New England Seasonal Snowfall Historical Records

[weatherwiz]

I will start off by saying how infuriating it is how irresponsible record keeping has been, especially with the debacle from the mid 1990's through the early 2000's. I totally forgot what happened (I know Will has explained it many times) but records were lost and there are also many questions about the accuracy of some of the data. I don't want to divulge into this any more because it drives me crazy. 

Anyways, one thing I've really thought about the past few years when discussing average snowfall (in relation to the season) is whether or not averages are being swayed by anomalous seasons which would be defined as outliers (when performing statistical methods). As far as I know, averages are just reported by taking the sum of seasonal totals and diving them by the number of years on record. Nice, easy, simple, however, when dealing with datasets which can have a wide array of variation, you run the risk of outliers. One great example of this is with the tornado database. If you were to calculate the average number of tornadoes from 1950 to say 2015, the mean is going to be skewed quite a bit by the 2011 season. Now there are other factors involved here but just making a reference. 

What is the point of this thread? Moreso a place to just collect my thoughts but also open to discussion and maybe through collaboration we can all create some sort of database which is legit and maybe even fill in the gaps on some of the missing years. For what I did below with BDL, I would like to do for other stations as well. I also hope I didn't make any errors below. I'm not usually up this late 

Anyways, I was curious with BDL. The data was obtained from 

https://xmacis.rcc-acis.org/

Year was defined as July 1 - June 30

Notes: The long-term average is 44.8''. With the outliers removed, the average drops to 43.4''. Not a huge drop. 

Again, this is just based on the data from acis. Who the hell knows what is right and what is not right. 

 

[Ginx snewx]

You tossed the high outliers and not the.lows?

[weatherwiz]

2 hours ago, Ginx snewx said:

You tossed the high outliers and not the.lows?

There were no low outliers (to my surprise) 

[dendrite]

What? Why?

Snowfall is naturally going to be skewed toward bigger seasons at lower latitudes. Why would you just pretend the 3 snowiest seasons never happened? 
 

Just find the median value or the range of snowfall that is +/- 1 or 2 SD from the mean.

Maybe you should’ve went to bed earlier.

[weatherwiz]

Just now, dendrite said:

What? Why?

Snowfall is naturally going to be skewed toward bigger seasons at lower latitudes. Why would you just pretend the 3 snowiest seasons never happened? 
 

Just find the median value or the range of snowfall that is +/- 1 or 2 SD from the mean.

Maybe you should’ve went to bed earlier.

Not pretending they didn't happen, just curious to see how outlier seasons skew the overall data/mean. I was anticipating there would be some lower outliers. 

But you can see I did have a column setup for SD and was going to do that next...just didn't get to it (although it takes like 20 seconds). 

At least with BDL right now, its interesting that the top 3 seasons are enough to increase the overall mean by nearly 2''. But at the end of the day that's not overly significant. 

[The 4 Seasons]

How is 14.7" not a low outlier or all the other teens? That's what like a third of average. What are you using to define what is a low/high outlier? Did you just use the top 3? If so why not the bottom 3?

The difference from 43.4" to 44.8" is 1.4", calling it nearly two inches is a bit of stretch don't you think? In reality it's closer to 1. 

I did notice that 30yr normal jumped quite a bit when we went from the 1980-2010 period to 1990-2020. I think it was ~45" before now it's 51.7. And yeah I agree about the record keeping and lack of data drives me crazy. At least they fixed the 2005-2015ish period that was all f-d up and missing about a year ago. 

[weatherwiz]

Just now, The 4 Seasons said:

How is 14.7" not a low outlier or all the other teens? That's what like a third of average. What are you using to define what is a low/high outlier? Did you just use the top 3? If so why not the bottom 3?

The difference from 43.4" to 44.8" is 1.4", calling it nearly two inches is a bit of stretch don't you think? In reality it's closer to 1. 

