Thoughts on possible next steps

[bpbond]

Ideas for discussion

Expanding the existing sensitivity analysis

• Include variability in climate data. Right now we're not doing this, but we could and probably should.
• Use real parameter uncertainty. Currently we use a fixed percentage for all parameters, but in reality some parameters are more uncertain than others, and we'd like to target parameters that are both uncertain and influential. We could work with SERC and COMPASS folks to come up with robust means and uncertainties for the Millenial parameters; maybe also use global databases (e.g. Soilgrids)?
• Focus on the flood event. @stephpenn1 already proposed looking at whether parameter sensitivity changes with flood disturbance. This is a cool idea--I don't know if I've ever seen anything like this, although I'd be surprised if its results differed at all from a straightforward sensitivity analysis.

Incorporating one or more of these expansions would absolutely produce a publishable paper! That said...

Fancier, more complete variance decomposition

From Alexey's paper:

Finally, we assessed which parameters contributed the most to parameter uncertainty for each model structure by performing a sensitivity and variance decomposition analysis following Dietze et al. (2014), LeBauer et al. (2013), and Raczka et al. (2018). Briefly, this analysis uses a spline to estimate the relationship between individual parameters and output variables (averaged across the entire simulation, 1902–1999) and then uses these splines to predict the variance in the output variable attributable to individual parameters (“partial variance”; see LeBauer et al., 2013 for more details).

From Dietze et al. 2014:

We can then look at the effect of each parameter after accounting for both the uncertainty of that parameter and the effect of other parameters. This is "partial variance":

This is akin to @leeyap 's relative importance analysis, or the idea of Type III Sum of Squares, but accounts for parameter uncertainty. It's pretty cutting-edge.

[bpbond]

And reminding us about MAAT

Hi Ben, Stephanie, (& Matt, Rose), 

Good seeing/meeting you at the PI meeting and hearing about some of your work. I'm following up re MAAT and Millennial and other soil C models. 

I'd be happy to help get you up and running with MAAT and Millennial in MAAT, and try to deal with any issues you might run into. The code is open-source and more or less up to date. Though still very much a work in progress, there is a lot of utility in where it's at now.   

I'm including Matt and Rose who've done all the work developing the soil models and I'm sure will be happy to help too. 

Feel free to give us a shout whenever you have the headspace to start thinking about this and we can take it from there. 

All the best, 

Anthony