Modeling Exposure, Vulnerability, and Response

class: center, middle
.title[Modeling Exposure, Vulnerability, and Response]
<hr>
.left-column[.course[BEE 6940] .subtitle[Lecture 14]] .date[May 1, 2023]
 
---
name: section-header
layout: true
class: center, middle

<hr>

---
layout: false
name: toc
class: left

# Table of Contents
<hr>

1. [Review of Risk Determinants](#review)
2. [Challenges and Approaches: Modeling Exposure, Vulnerability, and Response](#challenges)
3. [Key Takeaways](#takeaways)
4. [Upcoming Schedule](#schedule)

???
This is an overview of the topics we'll cover in today's lecture. The italics around the last topic reflect that it's an "optional" topic that we may get to if time allows.

---
name: review
template: section-header

# Review of Risk

---
# Risk Decomposition
<hr>

.left-column[
Common framework:

**Risk** as a combination of
- *Hazard*
- Exposure
- Vulnerability
- Response
]

.right-column[
.center[
![Determinants of Risk](figures/simpson_risk.svg)
<br>
.cite[Source: [Simpson et al (2021)](https://doi.org/10.1016/j.oneear.2021.03.005)]
]]

---
# Example: Winter Storm Uri
<hr>

.left-column[
.center[
![:img Risk Components for Electricity Services, 95%](figures/winter_storm_uri_needs.png)
]]

.right-column[
Severity of impacts to electricity-dependent services came from:
- poor winter-proofing of homes;
- worries about provisioning.

.cite[Source: [Reed et al (2022)](https://doi.org./10.1029/2021EF002621)]
]

---
# Example: Winter Storm Uri
<hr>

.left-column[
.center[
![:img Risk Components for Electricity Supply, 95%](figures/winter_storm_uri_electricity_supply.png)
]]

.right-column[
Severity of impacts to electricity supply came from:
- fragility of (isolated) grid to low temperatures;
- increased electricity demand.

.cite[Source: [Reed et al (2022)](https://doi.org./10.1029/2021EF002621)]
]

---
class: split-70

# These Determinants Can Be Complex
<hr>

.column[
![:img Uri Risk By System, 90%](figures/Reed_Uri_Systems.png)
]
.column[
Contributors to failures span systems, sectors, and scales.

.cite[Source: [Reed et al (2022)](https://doi.org./10.1029/2021EF002621)]]

---
template: section-header
name: challenges

# Challenges and Appraoches: Modeling Exposure, Vulnerability, and Response

---
# Overarching Challenges
<hr>

Several categories of challenges for modeling exposure, vulnerability, and response:
- Downscaling uncertainties;
- Data collection/availability at appropriate scales;
- Dynamic changes due to endogenous dynamics.

---
# Downscaling Uncertainties
<hr>

Downscaling uncertainties can play a large role in short-to-medium term and highly local effects.

These are often neglected due to the complexity/computational storage needs for navigating large scale ensembles.

---
# Downscaling: Daily Temperature
<hr>

.center[
![:img Uncertainty Decomposition for Annual Max of Daily Max Temps, 65%](figures/downscaling_temp.png)
.cite[Source: [Lafferty & Sriver (submitted)](10.22541/essoar.168286894.44910061/v1)]
]

---
# Downscaling: Daily Precipitation
<hr>

.center[
![:img Uncertainty Decomposition for Max Daily Precip, 65%](figures/downscaling_precip.png)
.cite[Source: [Lafferty & Sriver (submitted)](10.22541/essoar.168286894.44910061/v1)]
]

---
# Downscaling: Extreme Degree Days
<hr>

.center[
![Uncertainty Decomposition for Extreme Degree Days](figures/downscaling_heat.png)
.cite[Source: [Lafferty & Sriver (submitted)](10.22541/essoar.168286894.44910061/v1)]
]

---
# Downscaling: Dry Days
<hr>

.center[
![Uncertainty Decomposition for Dry Days](figures/downscaling_dry.png)
.cite[Source: [Lafferty & Sriver (submitted)](10.22541/essoar.168286894.44910061/v1)]
]

---
# Data Availability at Appropriate Scales
<hr>

What kind of data do we need to assess exposure, vulnerability, and response?

- Database of population/assets that might be exposued;
- How much damage could ensue from different events;
- Adaptation measures which have already been implemented;
- Extent to which further adaptation measures or responses might be triggered.

These are all high-resolution pieces of information which can be difficult to gather!

---
# Data Availability at Appropriate Scales
<hr>

For flood risk:
- Structure inventory in floodplain;
- Reasonable depth-damage curves (including already-undertaken adaptations);
- Socioeconomic/settlement data;
- Information about how people would respond to new policies/programs or dynamic hazards.

---
# Uncertainty in Damage Estimates
<hr>

.left-column[
How much does the estimate of potential damages affect risk?
]

.right-column[
.center[
![:img Sensitivity of Expected Damages, 90%](figures/zarekarizi_sensitivity.png)
.cite[Source: [Zarekarizi et al (2020)](https://doi.org/10.1038/s41467-020-19188-9)]
]
]

---
# Structural Uncertainty and Damages
<hr>

In addition to building stock uncertainties, structural uncertainty exists in mapping larger-scale hazards to local-scale assumptions.

For example, certain modeling groups and foundations may claim high accuracy, but their model isn't publicly available.

This raises the issue of implementation uncertainty, as well as those emerging from any modeling choices and issues of resolution/scale.

