Atmosphere science Book Review Hazards Physical oceanography

Beat the Heat: Predicting Eastern U.S. Hot Days using the Pacific Ocean

Article: McKinnon, K. A., Rhines, A., Tingley, M. P., & Huybers, P. (2016). Long-lead predictions of eastern United States hot days from Pacific sea surface temperatures. Nature Geoscience.

Background

You know those summer days that seem so much hotter than the rest, where you don’t want to move and can’t get cool without an air conditioner? Those conditions are harmful to farms too, with high temperatures robbing crops of water. Cities and towns also face difficulties as large groups of people become dehydrated or suffer heat stroke and need medical attention. What if you could predict when those hot days would come, maybe even a whole month in advance? Current seasonal temperature forecast models do a mediocre job of this. They’re fine for getting the gist of a season’s temperature range, but do poorly at predicting these anomalous heat events any further than a week ahead of time. The authors of the study we’re focusing on today are attempting to improve those models (at least for the Eastern U.S.) through new analyses of the relationship between terrestrial rainfall, Pacific Ocean surface temperatures, and high heat anomalies.

Methods

Defining Extremely Hot Days

This study begins by examining summer (June 24 to Aug. 22) temperatures in the Global Historical Climatology Network database form 1982 to 2015. An algorithm was used to define regions of the U.S. where temperature recording stations all experienced abnormally hot days at the same time. These regional divisions vary depending on how many divisions they want, but they settled on 5 as it provides a realistic sense of the U.S. climate zones (Fig. 1). They then focused on the Eastern region, as it includes both a large amount of farmland and several population centers.

To define an extremely hot day, the researchers began by looking at the warmest 5% of temperature stations on a given day and how high above the total (1982-2015) average those stations were. If those warmest 5% stations were at least one standard deviation above the average (a statistical term determining anomalous values), the day was tagged as an extreme heat day for the region.

Figure 1: Shows the region definitions used in the study. These are termed: the maritime west coast (purple), the arid interior west (yellow), the semiarid and subtropical southwest (pink), tropical Florida (green), and the humid and continental eastern U.S (blue).
Figure 1: Shows the region definitions used in the study. These are termed: the maritime
west coast (purple), the arid interior west (yellow), the semiarid and subtropical southwest (pink), tropical Florida (green), and the humid and continental eastern U.S (blue). (McKinnon et. al. 2016)

Predicting Extremely Hot Days

The team first looked at the feasibility of using the Standardized Precipitation Index (SPI) to predict hot days. This measurement of precipitation over a region can give clues about how much moisture is in the soil, which can influence how the atmosphere retains or loses heat. Using this data, if the running SPI sum over 30 days fell below the average SPI, a heat event was predicted 30 days later with a 71% true positive rate, and a 46% false positive rate (meaning a heat event was predicted but did not occur). Figure 2a displays this prediction skill in a Relative Operating Characteristic (ROC) curve, where the area under the curve indicates a model’s prediction skill. The dashed line indicates where a model would be if it predicted an event correctly just as often as it did incorrectly (making it as useful as coin flip). The higher a model’s curve is above that line, the better it is at usefully predicting heat events. You can see in Figure 2a other lines where the model was used to try and predict heat events closer or further away than 30 days, and how the skill decreases the further into the future it tries to predict. This model performed reasonably better than similar models in use today. In terms of an odds ratio, if the running SPI fell below the average, there was a three-fold increase in the probability of a heat event 30 days later.

ROC curves for both the SPI and PEP prediction models. The higher the curve above the dashed line, the more useful the model is as a prediction tool.
Figure 2: ROC curves for both the SPI and PEP prediction models. The higher the curve above the dashed line, the more useful the model is as a prediction tool. (McKinnon et. al. 2016)
Figure 3: The Pacific Extreme Pattern (PEP) at lead times of 50 (a), 40 (b), 30 (c), 20 (d), 15 (e), 10 (f), 5 (g) and 0 (h) days. The PEP itself consists of the horizontal blue-red-blue pattern across the green box, indicating cooler-than-average, warmer-than-average, cooler-than-average waters. The black lines are atmospheric pressure indicators, which showed scientists how the ocean affected weather patterns that then moved East across the country.
Figure 3: The Pacific Extreme Pattern (PEP) at lead times of 50 (a), 40 (b), 30 (c),
20 (d), 15 (e), 10 (f), 5 (g) and 0 (h) days. The PEP itself consists of the horizontal blue-red-blue pattern across the green box, indicating cooler-than-average, warmer-than-average, cooler-than-average waters. The black lines are atmospheric pressure indicators, which showed scientists how the ocean affects weather patterns that then moved East across the country. (McKinnon et. al. 2016)

The team next tried to use Pacific Ocean sea surface temperatures to predict hot days in the Eastern U.S. These two locations may seem disconnected, but the Pacific Ocean may influence weather patterns that travel across the country and then affect eastern terrestrial temperatures. In this analysis, daily average sea surface temperatures at various times ahead of heat events in the climatology database were averaged over the dataset (i.e. pick out the ocean temperature field 30 days before each hot day, then average all those). The scientists found that a pattern emerged, visible up to 50 days before a heat event and traceable right up until the day of the heat event (Figure 3), which they termed the Pacific Extreme Pattern (PEP). The pattern consists of an alternating sequence of cooler-than-average, warmer-than-average, then cooler-than-average waters in a certain region of the Pacific, shown in the green box in Figure 3. Using the occurrence of the PEP to predict hot days during the same period used before performed similarly to the precipitation predictor (Figure 2b). When they applied the PEP prediction to a pre-1982 database of ocean and terrestrial temperatures, the model also performed well.

Findings

These prediction systems perform well, and much better than seasonal time span temperature models. They also perform much better than established atmospheric variability models that have been used to predict anomalous hot days. In the future, the researchers want to combine these two predictive models to see if they can improve the accuracy even further. Being able to better predict these extreme heat events a month ahead of time can allow cities to prepare to handle the heat, farmers to add more water to crops or otherwise shelter them, and let the public plan their lives accordingly.

Leave a Reply

Your email address will not be published.