Climate Change for 24 US Cities with Temperature Record going back to 1895 PowerPoint Presentation

Slide 1 -

Climate Change US Cities Gaetan Lion, September 26, 2022

Slide 2 -

Climate Change at the US City level Mining the climate data from the National Oceanic and Atmospheric Administration (NOAA), I uncovered 24 cities that have yearly temperature records going back to 1895. The temperature data is extremely volatile. In order to uncover the underlying trend of such data, and smooth out the volatility, I took the following steps: I calculated the 10 year moving average on a yearly basis. So the first data point is in 1904 reflecting the average temperature over the 1895 to 1904 period; I took the 10 year difference in such temperatures as calculated in the first step. So, now the first data point is in 1914 reflecting the decadal difference in temperature between 1914 and 1904; Next, I used a LOESS Regression that instead of fitting a straight regression trend line, fits a curving line that fits this non linear data set much better. LOESS allows one to observe the changing trend over time (see M vs. V patterns later in this presentation) and the most relevant recent trend over the most recent 10 years. 2

Slide 3 -

The Cities with data going back to 1895 24 cities from 20 different States, including three in New York and four in Texas. 3

Slide 4 -

4 How does the Intergovernmental Panel for Climate Change (IPCC) look at temperature? The IPCC looks at global yearly temperature data as the anomaly or increase over the average temperature during the 1850 – 1900 period. It indicates that current temperatures are about 1 degree Celsius above that level. And, that to maintain a relatively unimpaired environment we should maintain this temperature increase to =< 1.5 degree Celsius by the end of this century. This seems extremely challenging at best. The IPCC indicates that a rise in temperature greater than 2 degrees would be already somewhat adverse for the environment. And, an increase near or above 3 degrees could be even more adverse. The IPCC has generated scenarios where temperatures could potentially increase to 4, 5, or even 6 degrees Celsius by 2100. And, such temperature increases would be associated with progressively more devastating impact on the environment.

Slide 5 -

5 A look at US cities temperature history Focusing on the 24 US cities, I used as a baseline the average over the 1895 – 1920 period because the NOAA at the city level does not go further back than 1895. Also, the 1895 – 1920 period did not show much temperature increase. This framework is relatively comparable to the IPCC at the global level. In view of the above, it is concerning that 18 of the 24 cities have already experienced temperature increases greater than 1.5 degree Celsius; 11 over 2 degree Celsius; and even 1 over 3 degree Celsius. And, that is right now! All mentioned temperature increases are very likely to be much greater by 2100. Data source: NOAA

Slide 6 -

6 Ranking cities by most recent 10-year temperature increase using LOESS estimates I grouped the cities in 6 different categories in terms of most recent temperature increase over the past 10 years (using LOESS smoothed estimates). And, I also looked at if this specific temperature increase appeared to be rising, decreasing, or flat. When I indicated “Decreasing?”, the interrogation mark suggests there is much uncertainty about the trend. 10-year most recent temperature increase using LOESS est. Data source: NOAA

Slide 7 -

7 Temperature increase estimates by 2100 Here I simply take the LOESS estimated temperature increase over the most recent past 10 years (also shown on the previous slide). And, I assume that this temperature increase pace remains constant until 2100. And, I calculate the temperature increase since the onset or the average over the 1895 – 1920 period. For instance, for the top line for Fresno, the temperature increased already by 2.2 degrees Celsius since the average of 1895 – 1920. the LOWESS estimates for the most recent 10 years indicate that temperatures keep on rising by 0.89 degree Celsius per decade. And, if we project this current pace of temperature increase, it results in an estimated temperature increase of 9.2 degree Celsius by 2100 since the 1895 – 1920 baseline period. As shown, this would describe a devastating scenario for Fresno’s environment. As described only 3 cities would remain within reasonably benign scenarios (blue and green zone). All other cities would be associated with progressively more adverse to ultimately devastating scenarios as you move up from yellow to the deep red zone. Data source: NOAA

Slide 8 -

8 Temperature increase between now (2021) and 2100 We are looking at four different forecasts or scenarios of temperature increase between now and 2100 measured in degree Celsius. The first forecast is based on our LOESS estimates, the second one is based on just projecting average long term historical trends. The next two forecasts are from the NOAA Climate Explorer that weight the average of 32 Climate Change models output. They model the yearly average daily maximum temperature. The NOAA models forecast at the county level, and aggregate US cities within such counties. The NOAA Higher forecast is associated with a continuation of greenhouse gas emissions. The NOAA Lower forecast assumes emissions peak in 2040, and drop very quickly thereafter. I feel the divergence in measurements do not affect the relevance of the comparison between the LOESS and NOAA models. Data source: NOAA

