Interpretation, sensitivity and uncertainty

Click to edit Master subtitle style

By Carmen Alvarado

Copyright PR Consultants B.V

CA.3i.v1.2006.1

Sensitivity checks

Consistency check Contribution analysis Anomality check (Mis)match between inventory and impact assessment Appropriateness of impact assessment Validity of subjective choices Uncertainty assessment
Check goal and scope: Have I applied the principles described here in a consistent and realistic way? What are the processes with the highest load; usually only 10 processes really matter Anything strange???? Use common sense Many LCI results are not recognised in impact assessment methods. Have important LCI results been missed? You should not use an impact assessment method without land-use in an LCA on bio-plastic.... Subjective choice cannot be avoided. Check if your conclusions are dependent on these choices by recalculating, using different assumptions Monte Carlo analysis (next slides) CA.3i.v1.2006.2

Copyright PR Consultants B.V

Sensitivity versus Uncertainty

Sensitivity

Choice of data

Uncertainty

Reliability (measured, estimated) Library Sample size Process Completeness/ Choice of impact assessment methods Representativity Missing data Geographical, technical and temporal deviations Calculate two or more different scenarios Indicate uncertainty range and distribution and calculate uncertainty
Copyright PR Consultants B.V CA.3i.v1.2006.3

Sensitivity for subjective choices

In an LCA on paper or wood products.

Is the net-uptake of CO2 in forests included yes/no Recycling process belong to the end of life, and the new iron produced can be seen as a benefit The recycling process is part of a next lifecycle National grids European average Marginal electricity Have I choosen a proper alternative in system boundary expansion What if I use mass in stead of economic value in multi output allocation
CA.3i.v1.2006.4

In an LCA on steel cans:

On electricity in Europe.

On allocation

Copyright PR Consultants B.V

Uncertainty

There are many sources for data uncertainty

Uncertainty in measurements and estimates Representatives of a process

I would have liked to used data on a specific truck with a specific load factor, but I only have a generic average I use a truck, but maybe some transport is by train also

Uncertainty of consumer behaviour in use and disposal phase Uncertainty on future disposal scenarios Uncertainty in impact assessment factors etc.

Define uncertainty ranges, and apply Monte Carlo.

CA.3i.v1.2006.5

Copyright PR Consultants B.V

Benefits of using uncertainty

Eco-invent unit process database has specified uncertainty for all LCI parameters; no extra work Add uncertainty in your own data, like:

Estimated data Use phase and EOL data, that is often very uncertain

Run Monte Carlo simulation on your first screening and determine how much certainty you need to add, not more, not less. What is the minimum we need to know for decision support
CA.3i.v1.2006.6

Copyright PR Consultants B.V

Monte Carlo analysis

Define a distribution and standard deviation on (every) input parameter. Randomly select a series of values according to the distribution Recalculate the LCA for every parameter Store each result, and plot the distribution of results

Copyright PR Consultants B.V

CA.3i.v1.2006.7

Result of a single LCI parameter, Chromium emission to air

Copyright PR Consultants B.V

CA.3i.v1.2006.8

Results of an analysis

Copyright PR Consultants B.V

CA.3i.v1.2006.9

Result of an analysis, single score

Copyright PR Consultants B.V

CA.3i.v1.2006.10

Correlated uncertainties

Component A and B use the same electricity, the uncertainty per run is the same Correlation must be taken into account. Only Unit processes can be used
component A

component A

Component B

metal

electricity

plastic

Copyright PR Consultants B.V

CA.3i.v1.2006.11

Comparison without correlation: Misleading

Comparison of product A made of 10 kg of steel and product B made out of 11 kg of the same steel

Product A, 10 kg steel
Copyright PR Consultants B.V

Product B, 11 kg steel
CA.3i.v1.2006.12

The difference between A and B

The previous result cannot be correct; the difference is a kg of steel, however uncertain the impact of steel production is. SimaPro presents the difference in impact (A-B); In this result, all outcomes have a negative value, so we are 100% sure that there is indeed a difference between A and B

Copyright PR Consultants B.V

CA.3i.v1.2006.13

Case: Comparing Converter and Electro steel (Eco-invent, EI99 H/A)

Land use Fossil fuels

Converter steel
Copyright PR Consultants B.V

Electro steel
CA.3i.v1.2006.14

Case: Comparing Converter and Electro steel (Eco-invent, EI99 H/A)

F o s si l f u e ls
Number of times converter > electro

L a n d u s e

N u m b er of ti m es c o n v er te r < el e ct ro

Copyright PR Consultants B.V

CA.3i.v1.2006.15

Difference between Converter and Electro steel (EI99 H/A)

Difference between Electro steel and converter steel is negative; this is the case in 30% of the runs
Copyright PR Consultants B.V CA.3i.v1.2006.16

Conclusions

With Monte Carlo, we can finally understand if differences between products are real Monte Carlo simplifies LCA, as you can determine if your uncertainties need to be reduced.

Screening data may already be enough to get significant results. No need to collect additional data if the results are significant

Without taking into account correlations, Monte Carlo results can be very misleading. Further work is needed to solve the A/B problem Uncertainty on impact assessment must be further developed.

Copyright PR Consultants B.V

CA.3i.v1.2006.17

Defining uncertainty ranges in practice

Uncertainties have a distribution Range is usually defined as standard deviation (SD) Highest probability is often used as the best guess value Results also have a standard deviation, and a medium or average.

