I’m wondering if it’s possible to apply to stochastic approach to carbon pools per land use to assess the uncertainty of input parameters (i.e. an all-at-once sensitivity approach)?
For example, a python script to perform monte carlo sampling bounded by upper and lower estimates of carbon pools?
I plan to apply a similar global sensitivity approach to the NDR and uban flood models to assess the individual parameter contribution to uncertainty as well as the impacts of the model input interaction.
I’m wondering a similar approach was applied before because what I found so far are local sensitivity analyses.
Very curious to hear your thoughts
Welcome to the forums @kal !
Could you clarify what you mean by an “all-at-once sensitivity approach”?
In general, though, a sensitivity analysis on the InVEST carbon model should be straightforward to set up because the model itself is so simple. The other models would be much more interesting because the models are a bit more complicated.
I wonder if @swolny or @jesseG might have some further suggestions or experience with sensitivity analyses with these models?
Before proceeding too far, I highly recommend reviewing Hamel and Bryant 2017, titled “Uncertainty assessment in ecosystem services analyses: Seven challenges and practical responses”. Also, check out the InVEST User Guide, where some model’s chapters provide guidance and resources on performing sensitivity analyses.
Any sensitivity analyses I’ve performed on InVEST models were focused on local or regional areas, not globally.
Hello @jdouglass & @jesseG
Thank you for your prompt comments!
I’m referring to global sensitivity analysis (e.g. Sobol sensitivity analysis) using the SALib python package. For instance, the carbon pools per land use would be varied all at the same time to identify the influential parameters, rank their sensitivity and determine the impact of input parameters’ interaction on the model’s overall uncertainty. I plan to apply the same approach to the urban flood and NDR models.
My area of focus is local (a metropolitan area) mix of building and public park land uses. I’ve read Hamel and Bryant 2017 paper and it’s inspired me to follow this approach.
Global sensitivity analysis has been applied to other ES models but I’m wondering if it was used with the above mentioned InVEST models.