by Rud Istvan
Here is a recent example of artful lack of disclosure in the climate change debate, on the possible negative impacts of climate change on food security.
The following chart is Figure 13 on page 28 of the 2011 NRC booklet, Warming World: Impacts by Degree.
The National Research Council is sponsored by the National Academy of Sciences, whose motto is ‘where the nation turns for independent, expert advice’. The chart’s message is dire if temperatures do rise by 3 degrees as the 4th IPCC assessment says likely. Such a rise, for example, would mean US corn yields might decline by half. That would massively disrupt US food supplies, and would cause starvation elsewhere since the US produces 40% of the world’s corn.
The NRC’s 2010 website[ii] contained the following version of the same chart, which they credited to the National Science Foundation. It is uncertain whether their misspelling of Africa was also credited to the NSF.
NRC’s 2010 website said the NSF chart’s information comes from table 5-4 of the 3rd IPCC, or more recent studies. Working Group 2 Table 5-4 runs several pages and summarizes many studies modeling both positive and negative crop yield impacts. WG2 ¶184.108.40.206 says, “In 2/3 of the cases, temperate crop yields benefit at least some of the time.” That is not shown at all by the NRC chart although US maize is grown in a temperate climate. The NSF chart is labeled as plotting the single worst of all modeled estimates. Not disclosing this information makes the booklet chart extremely misleading.
The 2011 NRC booklet itself is worse than misleading. Text accompanying the chart says, “Solid lines show best estimates”. That makes it overtly false; the worst modeled outcome is re-characterized as the best consensus estimate. And the single US maize statistical study portrayed by the NSF chart is itself false.
The worst US maize (corn) AGW model is a 2009 PNAS paper. [iii] This massive statistical analysis assessed US heat extremes (defined as continuous 24 hour days at or above some temperature) using US county level crop and reconstructed weather data from 1950-2005 for corn, soybeans, and cotton. It contains 105,981 observations for corn. Using an 8th order polynomial equation, it found a statistical threshold at around 29°C, above which yields were increasingly affected by extreme heat days. The statistical results were graphed to suggest that continuous 24-hour exposure to 40ºC (104ºF) reduces corn yield at least 5%, even though there were zero 40ºC days in the data. In Africa the same effect was later statistically found using detailed CYMMIT field trial data to be only 1% per day, arise above 30°C, and occur during only anthesis. [iv]
The PNAS finding is superficially plausible. All plants become heat stressed above some threshold temperature (growth ceases), and are killed by prolonged exposure to some higher temperature. [v] These thresholds vary by plant and cultivar, and also vary over the plant’s annual growth cycle. Heat stress during anthesis is well known to reduce cereal yields. Physiological responses are similar to those for drought stress, so depend on water availability and transpiration. Hot low humidity days (plus insufficient soil moisture) produce the greatest cereal heat stress. It is therefore usually drought associated. Plants can develop transient thermotolerance in hours if sufficient water is available. At least partial recovery of lost growth after heat stress is proven in sorghum and tomatoes. As a result, whether heat stress is cumulative is unknown (or perhaps indeterminate, since depending on too many factors like degree of stress, duration of stress, transpiration losses, soil water availability, and post stress recovery). The maize heat stress threshold (determined experimentally in greenhouses) during the seedling stage is about 35°C, and during anthesis is about 38°C, in both cases with sufficient water available. Values vary by maize cultivar.
Adding the IPCC AR4 global warming estimates to the historical temperature distribution data, the PNAS paper used its statistical 29°C threshold (rather than the experimental >35-38°C) to calculate that by 2100 there would be up to 15% of days above 35ºC, with above a 2.5% negative impact. Assuming a cumulative effect, the paper used its statistics to model that this warming would cause up to 60-70% decline in yield by 2100 under the higher warming scenarios. This is easy to verify. The US maize season is from April-August (hotter southern states) or May-Sept (cooler northern states). That is about (5*30) 150 days. If 15% were above the statistical 29°C threshold averaging a negative 2.5% daily cumulative impact, then the impact would be (150*0.15*-2.5) -56%.
The PNAS maize forecast can be examined for veracity using the previously known experimental maize heat stress details and a revised subset of the paper’s data (averages by state from 1980) placed by its now famous authors into the public domain. (Equivalent data for soybeans and cotton was not provided to the public, so the critique is limited to maize.) This enables visual parametric scrutiny of the paper’s veritas without using any statistics at all. Simply visually compare the data to the paper’s statements.
“Regression Models. We assume temperature effects on yields are cumulative over time and that yield is proportional to total exposure.” That assumption is not supported by prior experimental heat stress facts. More telling, if heat effects were cumulative, Kansas should always have lower yields than Nebraska, and Nebraska should always have lower yields than the other states. That is obviously not true in the paper’s own dataset (just compare the right and left charts), which directly show that maize heat stress cannot be cumulative. This detail was not available to peer reviewers, who only had the statistical results and not the later released graphical dataset. The basic model used to compute the statistics is flawed. So is the yield conclusion.
