By Andy May
The Old Farmer’s Almanac has been making yearly long-term weather forecasts for 230 years. We pay attention to them because they are normally 80% accurate. They did not do as well last winter but were 72% in predicting the direction of temperature change, and 78% accurate in the change in precipitation. This is pretty remarkable because while the U.S. weather forecasts are 90% accurate five days in advance, they are only 80% accurate seven days out. The Old Farmer’s Almanac forecasts are far less specific, they only predict the direction of change, but their forecasts are for twelve months in the future, quite impressive. Figure 1 is their forecast for the lower 48 United States, for this winter.
Their forecast is by season, not day-by-day and they only predict temperature and precipitation direction of change relative to the average for the area. The lower 48 United States are divided into the 16 areas shown on the map in Figure 2. If their forecast is correct, we are in for another cold winter this year.
While they predict the temperature changes for many locations in each region numerically, they only grade themselves on their predictions of the change in direction of temperature and precipitation by region. This is a lower bar than NOAA uses, but still significant. Remarkably their 80% accuracy target is nearly always met over the past 230 years.
How do they do this? Consider that the IPCC, even after spending billions of dollars, has never accurately predicted climate changes—and they admit it. Besides missing the amount of global warming, they have often missed the direction for significant periods of time. The IPCC wrote in AR6, their latest report:
“There is medium confidence that most CMIP5 and CMIP6 models overestimate the observed warming in the upper tropical troposphere over the period 1979–2014, in part because they overestimate tropical SST warming.” (page TS-37)
“… all report that CMIP6 models on average overestimate warming from the 1970s or 1980s to the 2010s, although quantitative conclusions depend on which observational dataset is compared against…” (page 3-15)
“That overestimated warming may be an early symptom of overestimated equilibrium climate sensitivities (ECS) in some CMIP6 models (Meehl et al., 2020; Schlund et al., 2020)” (page 3-15)
“… studies continue to find that CMIP5 and CMIP6 model simulations warm more than observations in the tropical mid- and upper-troposphere over the 1979-2014 period (Mitchell et al., 2013, 2020, Santer et al., 2017a, 2017b; Suárez-Gutiérrez et al., 2017; McKitrick and Christy, 2018” (Page 3-24)
“… some studies suggest that climate sensitivity also plays a role (Mauritsen and Stevens, 2015; McKitrick and Christy, 2020; Po-Chedley et al., 2021). Hence, we assess with medium confidence that CMIP5 and CMIP6 models continue to overestimate observed warming in the upper tropical troposphere over the 1979-2014 period by at least 0.1°C per decade, in part because of an overestimate of the tropical SST trend pattern over this period…” (page 3-24)IPCC AR6 quotes
It’s like shooting fish in a barrel and it gets old after a while.
Suffice it to say, the Old Farmer’s Almanac, with its 230-year-old algorithm, while not as specific as the IPCC forecasts, does better and has a longer track record. Ross McKitrick and John Christy tested the AR5 models versus observations in the tropical troposphere and nearly all of them failed the test in a statistically significant way (McKitrick & Christy, 2018).
In a 2021 presentation to the Irish Climate Science Forum in Dublin, John Christy presented his evaluation of the AR6 models, it is shown in Figure 3. The historical or hindcasted model results are displayed from 1975 through 2014. The forecasted results are from 2015 through 2020, these are steadily up in temperature. Using the weather balloon data (light green) we can see temperature went up in 2015, this is also true of the UAH satellite data for the tropics in that year.
The weather sonde and reanalysis data ends in 2016, so if we substitute the UAH lower troposphere satellite data for the tropics, we see that temperatures went down in 2016 and 2020. In other words, directionally, the IPCC was only correct four out of six years in AR6, a success rate of 67%.
In Figure 3, notice the direction of the AR6 modeled temperature change is wrong from 2005 to 2010 and 1995 to 2000, even though these years were hindcasted. That is, the modelers knew the answer and still got it wrong. The red, yellow-filled boxes are the average of numerous climate models, and the spread of model results is ±50% or more. Further the observations are half (-50%) of the prediction.
Six years of predictions are not a lot. Thanks to Ross McKitrick and John Christy we also have their analysis of the AR5 models, which forecast temperatures from 2010. The plot from their paper is shown in Figure 4.
The model projections in Figure 4 are only from 2010 to the present, or eleven years. In this period, they get the direction of temperature change correct 8 times or 73% of the time. They hindcasted the period from 2005 to 2010 and got everyone of those years wrong, which tells us a lot. It is interesting that the spread in model results between AR6 and AR5 has not narrowed. It is generally acknowledged that the direct effect of doubling CO2 is about one degree of warming, which is modest. The true debate is over the possible feedbacks to that warming. Some think they could be net negative, especially in the tropics (Lindzen & Choi, 2009), and the IPCC thinks they are net positive. However, the uncertainty in the net feedback has not narrowed in AR6, in fact, it is slightly larger, as illustrated in Figure 5.
As always, the largest uncertainty in the feedback are the effects of clouds, cloud feeback varies from slightly negative (cooling) to positive (warming). As you can see on the left side of Figure 5, the net uncertainty in AR6 is larger than in either the CMIP5 models or the CMIP6 models. Billions of dollars and we know less than we did when AR5 was published in 2013.