---
# Structure Inventory
<hr>

The [FEMA HAZUS database](https://www.fema.gov/flood-maps/products-tools/hazus) is commonly used to estimate structures at risk, but exposure estimates are sensitive to building stock uncertainties:

.center[
![Hurricane Ida Damage Estimate Differences](figures/manville_hazus_diff.png)
.cite[Source: [Hurricane Idea in Manville, NJ](https://hurricane-ida-in-manville-nj-rutgers.hub.arcgis.com/)]
]

---
# Uncertainty in Damage Estimates
<hr>

.left-column[
How can we account for:
- Elevation;
- Moving appliances/electrical equipment out of the basement;
- Other floodproofing measures.
]

.right-column[
.center[
![:img Sensitivity of Expected Damages, 90%](figures/zarekarizi_sensitivity.png)
.cite[Source: [Zarekarizi et al (2020)](https://doi.org/10.1038/s41467-020-19188-9)]
]
]

---
# Socioeconomic Data
<hr>

Census/American Community Survey data provide a means to get at socioeconomic data.

- Census is more thorough, but only conducted every 10 years.
- ACS: Relatively small sample size

> The ACS sample size is simply insufficient to provide high-frequency data at high spatial resolution with low uncertainty levels...
>
> .cite[Source: [Spielman et al (2014)](https://doi.org/10.1016/j.apgeog.2013.11.002])

---
# Socioeconomic Data
<hr>

This is a fundamental challenge:

**How to get high-resolution socioeconomic data without violating privacy or introducing biases?**

---
# Endogenous Response Dynamics
<hr>

How to account for how people respond to dynamic hazards?

---
# Hazard Perception and Anchoring
<hr>

.left-column[
![:img Biases in Estimates of Hazard Fatalities, 100%](figures/hazard_estimates.svg)
]
.right-column[
![:img Anchoring and Hazard Fatality Estimates, 80%](figures/risk_anchoring.svg)
]
.center[.cite[Source: Morgan et al (2002), Risk Communication: A Mental Models Approach
<br>Adapted from Lichtenstein et al (1978)]]

---
# Risk Tolerance Can Be Subjective!
<hr>

.center[![:img Saturday Morning Breakfast Cereal: Risk and Statistics, 29%](https://www.smbc-comics.com/comics/20130517.gif)
<br>
.cite[Source: [SMBC 05/17/2013](https://www.smbc-comics.com/comic/2013-05-17)]]

---
class: left

# Review: Attitudes Towards Risk
<hr>

Two scenarios:
1. A certain payout of $X$?
2. An uncertain payout of $X-d$ or $X+d$, both equally likely?

Note that both of these "bets" have the same expected payout $X$. But there are a variety of responses!

---
class: left

# Risk Neutral
<hr>

.left-column[
Linear utility function:
- Expected utility matches expected bet outcome.
]

.right-column[
![Risk Averse Utility Function](figures/risk_neutral.svg)]

---
class: left

# Risk Averse
<hr>

.left-column[
Concave utility function:
- Greater utility impact from the downside of the bet than upside.
- *Risk Premium (RP)*: "penalty" that would be acceptable to avoid the uncertainty of the gamble.
]

.right-column[
![Risk Averse Utility Function](figures/risk_averse.svg)]

---
class: left

# Risk Seeking
<hr>

.left-column[
Convex utility function:
- Greater utility impact from the upside of the bet than downside.
- *Risk Premium (RP)*: "bonus" that would be needed to avoid taking the bet.
]

.right-column[
![Risk Averse Utility Function](figures/risk_seeking.svg)]

---
# Implications for Modeling Responses
<hr>

1. How is the system state translated into risk assessment?
2. Are decisions modeled directly or more abstractly?
3. What is the:
  1. utility function of the decision-makers?
  2. mathematical representation of response?

---
# Challenges for Modeling Responses
<hr>

- Available information
- Resolution/scale of agents/decision-makers
- Agent/decision-maker representation
- Decision rules/utility functions

---
# Challenges: Calibrating Decision Rules
<hr>

How can we calibrate decision rules?

- Data on outcomes or decision processes (surveys, experiments?)
- Statistical calibration can be challenging.
      - Often intractable likelihoods.
      - Hard to assess data biases.
- Often done through hand-tuning.
      - Overfitting?
      - Overconfidence?

---
class: split-50
# Decision Structural Uncertainty
<hr>

.center[![:img Impacts of Model Structure on Outcomes, 55%](figures/abm-structure.png)
.cite[Source: Yoon et al, Structural model choices regularly overshadow parametric uncertainty in agent-based simulations of household flood risk outcomes, Accepted]
]

---
# Institutional Constraints
<hr>

.center[
![:img Constraints on Adaptation, 50%](figures/vulnerability-constraints.png)
.cite[Source: [IPCC Assessment Report 6, Working Group II](https://www.ipcc.ch/report/ar6/wg2/figures/technical-summary/figure-ts-007)]
]

---
name: takeaways
template: section-header

# Key Takeaways

---
# Key Takeaways
<hr>

- Modeling exposure, vulnerability, and response is challenging for a variety of reasons.
- Data concerns/insights into decision processes can benefit from stakeholder engagement and input.
- Should consider structural uncertainty in all facets (including damage estimates);
- Think in terms of generative social science/exploratory modeling, not just for agent-based models.
- Institutional constraints add more complexity. and are often neglected or simplified.

---
name: schedule
template: section-header

# Upcoming Schedule

---
# Upcoming Schedule
<hr>

**Wednesday**: Work on HW4

**Monday**: Project Presentations.

**Next Friday**: Project Posters Due.