium emissions scenarios such as RCP4.5.
  3. The sea level scenarios used in the AIR Report are higher than the recent IPCC consensus assessments and are arguably implausible for medium emissions scenarios.
  4. The AIR Report ignores the ‘elephant in the room’ that is of relevance to their target period to 2050: the Atlantic Multi-decadal Oscillation (AMO). A shift to the cool phase of the AMO would arguably portend fewer major hurricanes striking the U.S.

For reference on the topics of hurricanes, sea level and climate change, see these reports recently published by CFAN:

  • Special Report on Sea Level and Climate Change  [Download Report]
  • Special Report on Hurricanes and Climate Change [link]


When considering scenarios of future climate change, there is much scope for uncertainty. Plausible outcomes (especially on the high end) are weakly constrained and there is much disagreement among experts as to the likely range of outcomes (67% likelihood).

Outcomes of future climate change are associated with deep uncertainty [for further context, see my paper on Climate Uncertainty and Risk].  Under conditions of deep uncertainty, probabilities of outcomes are not very meaningful. At best, we can hope to plausibly  bound the range of outcomes.  The IPCC seeks to bound the likely range [67%] of outcomes.

While experts will inevitably disagree on what constitutes a plausible best or worst case scenario when the knowledge base is uncertain, I have developed a classification  based on the extent to which borderline impossible parameters or inputs are employed in developing the scenario [see my essay What’s the worst case?]. This classification is inspired by the Queen in Alice in Wonderland: “Why, sometimes I’ve believed as many as six impossible things before breakfast.” This classification articulates three categories of worst-case scenarios:

  • Conceivable worst case: formulated by incorporating all worst-case parameters/inputs  into a model; the outcome does not survive refutation efforts.
  • Possible worst case (borderline impossible). Includes multiple worst-case parameters/inputs in model-derived scenarios; the outcome survives refutation efforts (at least temporarily).
  • Plausible worst case:  Includes at most one borderline implausible assumption in model-derived scenarios.

This classification is used here in evaluating the plausibility of AIR’s scenarios out to 2050.

How much warming?

AIR’s assessment of the climate change impact on U.S. hurricane risk is driven solely by warming from manmade emissions. Specifically, the AIR Report focuses on emission/concentration scenario RCP8.5.

While the emissions/concentration scenario RCP8.5 is often referred to as a ‘business as usual’ scenario, it is in fact an extreme scenario that is implausibly high. RCP8.5 pathways are driven by: very high population growth, very high energy intensity of the economy, low technology development, and a high level of coal in the energy mix. Wang et al. (2017) and Ritchie and Dowlatabadi (2018) challenge the bullish expectations for coal in the RCP8.5 scenarios, which is counter to recent global energy outlooks and exceeds today’s known conventional reserves. Burgess et al. (2020) further highlight the implausibility of the RCP8.5 scenario owing to contradictions in the assumptions used in constructing the scenario. Pielke and Ritchie (2020) concluded that RCP8.5 is systematically misused in applications of climate model simulations, particularly for policy making purposes. Ritchie and Dowlatabadi (2018) recommend that RCP8.5 not be used as a benchmark for policy studies.

The 2019 World Energy Outlook Report from the International Energy Agency (IEA) challenges the near-term RCP scenario projections through 2040 [link]. The IEA examined three scenarios: a current policy scenario where no new climate or energy policies are enacted by countries; a stated policies scenario where Paris Agreement commitments are met; and a sustainable development scenario where rapid mitigation limits late 21st century warming to well below 2°C. The IEA projections through 2040 are close to the RCP4.5 scenario (a moderate emissions scenario) and much lower than for RCP8.5.)

On timescales to 2050, I would classify RCP8.5 as implausible and RCP4.5 as the likely scenario.  In terms of a plausible worst case emissions/concentration scenario to 2050, the forthcoming IPCC AR6 is also using a scenario equivalent to RCP7.0.  However, given that likely projections of emissions out to 2050 are plausibly close to RCP4.5, it arguably makes more sense to use your one ‘borderline implausible’ assumptions on future hurricane activity and sea level rise under the RCP4.5 scenario.

