2017 snow depth prediction

TL;DR: The Spencers Creek¹ peak snow depth for 2017 will be 163 ± 44 cm.

As you probably know, I predict the season peak snow depth at Spencers Creek using a simple seven-parameter multiple regression model based on well-known climatic modes and influences. My mark-IV prediction model goes as follows:

Spencers Creek peak depth (cm) = 899 – 0.311 x year – 13.8 x AAO + 2.09 x SOI – 8.80 x IOD – 5.36 x PDO – 143 x SST + 534 x aerosol

The parameters are explained in the notes at the end. Click on the graphs for sources.


Calendar year

It’s 2017, so take off yet another centimetre².   Eventually there will be none left.


Antarctic oscillation (AAO)³

This parameter is supposed to be the hard one, but last year it was far from my worst pick (that was the local sea surface temperature). AAO is persistent over months, but it can also change suddenly. There is some very limited numerical guidance available from global climate prediction models like the Australian Bureau of Meteorology’s (BOM’s) POAMA model (not public). Here’s how AAO is looking (linked image will update):

Antarctic oscillation

We need the winter average — the “3-month running mean” to the end of August. After a fairly strong negative phase (good for our snow), it looks like we’re headed back positive again. Note that there is a long term global warming uptrend in this parameter. I’m going with +0.5.


Southern oscillation index (SOI)⁴

SOI reflects ENSO, which models have had heading back into El Niño this year (meaning SOI generally below about -10 and staying thereabouts for several months). But recently some seem to be backing off a little from that position, for example BOM’s POAMA now has this (static image link; negative Nino3.4 temperature anomaly correlates with a positive SOI):

Internationally, the various model predictions look like this (the 2017 winter average that we want is ‘JJA’ — the third tick in from the right edge; static image link):
IRI dynamic models

I don’t think there’s going to be a winter El Niño this year. I’m adopting a winter average SOI of -5, which is negative for our snow but not by much.


Indian ocean dipole (IOD)⁵

BOM’s POAMA model has this (static link):

But as you can see, it’s well out of step with the international competition:

The Japan Meteorological Agency’s model is not on that chart (IOD was invented by a Japanese guy). JMA is more aligned with BOM:

It seems there’s some chance of a strongly positive IOD this winter, which would be negative for our snow. I’m not in that camp; I’m adopting a neutral 0.0.


Pacific decadal oscillation (PDO)⁶

My model uses the two year average to the end of winter. The PDO seems to be firmly in positive mode, after the 2-year average made it to +1.7 last year, a record for our snow depth measurement era. The average from September 2015 to March 2017 has been +1.4, which is what I’m adopting. That is a bit negative for our snow.


Sea surface temperature (SST)⁷

This is the thing I don’t seem to be able to pick. Here’s how the winter average has been going in recent years:


The trend in sea surface temperatures in our area of interest is relentless, and the correlation of that with our season peak snow depth is the strongest of the lot. One degree Celsius rise in local sea surface temperature loses us about a metre of expected peak snow depth.

Our winter SSTs have been near or below trend in recent years (they could easily have been much worse). Our monthly (all seasons) SSTs have recently taken a huge dip (hence my bad choice last year), from which they seem to be rapidly recovering:

SST_box SST_monthly_recent

That data is from the Hadley Centre’s HadISST dataset through March. Here’s what NOAA’s CFSv2 model is predicting for winter (anomalies vs 1999-2010, add about 0.5°C for 1951-1980 in our region; will update):

CFSv2 SST anomaly prediction for June-July-August 2015
(vs 1999-2010, add 0.5°C for 1951-1980)

And here’s BOM’s SSTs over the last available week (anomalies vs 1961-1990, add 0.1°C for 1951-1980; static link):

Latest BOM weekly SSTs vs 1961-1990, add 0.1°C for 1951-1980

So it’s warm in the Tasman but not in the Great Australian Bight, which is a standard pattern now, presumably reflecting an established mode shift since the reference interval. I think the chances are our local SST anomalies will still be a little below trend this winter, which remains strongly negative for our snow, just not as negative as might be. I’m adopting a winter average SST of +0.6°C vs the 1951-1980 average (detrended about 2015).


