In the last post, we looked at the heat potential for the week, the peak, and some uncertainty in the cooldown. So how did we do, using tools like the clusters, ECMWF EFI, and NBM? Pretty well! Shown below are just a few of the locations that tied or exceed their all-time record highs with data from acis. From these, we can see many California sites that exceed their all-time records, (in September nonetheless), and still remained warm on Friday. Further west in the Great Basin, Salt Lake City tied their all-time record on Wednesday, started cooling down on Thursday, and by Friday had returned to near normal. It should have not been very shocking to see these kinds of temperatures given the guidance and especially the ECMWF EFI values greater than 0.9 for many of these places.
So…what about the NBM?
Remember those NBM probability of exceeding 100 degrees plots we showed in the last post? Let’s revisit those, and add some observations. We can see a pretty good footprint by the NBM, and this was indeed a powerful messaging tool should we have needed it. These graphics were made from the NBM archive on AWS, and the synoptic data API. I do plan on sharing this code by the time the AMS Annual Meeting this January in Denver occurs.
And what about those clusters?
With their valid times in the past now, if we check back at the same WPC clusters we looked at before we can see how they did. This time, they look slightly different. They have an extra panel and an MAE score at the top. In the case of the first image, valid 00Z Friday, neither the grand ensemble mean, nor the biggest cluster was most representative, but it was a pretty thin margin, as cluster 1 (the largest), cluster 2 (the best performing with an MAE a whopping 1 meter better), and the grand ensemble mean all looked very similar anyway. The trough did indeed clip northern portions of the area at this time, and the ridge had begun to weaken for most (but still hanging on in California). And in the second clusters image from the same cycle for the next day (as in the previous post), we once again see something that shows through in bulk verification stats – a cluster has once again outperformed the grand mean, with cluster 1 having the lowest MAE representing a deeper, and little bit slower trough compared to the other clusters (except cluster 3, which had similar timing, just not the depth). This was just a glimpse at the power of ensembles, and using cluster analysis to make sense of spaghetti through distinct scenarios. As I mentioned before, there will be a post in the future with more of a deep dive.