The significance of rejecting the RCP8.5 in projections of insured losses out to 2050 is this. Any changes to hurricane activity that are dependent on increasing sea surface temperatures would be substantially muted (relative to RCP8.5).  More significantly, the greatest uncertainties in 21st century sea level rise projections are associated with possible large instabilities in the West Antarctic ice sheet arising from the highest temperature projections driven by RCP8.5. By eliminating RCP8.5, the highest sea level rise projections are eliminated.

Problems with the RCP8.5 scenario have been discussed in numerous posts at my blog Climate Etc., with references  to the primary literature as well as background information:

Citations for additional recent publications:

Amount of warming

With regards to the amount of warming associated with different emissions scenarios, projections from the CMIP5 climate models are provided in the table below [from IPCC AR5 Table SPM.2]. Substantially more warming is produced by the RCP8.5 scenario relative to RCP4.5.  The range in the projections for a given RCP scenario arises from the sensitivity of different climate models to warming from increasing emissions (primarily CO2).

It is important to understand that the temperature projections provided by the IPCC (from the CMIP climate model simulations) are not predictions of actual outcomes. Rather, they should be regarded as sensitivity analyses relative to increasing emissions. These projections neglect any changes in natural climate variability that would influence actual climate outcomes in the 21st century.

When considering scenarios of natural climate variability for the 21st century, there are reasons for thinking that the CMIP5 simulations may be predicting too much warming for the 21st century, even for the more plausible RCP4.5 emissions scenario:

  • Observed warming for the past two decades is less than the average rate of warming simulated by climate models. [link]
  • The ensemble of CMIP5 climate models do not sample the full range of likely values of equilibrium climate sensitivity to increasing CO2, neglecting the lowest 20% of the likely range from the IPCC AR5. [link]
  • Climate model projections do not include solar variability and volcanic eruptions, associated with plausible scenarios for a cooling effect in the 21st century. [link]

Specifically considering the amount of warming associated with the RCP4.5 scenario, my assessment is that temperature change  is very unlikely to exceed the upper bound of the IPCC AR5 likely range.

It is noted here that the CMIP6 temperature projections for the forthcoming IPCC AR6 are coming in warmer than CMIP5/AR5, with about half of the models having very high sensitivity to CO2.  It will be interesting to see how the AR6 evaluates this.  Topic for a blog post for another day.


AIR uses a sophisticated method to select scenarios from 100,000 stochastic simulations of individual hurricane seasons.

By 2050, AIR predicts a 15% increase of Cat 3, a 25% increase in Cat 4 and a 35% increase in Cat 5 (Cat 1 and 2 numbers are unchanged).  Not only does the AIR projection result in an increased proportion of major hurricanes, but also an increase in the total number of major hurricanes.

It is difficult to decipher exactly how these numbers were determined.  Figure 2 in the AIR report is based on Figure 7 from the GFDL statement.  GFDL’s text accompanying their Figure 7 is cited below:

“The Bender et al. (2010) study projected a significant increase (+90%) in the frequency of very intense (category 4 and 5) hurricanes using the CMIP3/A1B 18-model average climate change projection. Subsequent downscaled projections using CMIP5 multi-model scenarios (RCP4.5) as input (Knutson et al. 2013) still showed increases in category 4 and 5 storm frequency (Fig. 7). However, these increases were only marginally significant for the early 21st century (+45%) or the late 21st century (+39%) CMIP5 scenarios (based on model versions GFDl and GFDN combined). That study also downscaled ten individual CMIP3 models in addition to the multi-model ensemble, and found that three of ten models produced a significant increase in category 4 and 5 storms, and four of the ten models produced at least a nominal decrease. While multi-model ensemble results are probably more reliable than individual model results, each of the individual model results can be viewed as at least plausible at this time. Based on Knutson et al. (2013) and a survey of subsequent results by other modeling groups, at present we have only low confidence for an increase in category 4 and 5 storms in the Atlantic; confidence in an increase in category 4 and 5 storms is higher at the global scale.”

