Calibrate soil co2 single site

[AlexisRenchon]

TO DO:

• latest commit makes calibration model default looks good (like default run dk), but uses the netcdf LAI (not MODIS LAI) - which one do we want?
• think about good values for RAI and SAI, check if NEE still looks reasonable
• re-run calibration once mean params and prior spread looks good

wip

[kmdeck]

we may need to further scale nee_var by a constant to wait terms in the loss function equally. In our offline run with default parameters, the weighted RMSE for need is about 20x larger than the weighted RMSE for the heat fluxes.

julia> x = (lhf_model_at_obs .- qle_obs)./qle_unc;
julia> @show sqrt(mean(x.^2)), mean(x)
(6.310226345552107, -1.9822982854949414)

julia> x = (shf_model_at_obs .- qh_obs)./qh_unc;
julia> @show sqrt(mean(x.^2)), mean(x)
(5.226855236254443, 2.479323425387818)

julia> x = (nee_model_at_obs .- nee_obs)./nee_unc;
julia> @show sqrt(mean(x.^2)), mean(x)
(108.79570442024037, 19.221651412031065)

That is, we might want to scale the NEE uncertainty by 108/620.

[AlexisRenchon]

I'll leave this open for consideration... I'll try running as is for now