Environmental Life Cycle

Assessment
CEE 12-714 / EPP 19-714

Lecture 11: Modeling Uncertainty in LCA

February 21, 2018
 Administrivia
2

• HW3 due tomorrow, Feb 22nd

 Significant figures are important
 Watch your formatting – make it easy for grading!
 Comma format large numbers
 $ signs if relevant
 Page numbers and headers
 Last assignment with formatting points
• HW4 will be released tomorrow

2
More administrivia
• Monday, we start a 4-part series of lectures on
input output modeling, including an important
(fun) class exercise
 See Canvas readings
 Be on time!
• Tuesday, office hours will be an optional lecture
on stochastic modeling (more details than today)

3
 Group projects
4

• Proposals due no later than Monday, Feb 26,

midnight
 Submit sooner if you want earlier feedback
 I will provide feedback to you ASAP
 Canvas project deliverables opened, see for additional
details and rubrics
• Schedule team meetings with me to discuss
your draft proposals (optional)
 Team 4, today, 1 pm

4
Course trajectory
1. Introductions 10. Uncertainty
2. Life cycle thinking 11. Input-output LCA
3. Quantitative methods and 12. Process-matrix LCA
life cycle cost analysis
13. Hybrid LCA
4. ISO LCA framework
14. Impact assessment
5. Critical review
15. Structural path analysis
6. LCA data sources
16. Professional responsibility
7. Life cycle inventory
17. Carbon footprinting
8. SimaPro
18. LCA for big decisions
9. Handling multifunction
19. Project presentations
systems

5
 ISO 14040: Figure 1

Phases of an LCA
6

• Goal and scope

definition

• Inventory

• Impact assessment

• Interpretation
6
Uncertainty
Types, sources,
how will we deal
with it?
7
“A decision made without taking uncertainty
into account is barely worth calling a
decision.” Wilson (1985)

8
 From Lecture 3

Uncertainty vs. Variability

9

• Uncertainty: exists because of ignorance or lack

of data
 Likely reducible with further study
• Variability: exists because of heterogeneity or
diversity
 Unlikely to be reducible with more study

• We assume they are the same, and call them

both uncertainty
• And do “uncertainty analysis”
Adapted from Margrit von Braun, Univ of Idaho 9
 Deterministic Context in LCA

• Most studies still deterministic (point

estimates, no uncertainty)
 Result: a single LCI value (or list of single values)
• Typical use: hot spot or comparison
 Hot spot: which LCI value is bigger than all others?
 Comparison: which is lower, A or B?

• Typical analysis: Simply A < B

• Typical result: It depends

10
How are comparisons done in other
domains with data?
• How do we check whether A < B?
 We set up a hypothesis and apply statistical tests (e.g.,
t-test) and assess significance

• Why doesn’t LCA do this?

 Easy answer: there hasn’t been sufficient data, or
sufficient demand for this level of technical complexity.
That is changing.

11
ISO 14040 says…
• LCA addresses potential environmental
impacts; LCA does not predict absolute or
precise environmental impacts due to:
 relative expression of potential environmental
impacts to a reference unit
 integration of environmental data over space and
time
 inherent uncertainty in modelling environmental
impacts
 some possible environmental impacts are clearly
future impacts

12
ISO 14044 says…
• Data quality requirements should address
uncertainty of information
 data, models, and assumptions
• “Uncertainty analysis and sensitivity analysis
shall be done for comparative studies intended
for public release.”

13
Text from an actual LCA

How well does this statement

handle uncertainty?
14
Paper vs. plastic
• Lave et al compared energy use (electricity only) of
plastic and paper cups (1995)
• Plastic cup consumed ~50% less electricity
 plastic cup: 4,400 kWh
 paper cup: 8,600 kWh
• Updated to consider uncertainty (and total energy)
(Chen, 2017):
 plastic cup: 0.3 – 0.7 TJ
 paper cup: 0.3 – 0.4 TJ
 Original conclusions change
 Overlapping range suggest high potential for ~ same
energy use, or for lower energy use for paper cup

15
Back to the decision context
• Use care in using simplistic methods to support
decisions
• “A decision made without taking uncertainty into
account is barely worth calling a decision.”
(Wilson, 1985)
• Would you really want to redesign a process
around the result of a deterministic LCA?

