<h2>ERSST and Sea Ice</h2>

I use the NOAA ERSST V4 SST (Sea surface temperature) dataset as part of TempLS. It has the virtue of coming out promptly at the start of the month, and of course is the product of a lot of scientific work. But it has two nuisance aspects. One that I described last month, is that its 2x2° cells don't align very well with the coastal boundaries, and some repair action is needed. The other is the treatment of sea ice. ERSST returns values (if it can) for all non-land regions, and where there is sea ice, returns -1.8°C, which is the melting point of ice in sea, and so is indeed presumably the temperature of the water. But it isn't much use as a climate proxy there. Polar air over ice is often very much colder.

My aim is to mark these regions as no result, so that they will be interpolated, mostly from land. But that is complicated because, while -1.8 is clear enough, there are often temperatures close to that, which presumably mean mostly ice, or maybe ice for part of the month. So I have used a cut-off of -1°C.

I have been working recently with land masks to improve the accuracy of TempLS near coasts. My preferred version uses a triangular mesh with nodes at measurement points, so triangles will often be part land, part sea. It would be desirable to ensure that the implied interpolation uses land values for land locations. I'll post soon on how this can be done. But it sharpens the problem of sea ice, because the land mask doesn't recognise it. So I need to use some data, and ERSST is to hand, to mark this as land rather than sea.

So I have been reviewing the criterion for making that determination. I actually still think that -1°C is reasonable. To see that, I mapped the ERSST grid for Jan-Mar 2017 to show where the in-between regions are. I used WebGL.

It might seem that WebGL is overkill, since the polar regions can be easily projected onto 2D. But the WebGL facility makes it the easiest way. I just set all positive temperatures to zero, use the GRID type so I don't have to work out triangles, and then the color mapping automatically devotes the color range to the region of interest (and makes a color key).

So here is the plot (drag to see poles); in those months (radio buttons) it is Arctic that is of most interest. You can see that most of the region expected to be sea ice is in fact at -1.8C, and the fringe regions are intermediate. But there are also regions around the Canadian islands, for example, which show up as higher than -1.8, but would be expected to be frozen. A level of -1 seems to capture all that, without unduly modifying the front to clear ocean.

You'll notice the small white circle at the pole. That comes because ERSST goes from -88° to 88°. The grid triangles actually connect cell centers, which is the origin of the shading. It also means that if just one of the nodes is land, the triangle will be marked black. That is why the coast outlines aren't just rectangular.