The AIR Report cited a recent publication by  Kossin et al. (2020)  which provides an analysis of satellite-derived intensities of global tropical cyclones for the period 1979-2017.  The analysis was divided into two periods: Early (1979-1997) and Late (1998-2017).  The paper’s key result  is that there is greater probability of a tropical cyclone reaching major hurricane status (Pmaj) in the Late versus the Early period, reflecting a global average increase of about 5% per decade.  When broken down by individual ocean basin regions, this increase is driven by the North Atlantic, the South Pacific and South Indian Oceans.  Notably, the West Pacific (the region with largest number of tropical cyclones) shows a substantial decrease in the probability of reaching a major hurricane.  Shown below is the main table from Kossin et al. (2020; a corrigenda to this table in the original publication was issued).

The AIR Report also cites the recent assessments from  the World Meteorological Organization (WMO) Expert Team on Tropical Cyclones (Knutson et al. 2019b). This report concluded that 2oC of warming (consistent with RCP4.5 in 2100, or RCP8.5 at mid-century) is projected to impact hurricane frequency and intensity as follows:

  1. For hurricane intensity (maximum wind speed), there is 
medium-to-high confidence that the global average will increase. The median projected increase in lifetime maximum surface wind speeds is about 5% (range 1–10%).
  2. For the global proportion of hurricanes that reach Category 4–5 levels, there is at least medium-to-high confidence in an increase, 
with a median projected change of +13%.

Author opinion in Knutson et al. (2019b) was more mixed and confidence levels lower for the following projections:

  1. A decrease of global hurricane frequency, as 
projected in most studies
  2. An increase in the global number of very intense hurricanes (Cat 4–5).

Individual experts and individual model simulations have supported scenarios as extreme as that used by AIR for a warming of 2C, although the plausibility of a large increase in number (rather than proportion) of Cat 4/5  is disputed by experts.  At this point, I would conclude that such extreme outcomes for a 2C temperature increase cannot be falsified based on our background knowledge, although they are well outside of the likely range according to the IPCC and WMO assessments.  The implausibility of the AIR scenario rests on the extreme temperature projection from RCP8.5, and borderline implausibility of a large increase in the number of Cat 4/5 hurricanes.

Missing the elephants in the room

In projecting a substantial increase in U.S. major hurricane landfalls for the period to 2050 as a result of global warming, the AIR Report ignores two elephants in the room:

  1. The historical record for major hurricanes striking the U.S. that shows elevated activity prior to 1970, when sea surface temperatures were significantly cooler.
  2. The dominant role of the Atlantic Multidecadal Oscillation (AMO) in determining overall Atlantic hurricane activity, but particularly the number of major hurricanes.

The most striking aspect of any analysis of Atlantic hurricanes and U.S. landfalls is the large interannual to multi-decadal variations.  The Figure below (from NOAA AOML) shows the yearly values since 1850, for  major Atlantic hurricane counts. While the number of major hurricanes prior to 1944 is probably undercounted, it is noteworthy that the number of major hurricanes during the 1950’s and 1960’s was at least as large as for the last two decades.

The Atlantic Multidecadal Oscillation (AMO) has a strong influence on Atlantic hurricane activity, particularly the number of major hurricanes.  Since 1995, the AMO has been in a warm phase (associated with elevated Atlantic hurricane activity.)  The previous cool phase of the AMO (1970-1994) was associated with few major hurricanes.  The previous warm phase (1926-1969) was also associated with a high number of major hurricanes, althought sea surface temperatures were substantially cooler than the present warm phase of the AMO.  [For further information, see section 4.3.1 from my Report on Hurricanes and Climate Change].

By looking at the period from 1979-2017, you can see an obvious and large increase in the number of major hurricanes (which was identified by Kossin et al. 2020).  However, any attempt to relate the large trend since 1979 to global warming must account for the large number of major hurricanes during the 1950’s and 1960’s that was at least as large as the recent numbers which  occurred in the presence of significantly lower sea surface temperatures.  Simply extrapolating the trend from 1979-2017  (and assuming it was caused by increasing sea surface temperatures) produces a future increase in Atlantic major hurricanes that is clearly unjustified when the longer data record is examined.

