Much of the Western US finds itself in the middle of a late-season heatwave for the holiday weekend. As meteorological autumn begins, just how much longer will it last? Let’s turn to medium-range ensembles to help answer that question.
The Lead Up
Ensembles gave us a good long heads up on the heat last week. The first/left image below is the Extreme Forecast Index (EFI) from the ECMWF ensemble from last Sunday night, valid for Labor Day. EFI values are shaded, and can be summarized as how confident the ensemble system is in an extreme event, relative to its own reforecast climatology (making this a calibrated product since model bias is effectively removed). Values above 0.8 are usually quite significant. The black contours are Shift of Tails (SoT), and can be summarized as how extreme is the extreme solution – how extreme the outlier solutions in the ensemble are. While this advertised values of 0.8 scattered across the west, it was most concentrated in the Central Valley of California. To help understand what this EFI approaching 1.0 means, we can look at the ensemble in a box-and-whisker plot that also has ensemble climate.
In the second/right image from ECWMF charts, we’ll focus on the bottom plot, with the red and blue box and whisker plots. The box and whiskers represent the range of the ensemble forecast. The red lines and shading represent the model climatology- that is, past forecasts for the point (in this case, reforecasts for the past 20 years +/- 15 days of the period). Notice by the 4th and 5th, the entire forecast distribution lies above climatology – this is what an EFI near 1.0 means. In other words, ensemble systems latched onto this heat event at least a week in advance.
When Will it End?
But when will this end? We can once again turn to ensembles. If spaghetti charts come to mind with that word, we have no need to look at those. There was a time we could look at them, and if they resembled their namesake- “spaghetti”, we could throw our hands up and say there was too much uncertainty to even talk about possibilities (it is often tempting to do this with deterministic runs too). But we have new tools, thanks to artificial intelligence that allows us to sift through that spaghetti. But first, a quick peek at the ECMWF ensemble again.
From the above plots of ECMWF ensemble mean 500mb height and anomaly, we can see the western US ridge peak sometime Tuesday into Wednesday, and by Thursday night, the ridge has pretty much collapsed, with a tropical system near the Baja, and troughing taking over, especially for northern portions, which should bring relief to the heat. But these are means. And can mask important features and outliers, and in the end, represent a single solution, which may or may not be meteorologically consistent. I mentioned AI earlier, so let’s take a look at that.
The 5-panel plot above combines the American (GFS), Canadian (GEPS) and European (EPS) ensemble systems into a 100-member grand ensemble, then uses AI/ML (specifically, an Emperical Orthogonal Functions (EOF) and k-means clustering) to sort the would-be spaghetti into distinct clusters/scenarios. This particular product will always produce 4 clusters. The shading in this plot represents 500mb height departure from climatology, with the cluster 500mb height contoured. The numbers on top represent the ensemble members and resulting percentage of that ensemble in that cluster (C = CMC, G = GEFS, E = EPS). These clusters are valid 24 hours ending 00Z on Friday (Sept 9), just before the later ECMWF mean 500mb height image I showed earlier. For the most part, we can see the evidence of the pattern already breaking down, with troughing entering the west. Clusters 1 and 2 have the trough a little deeper and further along, with perhaps a slightly slower trough in cluster 3. Cluster 4 meanwhile, keeps the trough week and to the north, maintaining the ridge through this day. This solution (cluster 4) only has one ECMWF member in it, and only represents about 15% of the grand ensemble solution space, which is why we didn’t see such an indication in the EPS mean alone.
If we advance another day, we can see more differences emerge. We can see both timing and strength differences around the trough, as well as differences in the position and strength of the remaining ridge. Clusters 1 and 2 keep a strong ridge, but just off the west coast, while cluster 4 continues to pain only a glancing blow by the trough, with the ridge remaining over the Western US. Cluster 3 seems to indicate a trough digging through much of the west, with overall lower heights across the west. Notice that while it is likely that the heat will end in the back half of the heat as the ridge collapses, this is not a certain outcome, there is a small (15%) chance that it will indeed linger an extra day or two. This is always something to look for and keep in mind. If you didn’t quite follow these clusters, don’t worry. I can do another post with a deeper dive to walk through the magic. So what does this all mean? We can turn to the NBM to help refine our message, since it ingests all these models and more, and calibrates them on a high-resolution 2.5km grid.
The three images above show the probability of maximum temperature greater than 100F for Sept 7, Sept 9, and Sept 11. Note the widespread elevated values across the valleys of California, through the Great Basin, and even into the Montana plains. By the 9th, this decreases for most, but hangs on in California. Finally, notice how even on the 11th, there are still low probabilities hanging on in California’s Central Valley. This makes sense given what we saw in the clusters. Even though a low probability, adding duration to a heat event results in further cascading impacts, and should not be ignored. If you are interested in those plots/data, be sure to come to the AMS Annual Meeting this January in Denver – I have submitted an abstract with my efforts on this front. In short, this is dressed up python, run in Google Colab, grabbing data from AWS.
And finally, because I witness Salt Lake City, UT weather every day, gotta plug just how extreme the heat is, even here (where the EFI was closer to 0.8-0.9). For this, I used experimental NBM 4.1 text bulletins (since in v4.1, both the probabilistic component [for the shaded forecast range] and deterministic forecasts are both training to the METAR). Note the all but guaranteed stretch of continued record temperatures for the next several days. And while the 10th percentile almost falls back around normal, the 90th percentile still hangs out around 100 for the 10th. Of particular additional note, the all-time record for SLC is 107F, and the 90th percentile for Wednesday (Sept 7) has crept up to 106F. Notable in June, July, and as it turns out…even September.