+1 vote

I have been running Monte Carlo analysis of my models within EcoInvent 3.3, and they will give me hugely different results to those from analysis. 

For example, from Analysis I get a ReCiPe Midpoint(H) value for Climate Change (GWP100) of 15.09374 kgCO2eq.

Whereas from a 1,000 run Monte Carlo analysis, when I calculate the geometric mean after exporting the data, I get 76,376,446 kgCO2eq

This seems quite a difference. Does anyone find this?

asked by (240 points)
commented by (240 points)
Thanks... I've sent it as a google drive link
commented by (15.1k points)
Hi Tom, thanks for sharing. Probably this has something to do with negative signs in the input/output tables of processes. Is that possible? Do you have some negative numbers in your amount cells?
commented by (240 points)
I shouldn't do, unless there's an error somewhere. I've reduced the problem significantly by removing parameters from my processes, for some reason my parameters really upset the Monte Carlo. Now I'm getting a CO2eq impact of 15.09374 kgCO2eq still from a quick analysis, and about 3.6 kgCO2eq from the Monte Carlo. Still bad, but a massive improvement.

1 Answer

+1 vote
answered by (240 points)
I've started from scratch to see if I rebuild the model, to see what the problem is. It appears there's something wrong in the Monte Carlo if you have an input product which is also an output product. I use a lot of dummy processes to keep things clear in my models, especially when I want to correct an ecoinvent process which has the wrong output value. So this issue has caused me a lot of problems.

To replicate what I did, try this...

Create process A with an input of:
1kg carbon dioxide, liquid: market for carbon dioxide, liquid | carbon dioxide, liquid | Cutoff, U - RER (1;4;1;1;1)   

1m3 compressed air, 1200 kPa gauge: market for compressed air, 1200 kPa gauge | compressed air, 1200 kPa gauge | Cutoff, U - GLO (1;4;1;1;1)   

1kg potassium nitrate: market for potassium nitrate | potassium nitrate | Cutoff, U - GLO (1;4;1;1;1)
1kg water, deionised, from tap water, at user    : market for water, deionised, from tap water, at user | water, deionised, from tap water, at user | Cutoff, U - Europe without Switzerland (1;4;1;1;1)   

With an output flow you create of, say 1000 kg algae (these are made up numbers to demonstrate something).

Then make process B, which takes in 1000kg of algae and outputs 1000kg of algae
Create process C which takes in 1000kg of algae and outputs 1000kg of algae

Make sure A->B->C by setting the providers right.

The compare the Quick Analysis and Monte Carlo. They'll be different for B and C by a factor of 1000, because the models is based on the production of 1000kg of algae, and the Monte Carlo gets confused which process to take figures from if a flow goes through several processes.

Weird, right?

I tried this with EcoInvent 3.5 and OpenLCA 1.7.4, but it happens with other versions.

Anyone want to confirm this happens? Any idea why????
commented by (15.1k points)
Thanks for your information Tom. I posted your information on GitHub https://github.com/GreenDelta/olca-app/issues/60