16
Decisions that matter
• Presumed focus of book and course
 We’re doing LCA for big decisions
 Hybrid cars? Green energy? Change-the-world policy
questions
• Thus analysis should be as good as possible
• “It depends” studies perhaps just not “complete”
studies
• LCA is a robust methodology that requires
careful choice of methods and leverage of
quantitative tools.

17
Biofuels and land use changes
• Growing crops for food or fuel requires land
• Converting land use creates significant GHG
emissions
 e.g., forests to farms
 Biomass decomposition, disturbance of soils and other
activities
• Promoting biofuels causes crop price increases
 Triggers increased production and conversion of land
 More GHG emissions from increased biofuels
• Results of studies of indirect land use change (ILUC)
emissions vary

18
Biofuels and ILUC
• No way to measure ILUC directly
• No objective way to tell which estimate is correct
• Researchers have conservatively considered
ILUC effects in biofuel systems, concluding that
biofuels can reduce GHG emissions
• However using higher estimates of ILUC effects
on GHG emissions would reverse conclusion

19
Biofuels and ILUC
• Conclusion ‘depends’ highly on which ILUC
values are considered
• Use of any one of the ILUC estimates creates
very narrow (misleading?) conclusions

• The uncertainty matters.

20
Measurement vs Accounting

• Measurement: observable quantity with an

ideal way to measure; limits to precision
 Uncertainty range might be described as +/- 1%
 Could be improved with better measurement tools
• LCI accounting: may lack primary data, raw
data leveraged to estimate flows
 Uncertainty ranges likely to be appreciable
 Results roll up many, many processes
• Do stakeholders expect measurement-quality
results?

21
General Overview of a Model

22
Data or input uncertainty
• Measurement uncertainty
• Parameter uncertainty
 Survey errors
 Incomplete and missing data
 Unit conversions
• Geospatial uncertainty
• Temporal uncertainty
 Old data
 Forecasting

23
Data or input uncertainty:
Measurement
• Imperfect measurement technologies (or
measurers)

24
 Data or input uncertainty: Parameter
• Survey data
 Coverage: sample or census
 Reporting errors
 Compiling errors
 Data suppressed due to confidentiality claims
• Incomplete and missing data
 Flows that are not included
 Sectors that are not studied
 Data unavailable for smaller businesses

25
 Data or input uncertainty: Geospatial
• Different processes used:
 Titanium dioxide is made by a sulfide process in
China, and a chloride process in the U.S.
• Different electricity grids
 Regional natural or built resources
 Coal, gas, hydro, nuclear, renewables?
 Age and efficiency
 Pollution controls
 National vs. micro

26
Foreign production?
• Are inputs manufactured in same region as
target product?
• Production in other regions may be similar OR
substantially different
 Different levels of environmental regulation
 Or environmental protection

27
Data or input uncertainty:
Temporal
• Old data
 Technology changes
 All US titanium dioxide sulfide processes have halted
 Three major US titanium dioxide chloride processes have
dramatically reduced dioxin/furan loads
 Computing: Desktops, laptops, tablets, handheld
 Pricing changes: competition, innovation, resource
constraints
• Forecasting

Distribution of USLCI basis years 28

 Temporal uncertainty
29

Example: Iron and Steel in US, Japan, China

Williams, Weber, Hawkins, 2009 29

Model or method uncertainty
• Process-based model
• Cutoff
• Software tool
• Database
• Allocation

• Others…
 Aggregation
 Input-output model
 Impact assessment method

30
Model or method uncertainty:
Process-based models
• How good is the model’s resolution (number of
processes in the model)?
• How well characterized are the input and output
flows?
• Presumed to be linear

31
Model or method uncertainty:
Cutoff error

• Cutoff or truncation error in

process LCA: not all
processes are included
 Cutoff or dummy flows in
USLCI, AgLCI, Ecoinvent
• Matthews et al (2008):
average industry captures
only 25% of upstream
cradle-to-gate using carbon
footprint methods

32
Model or method uncertainty:
Cutoff error
• Leaving out disposal/recycling stage
 Often done due to scenario uncertainty
• Landfill, incineration
 Depends on product/scope
 Not usually important for energy
 Incineration might dominate for dioxin
• Recycling
 Can be significant
 For buildings, embodied materials energy is 30,000
GJ, out of 360,000 GJ over 50 year life cycle
energy
(Junnila, Horvath, Guggemos, Journal Of
Infrastructure Systems, 11, March, 2006) 33
Model or method uncertainty:
Software and databases
• Internal differences in software setups (SimaPro
versus Gabi versus OpenLCA)
• Default database setups
 Default scope boundaries
 Default flows
• Read the metadata

