Trying to Tune Alpha and Change Parameter Bounds

[ptrscll]

I tried to tune alpha and also tried using an even larger box for parameter bounds. Ultimately, these efforts produced some minor improvements in temperature accuracy at the cost of even worse performance on the CO2 data (since so far I've still been using NMSE with uncertainty as my error function). These trends can be seen in the plot below:

Interestingly, the optimal values of alpha were all quite close to 1, and for the larger box sizes, they were actually slightly less than 1. Based on this, would it make sense to try doing a sensitivity test for alpha like we did for S and ocean heat diffusivity?

(This PR is relevant to #8 )

[kdorheim]

@ptrscll yes let's try doing an sensitivity analysis with alpha!

[ptrscll]

Just did a sensitivity analysis - it looks like changing alpha is working as expected and (quite substantially) affecting Hector's output, as shown in this graph:

Based on the graph, I'm guessing the reason why optimal alpha values haven't been super high is that higher alpha values produce very cold and unrealistic temperatures for the recent past, offsetting any error reduction from cooler temperatures in the earlier historical period.