Statistical Analysis of Groundwater Model Outcome

#1
Hello
I have simulated a groundwater model on an integrated numerical model software (Mike She) and calibrated it.
My next step is to weigh the effect of each parameter used in the model by simply varying the parameters and running the model again then extracting the results file which in this situation is the spring discharge.
Any ideas of how to measure the weight of the varied parameters all together and how to assign a unified weight factor for each parameter on a certain scale.

Thank you
 

hlsmith

Not a robit
#2
If you are simulating these data, aren't you already setting the weights? Are data correlate or heterogeneity in interactions that make it too difficult to use the original simulated weights. Or is this like an agent-based model that you let loose to evolve?
 
#3
If you are simulating these data, aren't you already setting the weights? Are data correlate or heterogeneity in interactions that make it too difficult to use the original simulated weights. Or is this like an agent-based model that you let loose to evolve?
Actually it is a physically based model, where all parameters are actual (field based or lab for example soil porosity, precipitation, geology...etc) and the simulations basically run based on partial differential equations that represent groundwater flow from precipitation till it reaches the water table.
 

hlsmith

Not a robit
#4
Ahh, the last podcast episode on Science Friday (on NPR) talks about comparable models in weather forecasting of storm surges. I have not done these simulations, but I get the principles, and you put the weights in, so you can as you mentioned fiddle with them and see their effects (model sensitivities to distributions, etc.). Blackbox algorithms do this with Shapley values I believe. So you are now just looking for something like a variable importance metric with values for the included inputs, yes? Hmm, I don't have a solution for you, but it seems doable if YOU are selecting the probabilistic values and the functions (relationships between them). Perhaps look to the literature for a comparable published project and see what they have in the Methods sections or contact the corresponding author if you are still stumped. Sorry I am not a bigger help on this problem.
 
#5
Thank you hlsmith.
In simple terms I am varying the parameters in the model within a physical range and extracting the model output at the end what I need a general method to assess the weight of each parameters taking into account the variation of the model outcome results.