34
Model or method uncertainty:
Allocation
• Allocation methods yield different results
 No apriori or consensus
• How important is allocation in an LCA?
• Example: how should we assess personal
transport impacts associated with purchasing a
good?
 100% of impacts?
 Allocate by number of passengers, by number of items
purchased, other?
 Assess using scenario-based sensitivity analysis

35
Uncertainty in results
• Results are uncertain due to accumulated effect
of all data, inputs, assumptions, and methods
• No unique type of result uncertainty
• Rolled up uncertainties in results are most
visible and tangible expressions of uncertainty
 to both practitioner and audience
 This is the uncertainty that matters the most

36
Methods to address uncertainty
• Qualitative
• Semi-quantitative
 Significance heuristics
 Pedigree matrix
• Quantitative
 Ranges
 Sensitivity analysis
 Probabilistic methods and simulation

37
Qualitative
• Discus sources of uncertainty
• Textual summary without quantification

• Reliability: “All data for key processes are based

on measurements (primary data), so uncertainty
is deemed to be relatively low for this category.”
• Completeness: “Various processes include only
effects of direct production, leading to some
cutoff uncertainty.”

See Fig 7-6 for more examples 38

 Semi-quantitative: Heuristics
• Rule of thumb, preset “rule” for comparisons
 Example: uncertainties of energy and carbon emissions
~20%; other LCI categories more uncertain
 Differences <20% … inconclusive
 No solid science behind “20%”, but a useful screening
tool
• True quantitative better!

Maintain enough significant

digits for comparison:

0.6 and 1.4, not 1 and 1

39
Quantitative methods
• Ranges
• Sensitivity analysis
• Probabilistic methods and simulation

40
Using ranges to understand uncertainty
• Use ranges for inputs or outputs
• Recommended: use multiple data sources, not
single values

41
Using ranges to understand uncertainty
• Ranges quantitatively represents effects of
different assumptions/boundaries in underlying
data
 helping to show when they matter
• Appropriate graphical range representation:
“uncertainty bars”
 Linear representation of ranges of results with upper
and lower bounds
 AKA “error bars” in Excel

42
See also Lecture 3

43
Process Flow Diagram-based
Example with Ranges
• See E-resource for Ch. 7, ‘Complex Models from
USLCI with Uncertainty’
• Template for comparing modules to develop LCI
uncertainty ranges
• Example compares:
 Electricity from bituminous coal
 Electricity from lignite coal
 Electricity from anthracite coal
 Includes upstream (scaled) processes for coal mining
(by type) and transport (by type)

44
Note scale is logarithmic

45
 Covered also in
Sensitivity analysis Lecture 3, and
discussed in
Chapter 2 and 4
• Quantitative method
• Assess effect on results from changing a single
input
 Change one variable at a time; hold all others constant

• Other methods, such as multi-way sensitivity

analysis and simulation, can show effects of
changing more than one variable at a time

46
ISO and Sensitivity Analysis
14040:3.31 sensitivity analysis – systematic
procedures for estimating the effects of the choices
made regarding methods and data on the outcome
of a study

47
Sensitivity analysis
14044:4.3.3.4 Refining the system boundary –
“Reflecting the iterative nature of LCA, decisions
regarding the data to be included shall be based on a
sensitivity analysis to determine their significance…”
“The sensitivity analysis may result in
 exclusion of life cycle stages or unit processes when lack of
significance can be shown by the sensitivity analysis,
 exclusion of inputs and outputs that lack significance to the
results of the study, or
 inclusion of new unit processes, inputs and outputs that are
shown to be significant in the sensitivity analysis.”