Slide 9 -

9 Focusing on the two most interesting scenarios I eliminated two scenarios because they appear highly unlikely. I eliminated the “Historical trend” scenario because its long term trend (linear) was biased downward by the effect of SO2 emissions that caused a rapid drop in temperature between 1935 and 1970 (more info on this phenomenon a bit later in this presentation). So, this scenario fails to capture contemporary trend. I eliminated the NOAA Lower scenario as its assumption regarding greenhouse gas emissions peaking in 2040, and rapidly declining thereafter is too optimistic. We will study the divergence between the LOESS and NOAA models later in this presentation. Data source: NOAA

Slide 10 -

10 Observing the historical temperature data in degree Fahrenheit with LOESS estimates and Confidence Intervals to get an idea of the underlying trend, including directional shifts over time (up or down, decreasing or declining)

Slide 11 -

11 Anatomy of a graph The volatile black line depicts the 10 year change in temperature (Fahrenheit), based on 10 year moving averages. The blue line reflects this 10 year change smoothed out using LOESS regression estimates. The light blue zone around the blue line depicts a 0.9 or 90% Confidence Interval around the LOESS data point estimates. Data source: NOAA

Slide 12 -

12 Three different trend patterns: M, V, J M J V The 24 cities temperature increase trends typically fall into one of those three trend patterns: M, V, J. Data source: NOAA

Slide 13 -

13 All patterns show a common decline in temperature from about 1935 to 1970 because of a rise in sulfur dioxide (SO2), and an increase in temperature after 1970, in part, because of a decline in SO2. SO2 emissions result from volcanoes eruptions, forest fires, coal electricity generation, petroleum extraction, and manufacturing. SO2 emissions cause acid rain and ozone layer depletion. In the 1960s and 70s, the US, Canada, and Europe passed environmental regulations that rapidly curbed SO2 emissions within these regions.

Slide 14 -

14 Source: armstrongeconomics.com Source: “The Problem with Sulfur Dioxide”. September 9, 2016. cullycleanair.org SO2 emissions have rapidly declined in the U.S. But, the distribution of SO2 emission is very uneven. Although, higher SO2 emissions appear to be concentrated in the Midwest, South, and East; they also appear concentrated around cities elsewhere. However, disparity in local SO2 emissions may explain some of the divergence in cities temperature increase patterns since 1970.

Slide 15 -

Greater than 1 Degree Fahrenheit over 10 years (as specified) and Rising 15 These cities are in very serious trouble with rapidly rising temperatures. They are already relatively hot with average temperatures of 67, 71, and 74 degrees F respectively over the past 5 years (2017 – 2021) vs. 56 degrees F for the 24 cities average. If temperature continues rising at the current pace over the next 40 years, Fresno’s temperature could rise by 6.4 degrees Fahrenheit. The two Texan cities temperature would rise by about 4.4 degrees. These estimates assume that the temperature increase continue at current pace. The LOESS estimates trends suggest temperatures could rise faster than the current pace. This is an alarming prospect. Data source: NOAA

Slide 16 -

> 0.75 < 1.00 Degree Fahrenheit over 10 years (as specified) and Rising 16 These cities are also in trouble with rising temperatures. Their respective average temp. over the past 5 years (2017 – 2021) are: Albany 51, Baton Rouge 70, New York 56 F. If temperatures keep rising at the current pace, these cities would experience a rise in temp. from 3.0 to 3.8 Fahrenheit. But, the rising LOESS estimates trends suggest the temperature increases could be much higher. That is a concerning outlook. Data source: NOAA

Slide 17 -

> 0.75 < 1.00 Degree Fahrenheit over 10 years and Flat trend 17 Burlington has a couple of positive factors including a relatively cool average temperature over the past 5 years (48 degrees). And, the LOESS estimates suggests that the current trend in temperature increase is flat. Still, the current pace of temperature increase would result in a 3.3 degree Fahrenheit rise over the next 40 years. Burlington would still remain a very livable city. However, the economies of nearby ski resorts may be impaired during the Winter season. Data source: NOAA