The PNAS paper’s flaws go much deeper. It found that “Greater precipitation partly mitigates damage from extreme high temperatures”. That is consistent with experimental heat stress details. But such an interaction term was not included in the model. Even though disclosed as a real co-linearity, the possible rainfall/temperature interaction was expressly omitted for ‘statistical’ reasons:
However, omitting temperature-rainfall interactions will not bias predictions of average effects of temperature and rainfall, as we do not find a significant correlation between temperature outcomes and precipitation outcomes…
There is no reason to think temperature and rain outcomes would ever be covariant. It rains when it is hot, and it rains when it is cold.
But this omission rationale is logically flawed—corn really cares about the coincidence of hot and wet. That is well known experimentally. The information was readily available before the paper’s regression model was formulated. For example, the FAO publishes extensively for farmers around the world. (Maize has become the staple grain in parts of Africa.) According to the FAO, “The crop is very sensitive to frost, particularly in the seedling stage but it tolerates hot and dry atmospheric conditions so long as sufficient water is available to the plant and temperatures are below 45°C.”
The charted data show yield declines for all states in some years with distinctly more > 29ºC days (like 1980, 83, 88, and 2002), but not in others. These particular years are well-known US drought years. The sharp Ohio decline in 2002 compared to neighboring Indiana (both with about the same >29ºC days) is specifically attributable to local Ohio drought. Abnormally hot years without drought such as 1990, 92, 95, 2000, 01, and 06 did not have similar yield declines. (Dry Kansas after 2000 has had to curtail irrigation due to depletion of the Ogallala aquifer.) The negative yield effect occurs mainly in hot + dry years, not in hot + normal rain years, coherent with experimental heat stress.
The published data also show that average yields in recent hot + dry years like 2003 or 2006 were still better than yields in earlier cool + wet years like 1981 or 1990. Even Kansas’ worst recent yield in hot, dry 2006 was about 115 bu/acre, around the national average from 1985-95. The PNAS paper said,
“with wide geographic range in average yields, and with a three-fold increase in yields over the sample period (56 years)…[soil quality, technological change, and precipitation]… have strong statistical significance not reported here.”
These much more important factors were statistically ‘removed’ to focus just on temperature. Other future developments (new hybrids) may be much more important than climate change to future yields. This possibility was even acknowledged in the PNAS paper:
“greater heat tolerance still may be possible if greater returns for such innovation arise. Recently, a National Science Foundation- funded study completed a draft sequence of the corn genome, which might make it easier to develop new corn varieties with greater heat tolerance”
In fact, by 2010 the USDA ARS had identified at least 4 independent thermotolerant genetic traits for future hybridization. And in December 2011 the FDA and USDA approved Monsanto’s first genetically modified drought resistant corn, MON87460 with 6-10% yield improvement. The PNAS paper’s past data cannot be used for projecting long-term future yields without incorporating the other information ‘not reported’. But that is what NSF did, which NRC concealed.
The NSF chart truthfully reported one portion this paper’s erroneous statistical conclusion. The problem is that the paper’s data do not support it. The key model assumption is not true, a key known heat stress relationship was deliberately excluded, and the future influence of more significant yield factors (hybrid corn improvement) was not considered.
The NSF chart was ‘hearsay’ even before being mischaracterized by the NRC. Harvard President Lawrence Lowell supposedly wrote in 1909 that statistics, “…like veal pies, are good if you know the person that made them, and are sure of the ingredients.” This is still true a century later.
[iii] Schlenker and Roberts, Nonlinear Temperature Effects, PNAS.0906865106
[iv] Lobell et. al., Nonlinear Heat Effects on African Maize, Nature Climate Change 1: 42-45 (2011)
[v] Wahid et. al., Heat Tolerance in Plants: a Review, Environmental and Experimental Botany 61: 199-223 (2007)
Bio notes: Rudyard L. Istvan is Chairman and CEO of Third Stream Bioscience, Inc., commercializing a novel topical antimicrobial licensed from P&G. He is also the inventor behind and principal of NanoCarbons LLC, which has developed the world’s best carbon for capacitive energy storage. It is now licensed to a major European company. He holds a summa cum laude in economics from Harvard College, a JD cum laude from Harvard Law School, and an MBA from Harvard Business School as a Baker Scholar. He has a book Gaia, Musings on Sustainability and another book Arts of Truth in process.
JC comment: Yesterday Rud Istvan sent me a section from his forthcoming book Arts of Truth, I invited him to do a guest post on this. I am particularly interested in the general topic of unwarranted alarmism on climate change impacts. This is a guest post, and the views presented here are those of Rud Istvan, and implies no endorsement by me of any particular statements made in this post.