48
ISO and Sensitivity Analysis
14044:4.3.4.1: Allocation “…Whenever several
alternative allocation procedures seem applicable,
a sensitivity analysis shall be conducted to
illustrate the consequences of the departure from
the selected approach.”
• And in LCIA to show impact of different modeling
choices (normalization, weighting)
 To be covered later in the semester
• Required for “comparative assertions intended
to be disclosed to the public”

49
Sensitivity analysis
• “By hand” or using sensitivity analysis tools in
Matlab, @Risk, Excel (What-If analysis),
SimaPro (if you have a higher level license than
the course license), etc.
• Applies to inputs and assumptions (parameter
choices, allocation methods, etc.)
• Choose appropriate sensitivity ranges
 What makes sense?
 Not just automatic acceptance of +/-50% defaults

50
 Depicting sensitivity analysis
• Simple hi/lo chart
(Excel or Matlab)
• @Risk, TopRank or
other software to
generate tornado or
spider type graphs

51
Case study – Uncertainty in Process-
based flows

• Examine USLCI for all

US regional electricity
flows to show range of
potential CO2e
emissions per kWh

52
Leslie Abrahams, PhD
CEE/EPP Doctoral research

Spotlight on
LCA research

53
LNG Supply Chain

54
LNG Supply Chain

Currently, the export approval

process only requires an assessment
of localized environmental impacts
from the liquefaction facility

How would an increase in U.S. LNG

exports would impact global GHG
emissions from a life cycle perspective?

55
Electricity Generation
Exporting natural gas results
in 11% higher emissions than
domestic combustion

56
Displacing Traditional Energy
Sources for Electricity Generation

45%

13%

57
LNG Exports LCA
• Debate over fugitive emissions rate for
upstream natural gas life cycle stages. Handled
three ways:
 “Most likely” range commonly cited in literature
represented as triangular distribution (minimum 2%,
maximum 4%, most likely value 3%)
 Sensitivity analysis showing effects of fugitive
emissions rates across range encompassing most
values discussed in the literature (1−9%)
 Discussion of break-even fugitive emissions rate that
would change result of analysis

58
Sensitivity of life cycle emissions of LNG exports
for electricity generation to fugitive emissions
rates

59
LNG Exports LCA Conclusions with
uncertainty and sensitivity
• “…emissions from the liquefaction, shipping, and
regasification segments of the LNG life cycle are
fewer than 11% of the total life cycle emissions
of LNG exports for electricity generation based
on a 100-year GWP and 3% average fugitive
emission rate.”
• “This percentage would continue to decrease as
a result of increased methane leakage rates
and/or a 20-year GWP assumption.”

60
 LNG Exports LCA conclusions with
uncertainty and sensitivity considerations
• “Based on a sensitivity analysis of these results, the
key model parameters that can have a significant
impact on the LNG life cycle emissions are
 the end use efficiency,
 the GWP (both time horizon used, and value within a given
time horizon probability distribution), and
 the fugitive emissions rate.”
• “Other uncertain parameters, however, such as
liquefaction plant efficiency, tanker capacity, tanker
fuel, and shipping distance can vary widely without
materially affecting the overall life cycle emissions.”

61
 LNG Exports

Figure 2. Life cycle emissions for electricity generation from natural gas
exports (A) by life cycle stage for a 100-year GWP, and (B) for the complete
life cycle for both a 100-year and 20-year GWP.
62
Uncertainty … iterations
• Adding uncertainty allows for screening tools
that ‘screen’ based on:
 Magnitude of effect (likely most important effects)
 Uncertainty

 …Indicators of where additional data needed in

subsequent effort.

63
Environmental Life Cycle
Assessment
CEE 12-714 / EPP 19-714

Lecture 11a: Stochastic modeling in LCA (optional)

February 27, 2018

Uncertainty and probabilistic
methods
• We know we have uncertainty in…
 Inputs
 Outputs
 Model
• Goal: to understand and model appropriately
uncertainty so we can make better decisions
• Simulation can advance that objective

65
Uncertainty Basics
• Uncertainty is inherent in everything
 There are no right answers (just poor assumptions)
• We can analyze uncertainty using sensitivity
analysis

• We can model uncertainty by using ranges or

distributions for our input variables

66
Probability versus ranges
• Ranges imply at least a sort
of distribution
 But ranges quickly get
prohibitively complex
 5 variables only evaluated at
max/min points creates 32
different combinations to
evaluate if we abandon moving
only one value at a time
• Probabilistic modeling allows
us to move beyond the
limitations of ranges

67
Using probabilistic methods to
understand uncertainty
• Use data from various sources to generate
probability distributions for inputs
• Use spreadsheets or other techniques to track
the effect of these probabilities through the
model
• Generate results that also have probability
distributions