Slide 18 -

> 0.50 < 0.75 Degree Fahrenheit over 10 years (as specified) and Rising 18 The three cities have average temperatures of 48, 51, and 62 F respectively. Their respective temperature increase at current pace, as specified, is 0.73, 0.63, and 0.53 F over 10 years. Fortunately, the warmer of the three cities (Greenville) is associated with the lower pace of temperature increase. At current pace, Greenville’s temperature could increase by 2.1 degree Fahrenheit over the next 40 years. The other two cities could see increases in the 2.5 to near 3.0 range. However, of concern is that all temperature increases could accelerate when focusing on the LOESS estimates trends. Data source: NOAA

Slide 19 -

> 0.50 < 0.75 Degree Fahrenheit over 10 years (as specified) and Flat 19 Roswell has an average temperature of 64 degrees F over past five years. Buffalo… a temperature of 51 degrees F. Based on current pace, both cities may experience a temperature increase of about 2.4 degrees Fahrenheit. Both trends are relatively flat. The prospective temperature increases appear manageable. Data source: NOAA

Slide 20 -

> 0.30 < 0.50 Degree Fahrenheit over 10 years (as specified) and Decreasing? 20 These cities are interesting because the trend in the LOESS estimates (decreasing) appear divergent from the underlying data. This is especially true for Reno where the most recent 10 years show a rise in temperature > 1.5 degree F and rising rapidly relative to earlier data points. But, the LOESS estimates are associated with a temperature rise < 0.50 over the same period. Both cities have temperate averages of 56 and 47 degrees F, respectively. Based on current (LOESS) pace, their temperature could rise by a moderate 2 degrees F over the next 40 years. However, the LOESS curves could turn back up in the next few years and suggest a far more rapid temperature increase (especially for Reno). Data source: NOAA

Slide 21 -

> 0.30 < 0.50 Degree Fahrenheit over 10 years (as specified) and Decreasing 21 These cities are less exposed to climate change. Over the next 40 years, their respective temperature may increase by about 1.3 degree Fahrenheit or less, given that the LOESS estimates are trending downward. Data source: NOAA

Slide 22 -

> 0.30 < 0.50 Degree Fahrenheit over 10 years (as specified) and Rising 22 Cheyenne has a mild average temperature of 48 degrees F over the past five years. Its current pace of temperature increase is a moderate 0.37 degree F based on LOESS estimates. However, when looking at the underlying data, it is close to 1 full degree F. The rising trend, and the divergence between LOESS estimates and underlying data render prospective rise over the next 40 years rather uncertain. Just projecting the two different paces (LOESS vs. underlying data), temperature could rise between 1.4 and 3.6 degrees F over the next 40 years. Data source: NOAA

Slide 23 -

> 0.15 < 0.30 Degree Fahrenheit over 10 years (as specified) and Decreasing 23 These three cities do not appear overly exposed to climate change. Based on current pace, they could incur a 1 degree Fahrenheit increase over the next 40 years… or less given the decreasing trends in the LOESS estimates. Data source: NOAA

Slide 24 -

> 0.15 < 0.30 Degree and Flat or Rising 24 Spokane is another city associated with much uncertainty. Its LOESS estimates suggests a modest 0.23 degree F rise per decade, or less than 1 degree Fahrenheit over 40 years. But, if you look at the underlying data the most recent 10 year change in temperature is about 1.4 degree F. This would suggest a very broad range in prospective temperature increase over the next 40 years from 1 to 6 degrees F… ranging from the benign to the alarming. Data source: NOAA

Slide 25 -

< 0.15 Decreasing or Rising? 25 Another couple of interesting cities. The LOESS estimates trends suggest that these two cities will not experience much if any temperature increase over the next few decades if the current pace and LOESS trends persist. However, the underlying data shows rapidly rising temperatures with a current pace over past 10 years close to 1 degree Fahrenheit or 4 degrees over the next 40 years. Data source: NOAA

Slide 26 -

< 0.10 Degree Fahrenheit over 10 years (as specified) and Decreasing 26 Aberdeen is also associated with a bit of uncertainty. The LOESS estimates trend suggests an ongoing small decline in temperature going forward. The underlying most recent data suggests an increase of about 0.35 degree Fahrenheit over 10 years. That would translate into a 1.4 degree F increase over the next 40 years. Data source: NOAA

Slide 27 -

27 Revisiting a comparison between the LOESS and the NOAA Higher forecast for temperature increase out to 2100 vs history Data source: NOAA

Slide 28 -

28 Explaining the NOAA Higher forecast within The Climate Explorer graph Source: NOAA The NOAA Higher forecast is represented by the red line. It captures the weighted average aggregation of 32 separate climate models that forecast temperatures F at the county level. This is the forecast I use throughout this presentation. For further explanation of all other elements of this graph, please refer to the next slide that discloses the NOAA’s own explanation.