Medium is the middle value, so if you made 100 samples, the 50th sample is the medium Average is the weighted average of all samples Average can be heavily influenced by outliers

Copyright PR Consultants B.V

CA.3i.v1.2006.18

Distributions

Log Normal distribution Industrial Processes & Nature Enter best guess and (standard deviation)2 to cover 95% interval Normal distribution Random samples Enter best guess and 2*(standard deviation) to cover 95% interval Triangular Simple version of normal distribution; has no extreme values. Enter best guess plus min and max value Uniform No preference for an average Enter min and max value
Best guess

SD Best guess

SD

SD

SD

Best guess

Min

Max

Min

Max

Copyright PR Consultants B.V

CA.3i.v1.2006.19

How uncertainties are defined in Ecoinvent

All distributions are lognormal Seven factors determine the uncertainty
SDg 95 = = exp
2 g [ln(U1 )]2 +[ln(U 2 )]2 +[ln(U 3 )]2 +[ln(U 4 )]2 +[ln(U 5 )]2 +[ln(U 6 )]2 +[ln(U b )]2

Six of these (U1-U6) are determined by Pedigree matrix (next slide) Eco-invented always document scoring in pedigree matrix (1,2,1,1,4,2) Basic uncertainty uB is emission specific

Copyright PR Consultants B.V

CA.3i.v1.2006.20

Pedigree matrix
Score: Reliability 1
Verified data based on measurements

2
Verified data partly based on assumptions OR non -verified data based on measurements 1.05 Representative data from >50% of the sites relevant for the market considered over an adequate period to even out normal fluctuations 1.02 Less than 6 years of difference to our reference year (2000)

3
Non-verified data partly based on qualified estimates

4
Qualified estimate (e.g. by industrial expert) ; data derived from theoretical information (stoichiometry, enthalpy, etc.) 1.20 Representative data from only one site relevant for the market considered OR some sites but from shorter periods 1.10 Less than 15 years of difference to our reference year (2000) 1.20

5
Non-qualified estimate

U1

1.00 Representative data from all sites relevant for the market considered over an adequate period to even out norma l fluctuations 1.00 Less than 3 years of difference to our reference year (2000) 1.00 Data from area under study

1.10 Representative data from only some sites (<<50%) relevant for the market considered OR >50% of sites but fr om shorter periods 1.05 Less than 10 years of difference to our reference year (2000)

1.50 Representativeness unknown or data from a small number of sites AND from shorter periods 1.20 Age of data unknown or more than 15 years of difference to our reference year (2000) 1.50 Data from unknown OR distinctly different area (north america instead of middle east, OECD -Europe instead of Russia) 1.10

Temporal correlation

U3

Completenes s

U2

1.03 Average data from larger area in which the area under study is included

1.10 Data from smaller area than area under study, or from similar area

Geographical correlation

U4

1.00 Data from enter prises, processes and materials under study (i.e. identical technology) 1.00 >100, continous measurement, balance of purchased p roducts 1.00

1.01

1.02 Data on related processes or materials but same technology, OR Data from processes and materials under study but from different technology 1.20 Data on related processes or materials but d ifferent technology, OR data on laboratory scale processes and same technology 1.50 >=3

Further technological correlation

Data on related processes or materials but on laboratory scale of different technology

U5

2.00 unknown

>20

Sample size

> 10, aggregated figure in env. report 1.05

U6

1.02

1.10

1.20

Copyright PR Consultants B.V

CA.3i.v1.2006.21

Basic uncertainty
input / output group demand of: thermal energy electricity semi-finished products working materials transport services waste treatment services Infrastructure resources: primary energy carriers metals, salts land use, occupation land use, transformation waste heat: emission to air, water, and soil emission to water of: BOD, COD, DOC, TOC inorganic compounds (NH PO4, NO3, Cl, Na etc.) 4, individual hydrocarbons, PAH heavy metals from agriculture: NO PO4 3, from agriculture: heavy metals from agriculture: pesticides radionucleides 1.50 1.50 3.00 5.00 1.50 1.80 1.50 3.00 1.05 1.05 1.05 1.50 2.00 1.05 1.05 1.05 1.05 2.00 1.05 3.00 Ub input / output group emission to air of: CO2 SO2 combustion: NOX, NMVOC total, methane, N and NH3 2O Combustion: CO Combustion: individual hydrocarbons, TSM Combustion: PM10 Combustion: PM2.5 combustion: polycyclic aromatichydrocarbons (PAH) Combustion: heavy metals process emissions: individual VOCs process emissions: CO 2 process emissions: TSM process emissions: PM10 process emissions: PM2.5 from agriculture: CH NH3 4, from agriculture: N2O, NOX radionucleides (e.g., Radon -222) process emissions: other inorganic emissions Emission to soil of: oil, hydrocarbon total Pesticides heavy metals radionucleides 1.50 1.20 1.50 3.00 1.05 1.05 1.50 5.00 1.50 2.00 3.00 3.00 5.00 2.00 1.05 1.50 2.00 3.00 1.20 1.40 3.00 1.50 Ub

Copyright PR Consultants B.V

CA.3i.v1.2006.22

Wrap up

With Monte Carlo, we can finally understand if differences between products are real Monte Carlo simplifies LCA, as you can determine if your uncertainties need to be reduced.

Screening data may already be enough to get significant results. No need to collect additional data if the results are significant

Without taking into account correlations, Monte Carlo results can be very misleading. Uncertainty on impact assessment must be further developed.
CA.3i.v1.2006.23

Copyright PR Consultants B.V