Dear team,
It is really great that we can access the Pedigree scores directly now, and that we can calculate the additional uncertainty associated with this additional uncertainty. I have two observations, one of which may be a problem:
1) In a userdefined dataset, if one defines a distribution in the "Uncertainty" field, and then clicks in the "Data quality entry" field, the entered uncertainty information is not carried over in the "basic uncertainty". That is fine, once we know. I guess that since it is possible to export the "total" uncertainty to the uncertainty field, we would end up with a loop if it was then fed back in the Data quality entry field. So all good.
2) When looking at ecoinvent data, however, it seems that the uncertainty that is accounted for is only the basic uncertainty, without any consideration for the additional uncertianty associated with the pedigree scores. Looking at a very precise example: cryolite input in `aluminium production, primary, ingot  aluminium, primary, ingot  cutoff, U  CAQC`. In the original ecoSpold, we have:
Lognormal (Geometric mean=0.0004, Variance of logtransformed data=0.0006, Arithmetic mean of logtransformed data=7,824, Standard deviation=1.0502, CI/2wP, half range of confidence interval=1.2286, Variance of data with pedigree=0.0106) Pedigree matrix: 1 1 4 5 1
==>The basic uncertainty, expressed as the variance of the underlying normal, is 0.0006. As gsigma: 1.0247973619913604. This is the value reported in openLCA "uncertainty" field for this exchange.
==> The additional uncertainty, expressed as the variance of the underlying normal, is 0.01 (using the ecoinvent pedigree matrix). As gsigma, this is 1.1051709180756477.
==> The total uncertainty, expressed as the variance of the underlying normal, 0.0106, or 1.1084429706226131 as gsigma. Note that this is the value that is reported as gsigma in the "Data quality entry" page as total uncertainty, and that basic uncertainty is assumed to be nil in this page.
So, to sum up: (1) the uncertainty considered in uncertainty calculation seems to be uncertainty without pedigree, (2) the basic uncertanty displayed in the "Data Quality entry" is not correct (it is displayd as 1), and (3) the geosigma in the "Data Quality entry" is indeed the total, but is not calculated from the displayed data and is not used in the uncertainty field.
Hope this is clear and that it helps :/
Pascal
Uncertainty data in openLCA
Re: Uncertainty data in openLCA
Thank you very much Pascal, we will look into this!
Re: Uncertainty data in openLCA
Hi Pascal,
Many thanks for the detailed explanations.
to 1) There is a separate field for the "basic uncertainty" at the bottom of the window that pops up when you edit the Pedigree scores of an exchange. When you click on the button "Use as uncertainty value" the calculated "total uncertainty" is copied to the geometric standard deviation of the uncertainty distribution. Thus, the geometric standard deviation of the uncertainty distribution is intended to contain the "total uncertainty" (and not the basic uncertainty) and is independent from the values entered in the Pedigree scores.
to 2) When I wrote the EcoSpold 2 import I looked at the documentation of the uncertainty fields. In the XML file, your example looks like this:
The documentation of the `variance` field is (see http://www.ecoinvent.org/files/document ... format.zip):
With your explanation it is now clear that this is just the "basic uncertainty" and that we should use the field "varianceWithPedigreeUncertainty" instead ...
I will fix the EcoSpold 2 import and then we have to think how we could patch existing databases...
Thanks again,
Michael
Many thanks for the detailed explanations.
to 1) There is a separate field for the "basic uncertainty" at the bottom of the window that pops up when you edit the Pedigree scores of an exchange. When you click on the button "Use as uncertainty value" the calculated "total uncertainty" is copied to the geometric standard deviation of the uncertainty distribution. Thus, the geometric standard deviation of the uncertainty distribution is intended to contain the "total uncertainty" (and not the basic uncertainty) and is independent from the values entered in the Pedigree scores.
to 2) When I wrote the EcoSpold 2 import I looked at the documentation of the uncertainty fields. In the XML file, your example looks like this:
Code: Select all
<lognormal
meanValue="0.0004"
mu="7.824"
variance="0.0006"
varianceWithPedigreeUncertainty="0.0106" />
From the documentation, I understood that this is the value that I should use as distribution parameter (see my documentation of the conversion method here: https://github.com/GreenDelta/olcamodu ... r.java#L81)."Unbiased variance of the underlying normal distribution"
With your explanation it is now clear that this is just the "basic uncertainty" and that we should use the field "varianceWithPedigreeUncertainty" instead ...
I will fix the EcoSpold 2 import and then we have to think how we could patch existing databases...
Thanks again,
Michael

 Posts: 30
 Joined: 19 Sep 2014 21:31
Re: Uncertainty data in openLCA
Thanks for the speedy responses.
I will be on the lookout for the patch for exising databases (especially v3.3).
Pascal
I will be on the lookout for the patch for exising databases (especially v3.3).
Pascal
Who is online
Users browsing this forum: No registered users and 1 guest