Southern Hemisphere stratospheric aerosols⁸

There have been no major volcanic eruptions affecting the stratosphere, but low levels of aerosols persist (paywalled text) from minor eruptions and pollution. For a real-time guide to the stratospheric situation, the high-altitude Mauna Loa data (Hawaii … not SH) provides a rough indication:

Optical thickness is the negative log of transmission, so the left panel is an inverted view. The right panel is a direct stratospheric aerosol measurement by laser probing (LIDAR). I’m adopting 0.003, the same as last year.



In the mark-IV model, those seven parameters give a 2017 best-estimate peak snow depth of 163 cm — which is far from great but better than a kick up the…  The one standard deviation (‘±1σ’) error is 44 cm as before, so the range 119 – 207 cm would be expected to include about two-thirds of likely outcomes, if the parameters were perfectly known (which they certainly aren’t!). The estimated chance that the peak depth will exceed 2 metres is about 20%.

For comparison, a statistically thorough naive prediction would yield a 2017 peak depth of 175 cm, with a ±1σ range from about 118 – 238 cm. On that basis my prediction is for a below average peak depth (except that “average” might not mean a lot with our alpine climate changing so rapidly). It looks like this:

Spencers Creek peak snow depth pre-season prediction for 2017

Spencers Creek peak snow depth pre-season prediction for 2017



  1. Spencers Creek near Charlotte Pass, NSW, Australia, midway between Perisher Valley and Thredbo; data courtesy Snowy Hydro Limited.
  2. The raw downtrend is now about 0.8 cm/yr when forced to linear, or nearer 1.3 cm/yr allowing for the non-linearity. Only 0.3 cm/yr of that appears in the calendar year term of the prediction model, because it’s a linear model and because there are global warming trends in some of the other parameters.
  3. Antarctic oscillation (AAO), also called “southern annular mode” or SAM, is a measure of how tightly the circumpolar winds (“polar vortex” in one usage) blow around the pole. A loose pattern (negative AAO) leads to more polar storms reaching southern Australia, and more snow. The winter average AAO is used — the average of June, July and August.
  4. Southern oscillation index (SOI) is the difference between Tahiti and Darwin surface atmospheric pressure expressed as monthly standard deviations times ten. SOI is an indicator of the El Niño Southern Oscillation (ENSO), an east-west quasicycle in equatorial Pacific Ocean surface temperature and wind patterns which correlates with precipitation across much of Australia, including with alpine snow. A positive SOI is associated with more (and wetter) Australian snow. The winter average is used.
  5. Indian ocean dipole (IOD) is an ENSO-like variation in the smaller Indian Ocean, which correlates with winter precipitation across southern Australia, including with alpine snow. Negative IOD is associated with more snow (the sign is consistent with SOI, but we’re on the opposite side of the Indian Ocean). The winter average is used.
  6. Pacific decadal oscillation (PDO) is a long-cycle, largely north-south variation in the Pacific Ocean, which interacts with (and perhaps partially mediates) ENSO. Negative long-average PDO is weakly correlated with more Australian snow. The 2-year average to August is used.
  7. Sea surface temperature (SST) is that in the Great Australian Bight and northwest Tasman Sea, averaged over the box: latitude 30-37°S, longitude 115-160°E, and expressed as degrees Celsius anomalies from the 1951-1980 mean, detrended about 2015. Cool SSTs correlate strongly with more snow, but unfortunately local SSTs are rising rapidly with global warming. The winter average is used. More at my post here. (I detrend the SSTs to make the factors in the model equation appear more sensible, and to make them more comparable with previous versions. Because of the model linearity, linear detrending doesn’t alter the regression outcome — it just shifts the SST trend effect across to the “calendar year” term.)
  8. Stratospheric aerosol optical thickness is the average for the Southern Hemisphere at 550 nm wavelength (green visible light). Large optical thicknesses from big volcanic eruptions correlate with big snow seasons. The winter average is used. More at my post here.

Leave a Reply