Sea surface temperature and snow

It’s hardly news to suggest that our snow season is affected by Southern Ocean sea surface temperatures (SSTs).  The evidence is mostly anecdotal: some claim it’s the Great Australian Bight which controls things, others say it’s the southern Indian Ocean.  Here I take a Hadley Centre sea surface temperature reconstruction¹ and compare SST averages in various latitude-longitude boxes with season peak snow depths², to find out.

The HadISST dataset is a 1-degree resolution monthly sea surface temperature reconstruction extending back to 1870, mostly interpolated from measurements made by ships at sea (there’s a huge dataset), augmented in recent years by buoy measurements and satellite remote sensing.  It’s not perfect (especially in the Southern Ocean) but it’s adequate for our purposes.

After trying dozens of latitude-longitude boxes to average SSTs across, I found the strongest winter correlation with the one shown above: latitude 30-37°S; longitude 115-160°E. Yeah, that’s right, the Great Australian Bight plus the north-western Tasman Sea (land areas are excluded from the average). The correlation coeficient is 0.15 which is pretty modest, but a fair bit higher than any of the others used in my snow prediction formula. It looks like this (using winter SST anomalies vs the 1951-1980 mean):



Latitude-longitude boxes including more of the Southern Ocean, for example south and west of Perth, show much lower correlations (for example 30-50°S / 90-120°E gives an r² of only 0.03).

Southern SSTs of course show a substantial uptrend (about +0.8°C since 1954), just as the peak snow depths show a substantial downtrend (nearly half a metre lost since 1954).  To compare the two, it helps to remove the trends and just look at the variations about the trend lines.   (Doing so barely alters the correlation coefficient, which is dominated by the year-to-year variations.)  In the chart below I’ve reversed the temperature scale to aid comparison:



It’s apparent that the southern SST correlation explains some peaks and troughs well, but completely misses others — like the big Pinatubo seasons (1991 & 1992) and the woeful 2006.  I’ll be incorporating it into a revised prediction formula this year.


1.  UK Met Office’s Hadley Centre Sea Ice and Sea Surface Temperature data set (HadISST1):

2.  From the Spencers Creek snow depth record (near Charlotte Pass, Australia) from Snowy Hydro Limited, 1954-2013.