With regards to actual U.S. landfalls,  Klotzbach et al. (2018) have conducted a comprehensive evaluation of the landfalling hurricane data for the Continental U.S. since 1900.  The figure below shows shows the time series for major hurricane landfalls (Category 3-5). The largest year in the record is 2005, with 4 major hurricane landfalls. However, during the period 2006 through 2016, there were no major hurricanes striking the U.S., which is the longest such period in the record since 1900.

Table 6.1 shows a list of the strongest U.S. landfalling hurricanes. The strongest storm on this list occurred in 1935.  Of these 13 storms, 10 occurred prior to 1970.

While higher sea surface temperatures can improve the thermodynamic environment for hurricane intensification, Cat 4 and 5 storms can and have formed at much cooler surface temperatures in the Atlantic. With the exception of Hurricane Andrew and the Florida Keys Hurricane, each of these exceptionally strong hurricanes occurred during the warm phase of the AMO.

Scenarios to 2050

Because of the large interannual to multi-decadal variability in Atlantic hurricanes, it is very difficult to detect any signal from global warming in the historical record.

The largest increase in Category 4-5 Atlantic hurricanes is predicted by Bender et al. (2010). Owing to the large interannual to decadal variability of SST and hurricane activity in the basin, Bender et al. estimated that
 detection of an anthropogenic influence on intense hurricanes
 would not be expected for a number of decades, even assuming a very large underlying increasing trend (+10% per decade).

Given the dominant influence on Atlantic hurricanes of the Atlantic Multidecadal Oscillation (AMO), arguably the single most important factor for the next 30 years would be a shift to the cold phase of the AMO.   The timing of a shift to the AMO cold phase is not predictable; it depends to some extent on unpredictable weather variability. However, analysis of historical and paleoclimatic records suggest that a transition to the next cold phase is expected prior to 2050. Enfield and Cid-Serrano (2006) used paleoclimate reconstructions of the AMO to develop a probabilistic projection of the next AMO shift. Their analysis indicates that a shift to the cold phase should occur within the next 15 years, with a 50% probability of the shift occurring in the next 6 years.

The AMO not only influences the number of major Atlantic hurricanes, but also the preferred location of U.S. landfalls.  The average number of U.S. landfalling hurricanes in the previous cool phase of the AMO (1970-1994) is 1.24 per year, compared with 1.7 per year during the current warm phase (1995-2019). Florida and North Carolina showed markedly fewer hurricane landfalls during the previous cool phase of the AMO.

The timing of a future shift to the cold phase of the AMO remains uncertain.  Whether a future cold phase would have a comparable distribution of landfalls also remains uncertain.  However, a scenario of reduced U.S. landfalling hurricanes between 2020 and 2050 is justified by empirical evidence from the historical record in context of a possible (or even likely) shift to the cold phase of the AMO prior to 2050.  This scenario of lower U.S. landfall activity to 2050 is arguably at least as likely as the scenario put forward by AIR.

Sea level rise scenarios

The AIR report uses a 2017 NOAA Report on Sea Level Rise Scenarios for the U.S. The NOAA Report included an Extreme sea level rise scenario of 2.5 m by 2100, whose primary rationale was a paper by DeConto and Pollard (2016) that suggested increased likelihood of extreme sea level rise from Antarctic ice sheet instability.

The AIR Report selects two scenarios from the NOAA Report to bound its sea level rise estimates: Intermediate Low (0.24 m by 2050) and Intermediate High (0.44 m by 2050). The AIR Report includes local projections for New York City, Miami and Houston.  Projections from the NOAA Report (since 2000) are provided below for these four cities.  The black curve reflects observations through 2016.  Comparing the slopes of the observations with the NOAA scenarios indicates the observed sea level rise tracking most closely NOAA’s Low and Intermediate-Low scenarios.

