Jump to content
The Weather Forums

Recommended Posts

I wanted to share some of my current research - I think this work is particularly interesting for us here in the PNW. I am currently analyzing 500mb circulation fields for 1979-2019 across western N. America, using ECMWF's ERA5 reanalysis product.

 

Looking at just JJA, I ran the sample set of 3,772 JJA days over the 41 year period through an artificial neural network algorithm called a Self-Organizing Map (https://en.wikipedia.org/wiki/Self-organizing_map). This is essentially a clustering mechanism that produces composites of "archetypal patterns" or "nodes" that represent every day in the sample. Each of the 3,772 days is binned into one of the 6 nodes that best represents it.

 

Attached are two images:

 

- The first one shows the 6 archetypal 500mb patterns for 1979-2019 summertime days.

- The second image shows the 41-year time series for each of the 6 nodes. It's pretty neat how you can see the dramatic dropoff in nodes 1-4 since 2013, which are the nodes that don't feature widespread western US ridging. They have been appearing less often, with the exception of 2019 when the node 3 pattern returned in quantity. Node 1 in particular had a dramatic dropoff after 2011, reaching its lowest frequency in 2018. Also notice the high frequency of nodes 5 & 6 in recent years, and the relative absence of node 6 patterns in the 1980s.

z500_nodes.png

node_percent.png

  • Like 9
Link to post
Share on other sites

Awesome work. That flip to persistent summer ridging correlates nicely to the reversal in the PMM/AMM ratio and associated Hadley/Walker intensities in 2013. The tendency towards +NAO on the low pass also derived from that flip.

 

Hard to imagine it lasting much longer, though. It almost has to change over these next few years. Has already backed off somewhat, though the underlying low pass signal is still present (at a lower frequency than before, but it’s still definitely there).

  • Like 2
Link to post
Share on other sites

Great stuff.

 

I also found the fact that the summer of 1985 was dominated by node 2 to be interesting. That node appears to be associated with a sharp block located just offshore with cooler airmasses diving into the northern Rockies. That configuration can end up rather hot around here in the summer with increased offshore flow (as we saw in July 1985), but it quickly morphed into a cool to very cold pattern by the late summer into the fall.

  • Like 1
Link to post
Share on other sites

Confirming that 2017-18 were pretty unique. As were 2010-11.

Yeah, we saw both ends of the spectrum last decade. We've talked a lot about how 2010 and 2011 were a throwback to the old days, and this confirms it at a larger scale.

Link to post
Share on other sites

Great stuff.

 

I also found the fact that the summer of 1985 was dominated by node 2 to be interesting. That node appears to be associated with a sharp block located just offshore with cooler airmasses diving into the northern Rockies. That configuration than can end up rather hot around here in the summer with increased offshore flow (as we saw in July 1985), but it quickly morphed into a cool to very cold pattern by the late summer into the fall.

That's a great observation about 1985. Our heat that summer had very different dynamics from the recent massive ridge-dominated summers. Also speaks to the fickle nature of trying to project how heat extremes might change along the immediate west coast in the future, places west of the Cascades and west of the Coast Range in California. So much depends on the right longwave patterns recurring, as opposed to actual warming in the means.

  • Like 2
Link to post
Share on other sites
  • 2 weeks later...

I've read about and wanted to do some work with these SOMs. What software/language did you use to run these?

Looks similar to stuff I’ve done with the Basemap and matplotlib modules in python.

Link to post
Share on other sites

I've read about and wanted to do some work with these SOMs. What software/language did you use to run these?

Hey, I'm using the 'somtoolbox' in MATLAB to create these. As far as I understand, this is still the canonical choice which most published papers have used for their SOMs analysis. There are also implementations in Python and R that users have contributed, but I don't think any of them are as stable or trusted as the MATLAB implementation.

Link to post
Share on other sites

Looks similar to stuff I’ve done with the Basemap and matplotlib modules in python.

You nailed it, Matplotlib + Basemap for those maps. I ran the SOMs algorithm in MATLAB and then exported the data to Python for analysis and plotting. The time series chart was actually done in Seaborn. I would do everything in Python if it were up to me, but I have to jump to closed source platforms once in a while. :P

  • Like 1
Link to post
Share on other sites

Hey, I'm using the 'somtoolbox' in MATLAB to create these. As far as I understand, this is still the canonical choice which most published papers have used for their SOMs analysis. There are also implementations in Python and R that users have contributed, but I don't think any of them are as stable or trusted as the MATLAB implementation.

I wish I still had access to Matlab! Thanks for answering though. I will just have to go with pythons implementation. 

Link to post
Share on other sites

I wish I still had access to Matlab! Thanks for answering though. I will just have to go with pythons implementation. 

Yeah, that's the problem with Matlab. You have to be in academia, otherwise it's not really relevant (or accessible).

Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

Loading...
×
×
  • Create New...