Slide 29 -

29

Slide 30 -

30 LOESS vs. NOAA forecasts vs. history: Aberdeen vs. Fresno Based on historical and current trends, there is a huge divergence between the two cities. Aberdeen temperatures appear to clearly decline. Meanwhile, Fresno’s are exploding upward. The LOESS model reflects this divergence out to 2100 with Fresno’s temperature rising by 7 degree Celsius, and Aberdeen’s decreasing by – 1.3 degree Celsius. Meanwhile, the NOAA models completely miss out on this divergence and actually forecasts that Aberdeen’s temperature will grow faster than Fresno’s (+ 4.6 C, + 3.8 C, respectively). Data source: NOAA

Slide 31 -

31 LOESS vs. NOAA forecasts vs. history: Corpus Christi vs. San Antonio The two cities have archetype patterns of M and J (the M reflects most recent 10 year periods temperature increase slowing down; and J such temperatures accelerating upward). The NOAA models miss much of this large divergence as they have Corpus Christi and San Antonio temperatures increasing by similar amounts by 2100 at + 3.4 and + 4.5 degree Celsius respectively. Meanwhile the LOESS model fully captures this divergence with + 0.9 C for Corpus Christi vs. + 5.0 C for San Antonio. Notice that the LOESS and NOAA models are fairly close regarding San Antonio: + 5.0 C vs. + 4.5 C, respectively.

Slide 32 -

32 LOESS vs. NOAA forecasts vs. history: Great Falls vs. El Paso Another classic comparison between an M and a hybrid J/V shape. Great Falls show a declining rate of temperature increase in the most recent period. El Paso shows an ongoing rapid increase in such temperatures. The LOESS model captures that divergence with + 1.1 C for Great Falls and + 4.6 C for El Paso. Meanwhile, the NOAA models make little difference between the two with both around + 4.3 C to + 4.4 C over the same period. Notice that the LOESS and NOAA models are very close regarding El Paso: + 4.6 C vs. + 4.4 C, respectively. Data source: NOAA

Slide 33 -

33 LOESS vs. NOAA forecasts vs. history: Lexington vs. Albany Nearly same comment as for Great Falls vs. El Paso. LOESS reflects the divergence in trends (M vs. hybrid M/V). It estimates the temperature increase at + 1.2 C for Lexington and + 4.1 C for Albany. Meanwhile, the NOAA models make little difference between the two cities (about + 4.3 C to 4.4 C). The LOESS and NOAA models are very close with Albany: + 4.1 C vs. + 4.4 C, respectively. Data source: NOAA

Slide 34 -

34 LOESS vs. NOAA forecasts vs. history: Minneapolis vs. Baton Rouge Nearly same comment as for Great Falls vs. El Paso. LOESS reflects the divergence in trends (M vs. V). It estimates the temperature increase at + 1.4 C for Minneapolis and + 3.5 C for Baton Rouge. Meanwhile, the NOAA models gets the divergence between both cities in the wrong direction with Minneapolis coming in much higher at + 4.7 C and Baton Rouge much lower at + 3.4 C. The LOESS and NOAA models have almost the same estimate for Baton Rouge: + 3.5 C and + 3.4 C, respectively. Data source: NOAA

Slide 35 -

35 LOESS vs. NOAA forecasts vs. history: Cape Hatteras vs. New York In most recent periods, the temperature increase is decelerating for Cape Hatteras and accelerating for New York. The LOESS model captures that divergence with + 1.5 C for Cape Hatteras vs. + 3.4 C for New York. The NOAA models do not capture much of this divergence with + 3.3 C for Cape Hatteras vs. + 4.2 C for New York. Data source: NOAA

Slide 36 -

36 LOESS vs. NOAA forecasts vs. history: Cheyenne vs. Portland At the outset, temperatures in Portland are rising much more consistently than in Cheyenne. The LOESS model captures that nuance with Cheyenne coming in at + 1.6 C and Portland at + 3.2 C. The NOAA models do not differentiate much between the two with Cheyenne at + 4.5 C and Portland at + 4.4 C. Data source: NOAA

Slide 37 -

37 LOESS vs. NOAA forecasts vs. history: Greenbay vs. Greenville This is a classic M vs. V pattern. The LOESS correctly estimates that the temperatures are rising more slowly for the M than for the V. Greenbay comes in at + 1.8 C and Greenville at + 2.3 C. Meanwhile, the NOAA models miss out the correct divergence between the M and the V pattern with Greenbay coming in faster at + 4.7 C and Greenville slower at + 3.9 C. Data source: NOAA

Slide 38 -

38 The NOAA models just make very little differentiation between the 24 different cities (or counties) regardless of the marked divergence in the historical data. A simple LOESS model makes far more differentiation between the cities.