68
Using probabilistic methods to
understand uncertainty: INPUTS
• Not point estimates or ranges
• Distributions defined based on multiple data
sources and methods like curve fitting
• Actual data used to make distributions
 Fit, or assume a particular type of distribution given
data properties

69
2 Types of Probability Distributions
• Discrete distribution
 The uncertain quantity can only be a finite or countable
number of values
 Binary events (0 or 1), number of chocolate chips in a
cookie
 Binomial, Poisson
• Continuous distribution
 The uncertain quantity can take on any value in a
given range
 Temperature, age
 Normal, uniform, beta

70
Discrete and continuous distributions
Discrete Distributions Continuous Distributions

71
Choosing the right distributions

Normal Distribution
• Central tendency
common to many
physical and social
phenomena
• Beware of long tails!
• Not suitable for
variables that are
inherently positive or
strongly skewed
(weight, stock price)

72
Choosing the right distributions

Lognormal
Distribution
• Only positive values,
well defined
minimum limit
• Mean can be very
distant from mode
• Traditional
distribution for
reliability analyses

73
Choosing the right distributions

Uniform Distribution
• Reflective of truly
random, but
constrained quantities
• Good when
justification for
distribution
parameters is lacking

74
Choosing the right distributions

Triangle
Distribution
• Good medium
between normal and
uniform distributions
• Can be too restrictive
on values near the
max/min limits
• When in doubt… not
a bad choice.

75
Choosing the right distributions

Poisson
Distribution
• Discrete distribution
common to
queuing/queuing
problems
• cars arriving at a
traffic light
• telephone calls
arriving in a system
• photons arriving at a
telescope

76
STOP !
Do you really need a distribution?
• What doesn’t need a distribution?
 User criteria
 Discount rates, safety factors, etc.
 User decisions
 Which investment to select, etc.
 Anything that might be causally related to a variable
already in the model…

• Another individual’s criteria or decisions may be

correctly modeled as a randomized distribution

77
Justify distribution parameters
• Distribution parameters must be justified
 Can come from existing data
 From personal experience
 From similar known mechanisms in analogous
situations
• Pay attention to independence
 Is it valid to consider your inputs independent?

78
 Depicting Probability Distributions
• Probability Mass or Density Function
 Describe the density of probability of a random
variable taking on a value in the sample space
 PMF for discrete distributions (similar to a histogram)
 PDF for continuous distributions
X

PMF PDF
P(X = x)

Binomial Distribution (Discrete) Normal Distribution (Continuous) 79

 Depicting Probability Distributions
• Cumulative Distribution Function
 Probability that a random variable is less than or equal
to a given value in the sample space
 Used for both discrete and continuous distributions

CDF
P(X ≤ x)

CDF

Binomial Distribution (Discrete) Normal Distribution (Continuous)

80
Relationship between PMF/PDF and CDF
Binomial Distribution (Discrete)

PMF CDF
P(X = x)

P(X ≤ x)
Add

X Normal Distribution (Continuous) X

PDF

Integrate

CDF

81
Know how to read PDFs & CDFs
• Mean?
• Minimum, maximum?
• 5-95%?

CDF

82
When variables are distributions, how
can we use them?
• LCA models are based on mathematical
manipulation
 What are the rules when your variables are distributions?
• Some cases are easy
 Adding Normal distributions
 N(μ1,σ1) + N(μ2,σ2) = N(μ1+ μ2,√(σ12+σ22))
 Multiplying Lognormal distributions
 LN(a1,b1) x LN(a2,b2) = LN(a1+ a2,√(b12+b22))
• Very difficult for models of complex systems
 N(μ1,σ1) x LN(a1,b1) = ?
 Binomial(n,p) x [Exponential(λ) + Uniform(a,b)] = ?

83
Solution:
Stochastic (or Monte Carlo) modeling
• Allows for modeling of uncertain parameters
across their ranges of probabilities
• Output represents the resultant range of
possibilities…

12-706 84
How does the simulation work?
• A model is a set of calculations
• Variables are defined as distributions
• The model is run many times, each time taking
random samples from the variable distributions
• The resultant set of model results (outputs) is
evaluated as an output distribution

• Let’s see an example.

