NittanyWx

First big win of the winter for the AI models. Plowable coast. AIFS in particular did well in digging the SW and turning the trough more neutral vs positive a couple days back.

I like the slightly more amped trends we're seeing today as well.

I again point to the deep DGZ with this and think fluff factor will help.

I like the depth of the DGZ with this one. A little liquid goes a long way. Great test case for AI modeling here with the more amped and juicier solution and a notable difference in trough tilt. I think this is a measurable one for the coast at least. This vort here is holding the keys to the difference in more neutralish rough tilt vs more positive. I don't know the answer yet on whether AI is better poised to be the better predictor vs NWP on trough axis tilt and whether this is a use case where it has skill. I do know the Euro is further north than the GFS at 48 hours, has a less positive tilt and if I'd have to guess in most cases the trend in recent years is north late. I wouldn't necessarily dismiss plowable fluff at this stage for the coast... The confluence near us is the counterargument to that.

You posted a relational inference based on observed data, yes We're not arguing about that.

Yeah so my only point for this is if there's really something here I think we'd see it in actual statistical correlations with ENSO to snowfall. Whether that be rate of change of ENSO over the winter, a linear or partial correlation between regional observation and ENSO value itself...just something there that's more robust and can give us something to really sink our teeth into and find out what it's telling us and why. I think your points here aren't far from where I'm at too. But I'm more curious as to thought process above all else to see if there's something testable.

Asking you about your methodologies and whether they actually make sense meteorologically or whether it's an overfitting of observational data is stirring up controversy now?

Posted through the entirety of last winter and had a very good one forecasting wise. Had a very good one the year prior too.

Nice, so now you're calling my work sloppy because you can't find a meteorological reason for why your threshold makes sense other than it 'fits the data'. You've dodged the question 5 times now. I know what I am doing and your arrogance is really starting to show here. You haven't seen my work, I can promise you that. I don't work at a vendor.

The math does not hold man, I'm literally sitting here doing the testing and its an overfit. My best guess is found a value you think makes sense based on recent record and picked it because it fit observed data. All I've asked you, repeatedly I might add, is why 4" means something statistically and meteorologically. Why did you choose 4"? ENSO is not and never really has been a great predictor of snowfall here.

Those three stations are what I'm testing. Again, assumption made that is incorrect. Take an average of those three stations, that's your 'snowfall index'. Bench that snowfall index to ENSO past 30 years and you get a correlation that fails. So other than arbitrarily fitting it to the data, why does 4" mean something meteorologically?

You're making a lot of assumptions about how I do things and those assumptions are wrong. I used SAI as an sample of a 'near term' trend with a strong statistical correlation that showed forecastable value off a short sample size and ultimately failed as your sample size got larger. My correlation work largely ignores what happens before 1991. The correlation still fails to reach statistical significant when testing between ENSO, the SOI and any number of corresponding variables when you benchmark them to regional snowfall observations. You're speaking with someone who has done this professionally in the commodities space for 15 years. My methodologies discount severely anything before 1991 because it is a different climate regime. This is an industry that benchmarks to the 10 year normal and not 30 -myself included- so I don't need a lecture and explanation about understanding what regime we are in. This isn't a discussion about that, it's a discussion about overfitting to an arbitrary value. For the record, I also do detrended analysis to find statistical signals around warmer background trends. I am very cognizant of the CC forcing arguments you've made and how deeply you believe them. But as we got into an argument last year about Feb, we're getting into an argument here on what constitues actual statistical signal and more importantly *why* you used a threshold value other than fitting it to observational data. You still need to find a robust statistical *and* meteorological reason for 4" of snow meaning something for us to take any real stock in using it as a metric. I can test it, I can say 'hey, that's kinda interesting' but you're not giving me a real meteorological reason for why 4" is an actual threshold.

Yeah so for me to take any real value of this having meteorological significance I need to see statistical correlating variables, which is why I've done a lot of work around correlations and partial correlations involving ENSO states, rate of change involving the SOI, etc and found minimal forecast/predictability value for snowfall locally. I've been burned by threshold/relational things in the past (SAI being chief among them) and have since really been hesitant on overfitting data to find some grain of predictability to it.

What, specifically are you responding to? The statistical correlation between ENSO 3.4 and NYC seasonal snowfall falls well below statistically relevant skill. And, why, exactly is 4" a meteorologically significant threshold? If we magically get 3.87" of snow vs 4.03" inches of snow we are SCREWED! Sounds like overfitting to me, but what do I know.

Snow and Nina correlations are abysmal and you're not bound by prior analogs for it.