I did notice that 30yr normal jumped quite a bit when we went from the 1980-2010 period to 1990-2020. I think it was ~45" before now it's 51.7. And yeah I agree about the record keeping and lack of data drives me crazy. At least they fixed the 2005-2015ish period that was all f-d up and missing about a year ago. 

ha...I did the math in my head wrong, yeah the difference is 1.4'' I was thinking 1.8'' so that's that error. 

For the outliers I just did it in Excel. The work is at the top there to the right. 

I found Q1, Q3, and so forth. 

The lower limit was -10 and the upper limit was 83.2. 

[ORH_wxman]

1 minute ago, weatherwiz said:

ha...I did the math in my head wrong, yeah the difference is 1.4'' I was thinking 1.8'' so that's that error. 

For the outliers I just did it in Excel. The work is at the top there to the right. 

I found Q1, Q3, and so forth. 

The lower limit was -10 and the upper limit was 83.2. 

I’m surprised there were no -10 inch snowfall seasons. 

[weatherwiz]

1 minute ago, ORH_wxman said:

I’m surprised there were no -10 inch snowfall seasons. 

My math or statistics may be incredibly rusty but how do you handle situations with outliers where your lower limit would be a negative number but obviously its impossible for any of the data to be negative? I don't ever remember going over something like that. 

But the way these past few winters have gone, we may as well consider the negative. 

[ORH_wxman]

35 minutes ago, weatherwiz said:

My math or statistics may be incredibly rusty but how do you handle situations with outliers where your lower limit would be a negative number but obviously its impossible for any of the data to be negative? I don't ever remember going over something like that. 

But the way these past few winters have gone, we may as well consider the negative. 

So the problem is snowfall is not a normal distribution. It has a high tail because of the fact we can’t have negative numbers on snowfall. 
 

So the easiest way to exclude outliers equally in a fast/crude manner is to put your entire array of values in a column on excel (you prob already did this)….now go to “formulas” tab on the top of the spreadsheet and go to the drop-down menu that says “more functions” and then go to “statistical” and then scroll all the way down to “Percentile.exc”. Once you do that it’s going to ask you select the array of values, so click on the first value of your snowfall total, hold shift, and then select the last value so that it includes all of them in the array.

Now, below the array line there is a “K” value. For your 10th percentile value, type in “0.1” and you should see a low-ish snowfall value appear below the K line on the right. That is your 10th percentile snowfall value…write it down or type it someonewhere . Now do the same thing for 90th by inputing “0.9” for the K value. 
 

Once you have have your 10th and 90th percentile values, you can toss all values above the 90th and below the 10th if you want to exclude outliers in an even manner. This gets rid of the problem of trying to use methods you’d normally use for a normal distribution. 
 

Now you could do this for any percentile value you wanted. If you wanted the 50th percentile, you type in 0.5 into the K line…for 25th you input 0.25….you get the idea. 

[weatherwiz]

15 minutes ago, ORH_wxman said:

So the problem is snowfall is not a normal distribution. It has a high tail because of the fact we can’t have negative numbers on snowfall. 
 

So the easiest way to exclude outliers equally in a fast/crude manner is to put your entire array of values in a column on excel (you prob already did this)….now go to “formulas” tab on the top of the spreadsheet and go to the drop-down menu that says “more functions” and then go to “statistical” and then scroll all the way down to “Percentile.exc”. Once you do that it’s going to ask you select the array of values, so click on the first value of your snowfall total, hold shift, and then select the last value so that it includes all of them in the array.

Now, below the array line there is a “K” value. For your 10th percentile value, type in “0.1” and you should see a low-ish snowfall value appear below the K line on the right. That is your 10th percentile snowfall value…write it down or type it someonewhere . Now do the same thing for 90th by inputing “0.9” for the K value. 
 

Once you have have your 10th and 90th percentile values, you can toss all values above the 90th and below the 10th if you want to exclude outliers in an even manner. This gets rid of the problem of trying to use methods you’d normally use for a normal distribution. 
 

Now you could do this for any percentile value you wanted. If you wanted the 50th percentile, you type in 0.5 into the K line…for 25th you input 0.25….you get the idea. 

