Before Jim left for his much deserved vacation to Europe, he asked his students to do occaisional posts. And since Jim is away I thought I would post about something he would never post about: a winter outlook based on current weather. The basic idea is that there are long term atmospheric connections that we do not yet understand, but we can still see their effects.
So does our current dry October offer any insight on our winter ahead?
Below is the SWE plot from the 2004 - 2005 winter at the Snowbird Snotel. By November 1st a whopping 13.1 inches of precip had already fallen and the snowpack at Snowbird was already comparable to an average January 1st snowpack! This winter would go down as one of the snowiest since the snotel first started reporting in 1990.
Below is the SWE plot from the 1990 - 1991 winter. Total precip for October amounted to only 3.8 inches, 1 inch below normal, and the winter never recovered.
But how representative are these 2 winters? Below is a plot of October Precip versus the mean monthly precip for the following November - April. I slapped a best fit line over the data and the trend is up, but only with an r-squared value of .18 (1 means perfectly correlated, 0 means no correlation). This suggests that dry Octobers may increase the odds for a dry winter and vice versa, but keep in mind the sample size is small (24 winters), and the correlation is weak. For example if the outlier winter of 2004-2005 winter is removed, the trend is still up, but r-squared drops to 0.10.
Next I investigated the upper levels during winters after dry and wet Octobers. Winters ending in 2004, 2000, 2009, 1996, and 2013 had the 5 driest Octobers and below is the the mean 500 mb geopotential height anomalies for those 5 winters (Nov - Apr). There are above average height anomalies across the entire West suggusting ridging during those winters.
Next I looked at the winters after the 5 wettest Octobers which ended in 1995, 2005, 2007, 2008, and 2011. Although Utah has positive height anomalies, there are negative height anomalies over the Pacific northwest suggesting some upstream troughing.
Again these signals are very weak and may not pass significance tests, but it is interesting to see a signal. Why is there a signal? Perhaps seasonal to annual teleconnections, like ENSO, or something we haven't even discovered yet, weights the odds for above or below average precip during the winter and the signal begins in the fall. It's fun to speculate, but no one really knows the answer.
So where do we stand currently? Thus far the Snowbird snotel has received 1.3 inches of precip for the month. If we do not get any more this winter would be tied for the 2nd driest October, but there is still a lot of month left. Looking at the long range, however, I find it unlikely that Snowbird receives the 4.8 inches of precip necessary for an average October, which may slightly weight the odds for a below average winter.
Look for another post this week by another graduate student about the Arctic Oscillation and its role on our winter weather, and perhaps I will post again in November about how November precip is an even better indicator of December - April precip than October. Only if I am allowed of course.
post by Jeff Massey
Back of the envelope: Snow season is Oct-April, six months, so Oct is 1/6 = 0.16 or 16% of the snow season. If there is statistical independence in the totals each month, then one would expect that the variance accounted for by any given month is roughly 10-18%. One would expect that Oct would account for less than Feb, say, but looking at the average snow total chart shows the Oct1 to Oct31 slope to only be slightly less than mid-season slope, which is surprising to me, but maybe I'm just in the mountains more in Feb so I don't really notice... Could you do the same analysis on Feb? I'm betting r-squared is no more than 0.2
ReplyDeleteOn the scatter plot, is that really the preceding Nov->April that is plotted against the following October? Or did you mean to say it's the preceding October plotted against the Nov->April that follows?
ReplyDeleteGood catch Lisa! Yes I meant to say the preceding October
DeleteOne way you could further investigate this is to do a month-to-month correlation of monthly precipitation totals (not of existing snowpack or accumulated precip, obviously) to look for patterns. For example, first correlate October vs. November, then November vs. December, etc for the available period of record. If the correlation coefficients all turn out to be very small numbers or are quite variable (some positive and some negative, etc) then likely there is essentially zero predictability with regard to this. If the coefficients are more significantly non-zero, or show some sort of obvious trend as you go through the season, then this may have some significance in terms of guessing at future precipitation based on a given month. My guess is that for Utah these coefficients would be quite low and variable, implying essentially no statistical relationship between one month and the next. But maybe there is something there.
ReplyDeleteGood idea, David. My guess is that there would be some weak correlation between months, but perhaps not. I will try to run the numbers next week
DeleteIt seems to me that you may be looking for a multi-dimensional correlation. For example, best correlation may not be between time periods at the same location but any two time periods at perhaps different locations. In other words the best correlation for snowfall in Utah my be tied to average temperature in Tundrova Siberia in July.
ReplyDeleteWell done post. It just supports the idea that long term seasonal forecasts are very hard to predict. Would have been nice to see this correlated to other snotel sites. As we know there is significant variation between sites even within <100 miles.
ReplyDeleteI have correlated month-to-month snowfall for 27 North American ski locations over an average of 42 years. I do not have October data.
ReplyDeleteNovember to December +15%
December to January +10%
January to February +20%
February to March +6%
March to April +19%
These are consistent with jdm83's findings
2 month correlations (November to January for example) are all in the +5% range, which is about as close to random as you can get.
The one month correlations are weak also. If you look at the individual ski area one-month correlations, about 20% of them are negative.
My conclusion is that current weather has trivial influence on future weather more than a month in advance.