Open lawinslow opened 8 years ago
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.
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.
I prefer the more conservative approach too.
From Gretchen: