|
|
h1. Bayesian uncertainty analysis
|
|
|
|
|
|
{{>toc}}
|
|
|
|
|
|
h2. Description
|
|
|
|
|
|
$x^{yield}_{t}$
|
|
|
|
|
|
|
|
|
Parameter determination for processes in LPJmL is mostly achieved by literature values. In case, data are available, the estimation of uncertainties and evaluation of current parameterization is possible by Bayesian analysis. Here, we give an example how to use data for a better grass parametrization.
|
|
|
|
|
|
|
|
|
h2. Details
|
|
|
|
|
|
a more detailed description of it (data sources, data preparation, meaning, usage...).
|
|
|
Refer to references if you have any using footnotes[1]
|
|
|
|
|
|
h2. Technical Note
|
|
|
|
|
|
What you need are
|
|
|
* data
|
|
|
* the R package FME[1]
|
|
|
|
|
|
Following the procedure there, you can
|
|
|
* define a cost function that gives the deviation of the model output from the observations
|
|
|
* determine the local sensitivity to a large set of parameters
|
|
|
* evaluate the most prominent parameters that should be included in the Monte Carlo simulations by determining the collinearity
|
|
|
* fit the model to the data using the chosen subset of parameters
|
|
|
* perform the Markov chain Monte-Carlo simulations
|
|
|
* evaluate the uncertainty of the model output due to the parameter values accepted in the Monte Carlo chain.
|
|
|
|
|
|
|
|
|
h2. Developer(s)
|
|
|
|
|
|
Susanne Rolinski, Anja Rammig, Werner von Bloh
|
|
|
|
|
|
h2. See Also
|
|
|
|
|
|
[[Upscaling]], [[Downscaling]], [[Wiki]], [[Mathematical Description]], [[Missing wiki page]]
|
|
|
|
|
|
Links to other Wiki pages, that are related. It doesn't matter if the wiki page already exists or not.
|
|
|
Also link pages that do not exist yet! Links to existing pages are written blue,
|
|
|
links to non-existing pages are writtn in red.
|
|
|
|
|
|
|
|
|
h2. References
|
|
|
|
|
|
fn1. K. Soetaert and T. Petzoldt (2010): Inverse Modelling, Sensitivity and Monte Carlo Analysis in R Using Package FME. Journal of Statistical Software, 33, 3, 1-28.
|
|
|
|