[2015-09-21] Lakes are "failing" but not giving NA's

[lawinslow]

From Gretchen:

It also appears that some years are failing but not producing NA’s:

        WBIC count.GDD
197   187500        29
470   504700        17
616   741700        10
732   837300        10
1070 1571800         3
1902 2484400         3
2022 2624800        10
2038 2633700         9
2257 2858000         3

In the previous run, all lakes that were in there had GDD values for all years (N=34).

[lawinslow]

These all seem to be shallow lakes that are getting a kind of "stuck depth".

After digging further, during the winter, ice is taking up most of the volume of the surface layer. Once this layer gets thin enough, it is combined with the layer above it. Then, it seems to get stuck there, I suspect because our min and max layer thicknesses are really close to each other, 0.2 and 0.3 meters respectively. This means that the single layer is maxed out at 0.3 meters in thickness, but it has trouble splitting into two layers (which both need to be > min_layer_thickness) which would then allow the layers to grow in thickness.

Reducing the min_layer_thickness to 0.1 seems to help considerably.

Checking now if we can apply it universally without changing our model results.

[lawinslow]

Ok, if we cut min_layer_thick to 0.1 universally across all lakes it doesn't hurt us much, though bias gets worse.

Option	RMSE	Bias
min_layer_thick=0.2 (used previously)	2.58	0.050
min_layer_thick=0.1	2.57	0.059

If we take a more conservative approach, and just cut min_layer_thick for shallow lakes (where max_layer_thick is set to 0.3), our results change less.

Option	RMSE	Bias
min_layer_thick=0.2 (used previously)	2.58	0.050
min_layer_thick 0.2 or 0.1 depending on depth	2.57	0.051

Unless there are objections (@jread-usgs?) then I'm going with the conservative option.

[jordansread]

I prefer the more conservative approach too.