Ahhh thank you! I didn't even think of going about it that way. This makes so much more sense. 

[tamarack]

1 hour ago, ORH_wxman said:

So the problem is snowfall is not a normal distribution. It has a high tail because of the fact we can’t have negative numbers on snowfall. 
 

So the easiest way to exclude outliers equally in a fast/crude manner is to put your entire array of values in a column on excel (you prob already did this)….now go to “formulas” tab on the top of the spreadsheet and go to the drop-down menu that says “more functions” and then go to “statistical” and then scroll all the way down to “Percentile.exc”. Once you do that it’s going to ask you select the array of values, so click on the first value of your snowfall total, hold shift, and then select the last value so that it includes all of them in the array.

Now, below the array line there is a “K” value. For your 10th percentile value, type in “0.1” and you should see a low-ish snowfall value appear below the K line on the right. That is your 10th percentile snowfall value…write it down or type it someonewhere . Now do the same thing for 90th by inputing “0.9” for the K value. 
 

Once you have have your 10th and 90th percentile values, you can toss all values above the 90th and below the 10th if you want to exclude outliers in an even manner. This gets rid of the problem of trying to use methods you’d normally use for a normal distribution. 
 

Now you could do this for any percentile value you wanted. If you wanted the 50th percentile, you type in 0.5 into the K line…for 25th you input 0.25….you get the idea. 

I'm lazy.  I'd merely sort from high to low, note n (looks like 113 since 97-02 are msg) and excise the top/bottom 11 of the column.
(I also like the median in accounting for outliers.  Doesn't always work, Jan snow here has the median 1.6" above the mean.)

[RUNNAWAYICEBERG]

On 2/2/2024 at 9:25 AM, ORH_wxman said:

I’m surprised there were no -10 inch snowfall seasons. 

In the past couple winters, there has been. A lot of snow has been stolen from sne…

[weatherwiz]

@ORH_wxman Has snowfall for ORH been measured at the same location since records began or did the location change at some point?

I've gone through and compiled a list of seasons below the 10th percentile and above the 90th for the major climo locations and for ORH I found it super interesting that since the 1950's any anomalous season is likely to be in the top percentile vs. the lowest. The first 50 years on record any anomalous season was virtually on the lower side

And for those wondering, there is a reason why I am doing this. I've never dabbled much into the historical snowfall climatology but now that I am getting back into seasonal forecasting, it's important I better understand the climatology and I like playing around with data in different ways. 

[ORH_wxman]

22 minutes ago, weatherwiz said:

@ORH_wxman Has snowfall for ORH been measured at the same location since records began or did the location change at some point?

I've gone through and compiled a list of seasons below the 10th percentile and above the 90th for the major climo locations and for ORH I found it super interesting that since the 1950's any anomalous season is likely to be in the top percentile vs. the lowest. The first 50 years on record any anomalous season was virtually on the lower side

And for those wondering, there is a reason why I am doing this. I've never dabbled much into the historical snowfall climatology but now that I am getting back into seasonal forecasting, it's important I better understand the climatology and I like playing around with data in different ways. 

ORH pre-airport site is anything prior to 1947-48 I think. 

[ORH_wxman]

And yes, there was a pretty awful stretch of snowfall between 1930-1955 at ORH. It covers both sites, but even if you bumped up the totals slightly from the old site prior to 1947, you’d still have low totals in the 1930s and 40s. 

[weatherwiz]

2 minutes ago, ORH_wxman said:

And yes, there was a pretty awful stretch of snowfall between 1930-1955 at ORH. It covers both sites, but even if you bumped up the totals slightly from the old site prior to 1947, you’d still have low totals in the 1930s and 40s. 

It's fun doing this stuff. When you look at the data, especially if you graph it, you can easily see we go through stretches of horrible periods and periods where we get nailed. While I think most understand that, what drives me nuts is how some blame the lull now on AGW. Now, I am a believer that human activity is escalating climate change, but I'm not going to blame every anomalous weather event and what not on AGW. 