The year 2017 (when the NOAA Report was published) arguably marked the peak influence of the DeConto and Pollard (2016) paper, with subsequent analyses backing off from this extreme scenario of Antarctic ice sheet instability to occur during the 21st century [link].  Most significantly, subsequent IPCC assessments are not producing exceptionally high projections of sea level rise, even for the RCP8.5 scenario. .

The 2019 IPCC Special Report on Oceans and Cryosphere in a Changing Climate (SROCC; Chapter 4)  provides sea level rise projections for the period 2031-2050, and 2046-2065.  To compare with the values used by AIR for 2050, the SROCC sea level rise projections are averaged here for the two mid century periods: for RCP4.5, 0.16 to 0.28 m; for RCP8.5, 0.19 to 0.33 m.  From what I have seen in the Second Order Draft, the sea level rise projections for the forthcoming AR6 are coming in lower than the SROCC values.

The sea level rise scenarios referred to in the AIR Report are quite high, with the intermediate-high sea level rise scenario being well outside the likely bounds of the RCP8.5 scenario in the IPCC SROCC Report (2019).  Relative to the more plausible RCP4.5 scenario for 2050, the AIR Report’s low end scenario of 0.24 m is near the top of the SROCC likely range, while AIR’s high end scenario of 0.5 m is well outside of the likely range even for RCP8.5; this large value is arguably implausible for RCP4.5.

So, what is the plausible worst case scenario for sea level rise by 2050? The role of instability in the Antarctic Ice Sheet on the plausible worst case scenario for 21st century sea level rise  is a very fast moving area of climate research, with much uncertainty and disagreement among experts.  However, once the high levels of warming associated with RCP8.5 are eliminated from consideration, then extreme instability of the West Antarctic ice sheet and large values of sea level rise (e,g. the Intermediate-High scenario) become increasingly unlikely, if not implausible, for  moderate rates of warming associated with RCP4.5, especially prior to 2050.Conclusions

Not surprisingly, the combination of implausibly high sea level rise projections and increase in the number of major hurricane landfalls in the U.S. results in a projection of substantial increases in damage and losses.  From the AIR Report:

“The results of the analysis show that increased event frequency and sea level rise will have a meaningful impact on future damage. The growth in the number of stronger storms, and landfalling storms overall, increases modeled losses by approximately 20%, with slightly larger changes in areas such as the Gulf and Southeast coasts where major landfalls are already more likely today. The loss increases extend to inland areas as well, as stronger storms may penetrate farther from the coast.

The impacts from sea level rise, using the analysis of storm surge for New York, Miami, and Houston suggests that by 2050, sea level rise may increase storm surge losses by anywhere from one-third to almost a factor of two, with larger impacts possible when combined with increases in the number of major storms. The results suggest that an extreme surge event in today’s climate may be twice as likely to happen 30 years from now.”

While predictions for global climate change out to 2100  are  weakly constrained  and highly uncertain, we have a much better idea of the constraints on likely outcomes out to 2050. It is expected that variations in the Atlantic Multidecadal Oscillation will continue to dominate the statistics of Atlantic hurricanes for the next several decades, with any signal from global warming being difficult to discern amidst the natural variability.

The plausible worst case scenario has an important role to play in some decision making frameworks.   However, the plausible worst case scenario requires assumptions that are justified and not falsifiable based on our background knowledge.  The scenarios for climate change out to 2050 presented by AIR are based on an implausible emissions scenario that produces an implausible amount of warming by 2050.  Even if this large amount of warming is accepted as plausible, the scenario for substantially increased  numbers of major hurricane landfalls impacting the U.S. is judged to be very unlikely if not borderline implausible.  The intermediate-high sea level rise scenario in the AIR Report extends well beyond the likely range of the most recent IPCC assessment, even for RCP8.5.

Apart from the slow creep of sea level rise which in recent decades has been tracking the low end of the RCP4.5 scenario, there is little justification for expecting a noticeable increase in insured losses associated with U.S. landfalling hurricanes by 2050.

This article appeared on the Climate Etc. website at https://judithcurry.com/2021/02/15/assessment-of-climate-change-risk-to-the-insurance-sector/


Join our effort to build a greener future.

  • This field is for validation purposes and should be left unchanged.