I am doing source apportionment study for PM2.5 and PM10 pollutants. TO do that, I use EPA PMF 5.0 model. There uncertainty data is to feed as one of the input.
Can any body help me how to estimate uncertainty of my data?
The method prescribed in model for uncertainty estimation is
UNC = (5/6)*MDL
MDL = Species-specific method detection limit
So what is MDL refers here?
Hi David
Yes your correct.
But in EPA PMF manual it is mentioned MDL as
"Species-specific method detection limit"
Maybe these references are also helpful:
- Paatero, P., 1997. Least squares formulation of robust non-negative factor analysis
Paatero, P., Tapper, U., 1994. Positive matrix factorization: A non-negative factor model with optimal utilization of error estimates of data values. Envirometrics 5, 111–126
Best,
Sascha
Dear Rajendran
First, you have to go for pretreatment of measured concentration. Means if you have missing values, determined values, and BDL values. Missing value can be replaced by geometric mean, BDL can be replaced by MDL/2. Method detection limit (MDL) will vary with the instrument which you are characterising species.
For uncertainty, determined value of can replaced by Pij *Xij+MDL/3 where Pij the uncertainty proportion which was estimated in Kim and Hopke 2004a, b. Missing value will be replaced by 4*geometric mean, and BDL value can be replaced by (5/6)*MDL.
Using (5/6)*MDL in uncertainty file only for BDL value came in measured concentration, not for all.
Dear Jaiprakash
From your suggestions,
If my instruments lower detection limits is say 2 micro gram / m3
then the MDL of the instrument is 2 micro.gram / m3
so uncertainity is (5/6)*2 = 1.67 micro.gram / m3
But here for each and every sample concentration, we have to feed uncertainty,
so how do I estimate that?
Can you share the paper of kim and hopke 2004.
The uncertainty ought to be determined by comparing the value as measured by the particular instrument (or method) versus a well characterized standard, e.g., NIST standard reference materials, or against a reference method. Otherwise, one should at least attempt evaluation by propagating the measured uncertainties of all of the components of the measurement device.
Dear Marimuthu Rajendran Ezhilkumarn
In matrix will be as number of samples and no. of species, I think you know
May be, some species concentration will below detection limit (BDL) so this species concentration can be replaced by (5/6)*MDL.
I am sharing paper of mine which work on PMF and also sharing paper of Kim and Hopke.
Sorry for late reply.
Dear Ezhilkumar Marimuthu Rajendran , I've found this conversation about EPA PMF 5.0 and I'm looking for an information about the last version of EPA's user guide : it was written in 2014,this is the last version or there is a new it?
Thank you for the answer
Dear Nappi,
This is latest version. u can check on the official website.
Dear Marimuthu,
I am using PMF also to do source apportionment. Until now i still meet the challenge to determine MDL because the laboratory where I analyzed my sample was not provide this data. Is there any recommendation to find this data ?
thank you
Best regards
Dear Seny, u can repeat you sample and see the how much varaition from previous. So, it would be uncertainity of species.
Or
U can see the Kim and Hopke 2004 paper, they reported uncertainity of species.
Sir, MDL refers to Method detection limit. It is the limit of the procedure or instrument with which you are doing your chemical characterization. It can be different for different elements, even done through same instrument. You can check that from official instrument website. It is specific for a particular instrument.
To feed the uncertainty, one can do the the replicate analysis for measuring the concentration of the species and the obtained standard deviation can be used as an uncertainty.
When quantification use a linear regression , MDL can be estimated analyzing 10 blank samples, then standard desviation (s) is calculated and the expresion DLM= 3s/m is applied, where m is the slope of equation of regression
- Badges
- Science method