85
Intro Resilience Trips Efficiency Air Conclusions

Annual Waterways Shipments, 2012

Nationwide: 2.3 billion tons
Port of Pittsburgh: 35 million tons
Port of Huntington: 53 million tons
(USACE)
Source: Port of Pittsburgh Commission

86
Regional Inland Waterway
Dataset from
Corps of
Engineers

56,000 records of
each lockage in
2010

Derive time to
travel between
locks, delays
before locking,
time locking
Source: USACE
87
 Issue: Long delayed infrastructure
improvement causes trip delay
• Most Corps analyses focus on delays at one
lock
• Most trips traverse multiple locks, with
compounding delays
• Infrastructure improvements reduce trip time
• Challenge: how to measure a “trip”?
• Dataset yielded distributions of
– Time to travel between locks
– Delays before locking
– Time in lock

88
Charleroi Locks and Dam
90
91 of
25
 Modeling transit time

Coal to Fort Martin Powerplant

(EIA)
2.75 million tons in 2010 92
Shipping Time Components
• Stochastic model of time savings from
completion of the Lower Mon Project
Total travel time (hours)= Time underwayhours +
Probability of delay * Delayhours +
Lockage timehours + Reconfigurationhours
• 15-barge tow shipment from Powhatan No. 1
mine to Fort Martin
– 200 miles, 11 locks, 164,000 tons in 2010
– 17,200 to 22,100 tons per tow
Setting up a travel time model
Total travel time (hours)=
Time underwayhours
+ (Probability of delay * Delayhours )
+ Lockage timehours
+ Reconfigurationhours

+[ * ]+ +4
Pearson Binomial Lognormal Loglogistic
Monte Carlo (Stochastic) Simulation
Take K random samples of each distribution

Fit a distribution and

Uncertain Variables Random Random
… determine the relevant
(with parameters) Sample 1 Sample K parameters

4.1 … 3.2
Plot a histogram
1.2 … 0.3 and/or
a PDF

50 … 76

Calculated output
for each set of 1209 … 1638 Produce K output values
random samples
95
Model results

Stochastic dominance

Jumbo barge shows

measurable
improvement

Mixed tows show little

improvement
LCA of the Yellow Pages:
Does Pittsburgh Still Need the Big
Book?
Functional unit: one
volume of “Yellow
Pages,” including its
production,
distribution, and
disposal

System boundary: all

households in Allegheny
County; excludes
commercial and
institutional
establishments 97
Yellow Pages LCA Stages
• Phonebook production
• Freight of phonebooks from manufacturing plant
to a Pittsburgh warehouse
• Warehouse storage prior to distribution
• Distribution
• Production of fuel used in distribution
• End of life

98
Sample Distributions Used in
Yellow Pages LCA

99
Distribution fitting with @Risk

100
Presenting LCA results considering
uncertainty

101
Stochastic modeling tools
• Simple models – Excel
• Palisades DecisionTools Suite (@Risk)
 Excel add-on makes it easy (and Help Content
is super)
 Cluster computers
 Download from IT Services
 https://userguide.its.ce.cmu.edu/computing-
help/software/
• Other courses cover the underlying math
and more sophisticated application in much
more depth…
102
Using probabilistic methods to
understand uncertainty: RESULTS
• Result of simulation is a probability distribution
• Greater iterations of simulation, resulting output
distribution should converge to actual
distribution of results

103
Using probabilistic methods to
understand uncertainty: RESULTS
• Results as probability distributions (e.g., pdf/cdf)
let us leverage probabilities to provide decision
support beyond what is possible with ranges
 Range-based analyses only show where ranges
overlap
 Probability distributions can compare distributions to
determine the likelihood that performance of A is
superior to B

104
Summary
• LCA involves considerable uncertainty
• Can be reduced with careful analysis of
underlying data and production processes
• Relative desirability of design alternatives can
be assessed with greater confidence than
overall impact of a single alternative
• Unrealistic to expect that we could support a big
decision without accounting for uncertainty.

105
Questions?

106
Next
• HW3 due tomorrow, Feb 22nd
• Monday, project proposals (goal and scope) due by
midnight
• Monday, we start a 4-part series of lectures on input
output modeling, including an important (fun) class
exercise
 Chapter 8 thru Ex. 8-2.
 See additional Canvas readings
 Be on time!
• Tuesday, office hours will be an optional lecture on
stochastic modeling (more details than today)

107