Slide 39 -

39 LOESS estimated temperature increases are in average a lot lower than the NOAA. But, the LOESS estimates are far more disperse with a far wider range of forecasts. Meanwhile, the NOAA forecasts all converge closely to their average of 4.2 degree Celsius. Data source: NOAA

Slide 40 -

40 Why are the two models forecasts so different? They have a very different structure

Slide 41 -

41 LOESS model strength More often than not, the LOESS model does capture the historical trend, and the most relevant contemporary trend in the data. And, its forecasts clearly differentiate between two cities diverging trends (M vs. V shape). We saw this comparison between the two cities earlier. And, the LOESS model estimated that temperatures out to 2100 would rise more slowly in Minneapolis (+ 1.4 C) vs. Baton Rouge (+ 3.4 C). This seems like a reasonable forecast. Data source: NOAA

Slide 42 -

42 LOESS Model Weaknesses LOESS regression can overfit the historical data. LOESS regression is not well catered to long term forecasting (very few models are). This LOESS model seems in some cases to have a lagging effect, whereby its underlying regression trend curve does not seem to capture the actual trend in the recent actual data points. For Astoria and Aberdeen, the LOESS model suggests that the recent trend curve is associated with either very low temperature increase or even negative temperature change over the most recent 10 years. But, the actual data shows very rapid change in temperature. Projecting this LOESS trend out to 2100 could underestimate temperature increase by a huge amount. As structured, this LOESS model does not use any exogenous greenhouse gases variables. Therefore, it can’t be sensitized to different greenhouse gases emissions scenarios. Data source: NOAA

Slide 43 -

43 NOAA model(s) strengths The NOAA model can sensitize the forecasts to different greenhouse gases emission scenarios. Using this 32 model approach, it allows to generate very interesting uncertainty bands around the weighted average estimates. This makes for a valuable substitution to Confidence Intervals around the model’s estimates. This framework is associated with spectacular data visualization. Source: NOAA

Slide 44 -

44 NOAA Model(s) Weaknesses Even though it is currently shown at the NOAA website as its most current version, this model relies on historical data that is already over 15 years out of date. Since then, temperature trends have changed a lot. Even more puzzling is that NOAA has the most recent historical data going out to 2021. In view of the above, it is not surprising that the NOAA forecasts does not reflect contemporary historical trends. Going back to Minneapolis and Baton Rouge, even though Baton Rouge’s temperature has been rising much more rapidly than Minneapolis since right around 2006 (when the NOAA model historical data stops), this model forecasts that Minneapolis temperature will rise faster than Baton Rouge ( + 4.7 C vs. + 3.4 C, respectively). Overall, this model makes little differentiation between the various counties/cities. See box plot to the right. For all its model complexities (aggregation of 32 models), the NOAA could have just about taken the average + 4.2 C out to 2100, and it pretty much would have reflected the forecast for any of the mentioned cities. Data source: NOAA

Slide 45 -

45 NOAA Model(s) Weaknesses continued If you look closely at the historical visual data (for New York in this specific case), the NOAA model does not appear to fit it well. The NOAA does not disclose the weighted average estimates during the history which should have been represented by a black line. The NOAA just showed the minimum and maximum of their 32 models. If the NOAA model fit was good, the minimum and maximum estimates should be symmetric around the historical temperature data points. They are often not [symmetric]. This is a reflection of a poor historical fit. Source: NOAA

Slide 46 -

46 Considerations Complex does not mean better. The NOAA model(s) fails on a couple of critical counts. First, it does not appear to fit the historical data well enough to make for relevant forecasts. Second, it does not differentiate much between various counties/cities. The LOESS model, as mentioned, is associated with all the limitation of LOESS. It may overfit the historical data, and it is not really appropriate to generate long term forecasts (as stated, few models are). Nevertheless, this far simpler LOESS model appear to fit the historical data better, and differentiates far more between cities than the NOAA model(s). This is not a ratification of the LOESS methodology, but more of a criticism of the NOAA model(s) that needs to go through a complete revision to ensure resolving the mentioned deficiencies.