Guest post by Steven Goddard
I have been noticing in recent weeks that NSIDC extent is much closer to their 1979-2000 mean than NANSEN is to their 1979-2007 mean. This is counter-intuitive, because the NANSEN mean should be relatively lower than NSIDC – as NANSEN’s mean includes the low extent years of the 2001-2007 period. Those low years should have the effect of lowering the mean, and as a result I would expect the NANSEN current extent to be equal to or above the 1979-2007 mean.
I overlaid the NANSEN graph on top of the NSIDC graph below, and it is easy to see how large the discrepancy is. In fact, the NSIDC mean sits at about one standard deviation below the NANSEN mean – which makes little sense given their base time periods. It should be the opposite way.
(Note – the NANSEN and NSIDC measuring systems are not identical, and I had to make a shift along the Y-axis to line them up. However, the X and Y scales are identical for both graphs in the overlay image.)
Nansen uses a different algorithm to calculate the sea ice extent. The algorithms differ in the way combine the raw data together to estimate extent. As long as one uses the same algorithm, the stories are all the same, but the details can differ, more so at certain times of year. When there is a diffuse, broken up ice edge and melt is starting is one such time.
I suspect the Bering Sea is probably the region resulting in most of the differences. While our algorithm shows the region has mostly “ice-covered” the ice cover there is very fragmented, broken-up, and thin.
The other thing that’s important to mention is that I was referring simply to discrepancy between how close the current lines are to climatology. However, there is also generally an “offset” between algorithm outputs – a bias or mean difference between the algorithms that is fairly consistent throughout the record. That is why NSIDC’s climatology is different than the Nansen climatology.
The important thing to remember is that there is a good consistent record from the passive microwave data as long as you consistently use the same algorithm and the same processing. But you can’t mix and match products.