 

[ORH_wxman]

1 minute ago, weatherwiz said:

It's fun doing this stuff. When you look at the data, especially if you graph it, you can easily see we go through stretches of horrible periods and periods where we get nailed. While I think most understand that, what drives me nuts is how some blame the lull now on AGW. Now, I am a believer that human activity is escalating climate change, but I'm not going to blame every anomalous weather event and what not on AGW. 

 

It’s inherently tough to “blame” things on multiple causes. Our brains aren’t really wired that way…we want to have a quick and easy one-stop-shop explanation. That’s why you get so many people who bristle anytime you try and assign multiple causes for a weather event or season. As someone who has looked at so much climate data since I started studying this stuff over 2 decades ago, I was already aware of the large swings we see in the past so when we get them in the present, I don’t feel the need to pretend it’s a new phenomenon. It’s easier to break warm records now and harder to break cold records due to CC. Aside from that, I’m usually hesitant to put too much attribution on CC for other weather phenomenon since those are much messier datasets and you lose the robustness of the relationship. This is particularly true for snowfall in New England.
 

Even temps are a bit messy…they increase long term but we see large temporal and spacial variations (I always tell people to look at the 2-3 decade trend in the N plains/Rockies)…sometimes natural variation is working in the same direction as CC and sometimes it is not. 

[weatherwiz]

44 minutes ago, ORH_wxman said:

It’s inherently tough to “blame” things on multiple causes. Our brains aren’t really wired that way…we want to have a quick and easy one-stop-shop explanation. That’s why you get so many people who bristle anytime you try and assign multiple causes for a weather event or season. As someone who has looked at so much climate data since I started studying this stuff over 2 decades ago, I was already aware of the large swings we see in the past so when we get them in the present, I don’t feel the need to pretend it’s a new phenomenon. It’s easier to break warm records now and harder to break cold records due to CC. Aside from that, I’m usually hesitant to put too much attribution on CC for other weather phenomenon since those are much messier datasets and you lose the robustness of the relationship. This is particularly true for snowfall in New England.
 

Even temps are a bit messy…they increase long term but we see large temporal and spacial variations (I always tell people to look at the 2-3 decade trend in the N plains/Rockies)…sometimes natural variation is working in the same direction as CC and sometimes it is not. 

Well said, this is an excellent post! Couldn't agree any more with this.

[tamarack]

2 hours ago, ORH_wxman said:

It’s inherently tough to “blame” things on multiple causes. Our brains aren’t really wired that way…we want to have a quick and easy one-stop-shop explanation. That’s why you get so many people who bristle anytime you try and assign multiple causes for a weather event or season. As someone who has looked at so much climate data since I started studying this stuff over 2 decades ago, I was already aware of the large swings we see in the past so when we get them in the present, I don’t feel the need to pretend it’s a new phenomenon. It’s easier to break warm records now and harder to break cold records due to CC. Aside from that, I’m usually hesitant to put too much attribution on CC for other weather phenomenon since those are much messier datasets and you lose the robustness of the relationship. This is particularly true for snowfall in New England.
 

Even temps are a bit messy…they increase long term but we see large temporal and spacial variations (I always tell people to look at the 2-3 decade trend in the N plains/Rockies)…sometimes natural variation is working in the same direction as CC and sometimes it is not. 

All good points, and the long look is crucial.  We've had 5 events with 3.25" or greater, including 2 with 4"+, in the period Oct 22-Dec 23, only 15 months.  However, June 98 thru Sept 99, 16 months, also had 5 such events and 4 of those topped 4".  Looking the other way, 2001-04 had no such events and 2018-21 had only one.  (Context: From May 1998 thru the present, we've had 29 storms of 3.25"+, about 1.2 per year.)

Edit:  The late lamented Farmington co-op has winters 1893-94 thru 2021-22, so 'n' = 129.  Their average winter is 89.6" and the bottom 10% (13th lowest) had 60.6" and the top 10% winter had 122.6".