The Old Farmer’s Almanac Methodology
The Old Farmer’s Almanac tries to predict the direction of change in both precipitation and temperature. The IPCC model ensemble, over two short forecast periods, achieves an accuracy of 67% to 73%. In 230 years of forecasts, the Old Farmer’s Almanac is normally 80% accurate. How do they do this?
Their primary metric has always been solar activity, as measured by sunspots. The predominant input they use in their forecasting model is the solar cycle and it has been the most important input for 230 years. They have found that the solar cycle strongly affects weather teleconnections, these are longer-term weather patterns such as the North Atlantic Oscillation (NAO), the Pacific/North American Index (PNA), and ENSO (La Niña/El Niño). Teleconnections are very large-scale weather patterns that can sometimes have global weather effects. For example, El Niños and La Niña’s in the equatorial Pacific strongly affect North American weather. What the forecasters at the Old Farmer’s Almanac do, is forecast the changes to the 32 teleconnections they monitor using the state of the solar cycle as a guide. Then they use the teleconnection predictions to build their forecast.
The teleconnection predictions and the predicted impact of each are guided by their 230 years of detailed records. For more on how Old Farmer’s Almanac uses teleconnections in their forecast see this blog post by Mike Steinberg.
The methods used by the Old Farmer’s Almanac stem from an algorithm developed by Robert B. Thomas in 1792, when George Washington was still president. It has obviously been refined over the years, but the basics are still the same. The staff assure us that folklore, such as acorns, persimmons, apples and wooly caterpillars are not used in their process. Thomas firmly believed that changes in solar activity, as revealed through sunspots, are the primary influence on our weather. Figure 6 is a portrait of Robert Thomas.
Like NOAA, the Old Farmer’s Almanac uses 30-year climate temperature and precipitation averages as the basis for their predictions. NOAA’s averages are often called “climate normals.” So, when the Old Farmer’s Almanac says warmer than normal or colder than normal, the normal they are referring to is NOAA’s most recent 30-year average.
The IPCC has concluded that solar variability is unimportant for predicting climate change since, in their view, it just goes up and down every 11 years or so and has no long-term trend. This idea is discussed here, especially see the third figure, it is from the AR5 IPCC report and shows they assume the long-term change in solar output is zero to 0.1 W/m2. Thus, we have a 230-year solar-based algorithm, with a good track record, pitted against a relatively new algorithm that ignores solar variability, costs billions of dollars, and has a poor track record over the past 30 years.
“The Old Farmer’s Almanac‘s long-range forecasts are based predominantly upon solar activity, with their basis being that changes in activity on the Sun do indeed directly cause changes in weather patterns on Earth.”
As Steinberg mentions, the IPCC and other agencies do not believe that solar variability matters, however, the Old Farmer’s Almanac believes changes in the Sun control our climate to a large degree. He cites recent work by Russian meteorologists that postulate that very small changes in the Sun can affect Earth’s thermosphere, and that these changes can work their way down to the troposphere and affect our weather.
The thermosphere and the mesosphere are quite high, and their air density is quite low, so the effect on the troposphere is probably small, but the stratosphere definitely affects our longer-term weather, as past wobbly polar vortexes can attest (Kretschmer, et al., 2018). Solar UV varies more than total solar irradiation, and as solar UV output changes so does the stratosphere.
So, we have a 230-year-old seasonal weather forecasting method that depends upon solar activity and historical records of the effects of weather teleconnections at different stages of the solar cycle. In opposition, we have a billion-dollar set of complex computer models that assume the Sun doesn’t matter and greenhouse gases control the climate (Lacis, Schmidt, Rind, & Ruedy, 2010). The 230-year-old method has a good record, and the billion-dollar models, not so much. The billion-dollar models can’t even get it right when they know the answer!
Criticism of the Old Farmer’s Almanac and the younger Farmer’s Almanac (a different publication) generally revolves around their use of solar variability to make the forecast. Meteorologists, such as Marshal Shepherd, will say there is no evidence that solar activity influences weather. Perhaps, but if so, why are the Old Farmer’s Almanac predictions better than the IPCC predictions when compared in an apples-to-apples comparison, as we have done above? Critics of the IPCC forecasts suggest they ignore evidence that solar variability is important (Connolly et al., 2021).
The Old Farmer’s Almanac accuracy has dropped a bit over the last decade, and is now in the range 72% to 79%, but they are working to get back to their normal 80% range. Either way, they are clearly more accurate than the billion-dollar IPCC climate models when forecasting trends a year in advance or in the past. The magazine is only nine dollars. No wonder it is an Amazon best seller.
Look at it another way, if the Old Farmer’s Almanac has a better forecast record, for 230 years, than the IPCC has for 30 years, doesn’t that tell you something? What is the IPCC doing with the billions of dollars we give them? Their forecasts are almost straight lines, why not replace them with an Excel spreadsheet? Once I made a mistake in predicting the performance of a gas well, my boss said he was going replace me with a pair of dice. He didn’t thankfully, but he had a valid point. I had set my forecasting target and missed it, a model’s value is determined by its track record.
Download the bibliography here.