In collaboration with co-authors Andy Rhines, Martin Tingley, and Peter Huybers, I recently published a paper regarding long-lead predictions of hot days in the Eastern United States. You can find the paper here.
During the summer of 2016, I made experimental predictions for hot days using the methodology outlined in the above paper. A discussion of those predictions, and what actually occurred, can be found below.
After a bit of a late start, I am now publishing predictions for the current summer. If time allows, I will update the prediction figures so that prior predictions are also visible. However, for now, the predictions will be presented in the same format as last year.
Below is a time series of the predicted probability of an Eastern US hot day based on the most recent SST anomalies from the NOAA OISSTv2 dataset. Note that the climatological probability of a hot day is 0.15, and that predictions are not made more than 50 days in advance.
The predictions are made based on the most recent SST anomaly in the midlatitude Pacific:
A discussion of summer 2016 predictions
So, how did 2016 go? First, a quick recap of how hot days were defined for the purposes of prediction. First, we calculated a metric called T95 that was defined as the spatial 95th percentile of temperature anomalies in the Eastern US -- in other words, T95 provided the temperature anomaly that was exceeded by the hottest 5% of the Eastern US. Then, a hot day was defined as one when T95 was greater than 6.5° Celsius, or 11.7° Fahrenheit.
Many readers may recall that 2016 felt like a very hot summer, with many warnings about heat domes and suggestions to stay inside. However, from the perspective of T95, there were only two "hot days" during the summer: June 27, and July 23. And these days barely made the cutoff, with values of T95 of 6.87°C and 6.53°C respectively. Overall, the summer of 2016 had an average T95 of 4.53°C, which is just below the average across the prior 34 summers of 4.60°C.
Why didn't summer 2016 pop out as a very hot summer using our metric? The main reason is that T95 only takes into account daily maximum temperature, not relatively humidity. Many of the heat warnings over the summer were related to the heat index, which is "is a measure of how hot it really feels when relative humidity is factored in with the actual air temperature." This summer highlighted the importance of including humidity in future prediction studies, although historical relative humidity data are much more limited than for temperature, which is why the original study focused only on temperature. One other factor to keep in mind is that our predictions are for anomalies on top of a linear trend in order to try to remove effects of climate change. This formulation was intentional, because we wanted to make sure that we were not simply developing a model that could predict a trend, rather than intraseasonal variability in hot weather. That said, the Eastern US has not shown large increases in summer daily maximum temperature over the past 35 years (see my recent paper here).
The image below (similar to Supplemental Figures 17-50 from the original paper, which can be found here) summarizes the predictions (upper panel) as well as the observed values of T95 (lower panel). The cool colors in the upper panel indicate predictions for lower than average probability of hot days, whereas the warm colors indicate higher than average probability of hot days as a function of lead time (vertical axis) and date predicted (horizontal axis). For most of the summer, the predictor tended to be slightly negative, but rarely deviated too much from average values.
The lower panel shows the observed values of T95. The gray dashed line indicates the cutoff for a hot day. As noted above, two non-consecutive days during the summer exceeded this threshold slightly, but overall the summer was average.
To conclude, the predictions for 2016 were relatively accurate. The predictions generally suggested that the summer would have slightly lower-than-average probabilities of hot days, and the rate of hot days was 3% (2 out of 60), which is lower than the climatological rate (15%, or 9 out of 60).
That said, the summer also emphasized short-comings of the method. First, the predictions were not designed to predict an elevated heat index, and so did not provide warning of the humid weather that plagued much of the East Coast during peak summer. Second, the predictions are for a relatively large region, and so were not able to predict the record warmth in the Northeast during August.
Questions, comments, thoughts? mckinnon (at) ucar (dot) edu