1 of 48

Multi-hazard Risk Assessments of Hospital and Power Systems for Short-term Recovery

Luis Ceferino

Assistant Professor

Civil and Environmental Engineering Department

University of California, Berkeley

November 4, 2024

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

1

2 of 48

Infrastructure Systems must be prepared to Respond to Large Emergencies

Data Source:

EM-DAT, 2019

More than 100k injured after 2023 Turkey Earthquake (Mw 7.8)

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

2

3 of 48

Motivation

The vulnerability of urban systems is a critical problem in disaster risk management (DRM) worldwide

To create more robust and informed policies for DRM, communities need�better understanding of:

What factors govern the performance of infrastructure systems during short-term recovery?

How can we enhance such performance, especially in large systems?

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

3

4 of 48

Objectives

Showcase new methods to quantify large infrastructure systems’ performance under heavily uncertain single and compounding extreme events

Demonstrate how these methods can inform strategies to improve short-term recovery, e.g., for hospital and power systems

Interdisciplinary approach for better hazard assessments, decision-making aids, and government policies (McEntire, 2017; Mileti, 1999):

These projects lie at the intersection of structural and systems engineering, natural hazards modeling, and emergency medicine.

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

4

5 of 48

Earthquakes

Hurricanes

Heatwaves + cyberattacks

Hospitals and Power Systems in Short-term Recovery

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

5

6 of 48

Earthquakes

Hurricanes

Heatwaves + cyberattacks

Hospitals and Power Systems in Short-term Recovery

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

6

7 of 48

Hospitals drastically reduce their capacity

2016 Ecuador Earthquake (Mw 8.0)

Iskenderun Hospital

2023 Turkey Earthquake (Mw 7.8)

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

7

8 of 48

Objectives

Develop a method that captures

Reduced capacity of hospitals
High demand of healthcare
System behavior of hospitals

Leverage optimization techniques to design effective policies

Application: M 8.0 earthquake in Lima

10 M people in nighttime scenario
L: 160 km, W: 80km (Strasser et al., 2010)

Strategies for Emergency Response in Hospital Systems

Ceferino et al., Bull. Seismol. Soc. Am. 110, 2498–2518 (2020)

Ceferino et al., Bull. Seismol. Soc. Am. 111, 3356–3373 (2021)

Earthquakes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

8

9 of 48

Probabilistic formulation for Multi-scale Earthquake casualty modeling �

Ceferino et. al, ASCE-ASME J. Risk Uncertain. Eng. Syst. Part A Civ. Eng. 4, 04018023 (2018)

Earthquakes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

9

10 of 48

Demand:

Ground motions

Ground shaking model (Abrahamson et al. 2016)
Shaking spatial correlations and cross-correlations (Markhvida et al. 2018; Goda & Aktinson, 2009)

Structural Damage

1.5 M buildings (SARA-GEM)
Fragility functions (Villar et al. 2017)

Casualties

Nighttime scenario

Ceferino et al., Earthq. Spectra 34, 1739-1761 (2018)

4.7k (on average) earthquake patients will need operating rooms

Earthquakes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

10

11 of 48

Capacity:�

Ground Motions

Structural Damage

Data on 41 healthcare campuses with ~700 buildings (Santa Cruz et al. 2013)

Post-earthquake Functionality

Structural damage + Hospital Safety Index (WHO)
Focus on operating rooms and ambulances

48% (on average) of operating rooms will be available

Earthquakes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

11

12 of 48

Spatial mismatch between healthcare demand and capacity

Hospital centralization

Functional operating rooms: 59% located at the center

Injuries in the periphery

Injuries needing surgical procedures: only 13% located at the center

Earthquakes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

12

13 of 48

Hospitals behave as a system: Minimum Cost- Flow Model

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

13

14 of 48

Time-varying and heterogenous network model

Earthquakes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

14

15 of 48

Optimization of Performance

Constraints:

(3) Positive number of patients waiting in triage

(2) Operating room and ambulance constraints + Positive flow

Flow conservation:

Waiting Time

Ambulance trips

Linear Programming (LP) Problem

Earthquakes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

15

16 of 48

Emergency plans for Operating Rooms (ORs)

Low coordination

Baseline 1

Hospitals without ORs send patients to closest hospital with ORs

Baseline 2

Hospitals without ORs send patients to hospitals with most ORs

High coordination

Policy 1

Patient transfers across the whole system
Share ambulances

Policy 2

Deployment of 15 ORs in field hospitals (+8% capacity)

Earthquakes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

16

17 of 48

Average patient waiting times

Baseline 1: To closest hospital

Baseline 2: To largest hospital

Policy 1: To any hospital + sharing ambulances

Policy 2: Deploying field hospitals (+8% capacity)

Mean values

Baseline 1: 30 days

Baseline 2: 22 days

Policy 1: 10 days

Policy 2: 8 days

Ceferino et. al, Nature Communications 11, 4325 (2020)

Earthquakes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

17

18 of 48

Patient Treatment and Transfers

Low Coordination (LC)

To closest hospital (LC)

Only 28% of patients are treated in the centric districts (59% capacity)
Ambulance capacity is bottleneck

Strategic transfers + shared ambulances (HC)

56% of patients are treated in the centric districts (close match to capacity)
Ambulance capacity is not a bottleneck
The algorithm places field hospitals in the periphery

High Coordination (HC)

Earthquakes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

18

19 of 48

Preliminary Results of Acute Care Bed Functionality

M 7.2 Earthquake in Hayward Fault (HayWired Scenario)

Earthquake Scenario

Post-EQ Bed Functionality

Earthquakes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

19

I am currently investigating the hospital system in the Bay Area

The Northridge earthquake left several building in almost 10 acute care hospitals unsable triggering evacuations.

I have created a risk analysis model considering that structuctural and nonstructural vulnreabilities, reported by the California Department of Health Care and Information.

I would like to study (1) accessibility during short term recovery (as in my presentation) and also long-term recovery for predisaster level emergencies. We have seen that accessibility to emergency care worsens for long time in a case study with Dr. Alisjahbana

And (2) also I would to assess what is the best strategy for retrofitting hospitals’s structural and non stru tural components in the context of the deadline these hospitals have in 2030 and their financial stress.

20 of 48

Changes in travel time to closest functional hospital per zip code

Takes longer to arrive at hospital

Richmond (x5)

Earthquakes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

20

21 of 48

People will move from East to West to seek medical treatment, many through vulnerable roads and bridges

Ceferino et al. (2024)

Pre-earthquake conditions

Post-earthquake conditions

Earthquakes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

21

22 of 48

Conclusions

Key features of emergency response

High demand, reduced capacity, and system behavior in large urban centers

New formulation for:

Capturing these critical features
Optimization for designing emergency plans

Application to Lima subjected to a Mw 8.0 earthquake shows:

4,700 patients (mean estimates) will need ORs.
OR functionality will be reduced by 44% (mean estimates).
High-coordination policies pay off: reduction of waiting times by a factor of 3, potentially saving lives

Application to the Bay Area to M 7.2 earthquake ongoing

Earthquakes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

22

23 of 48

Earthquakes

Hurricanes

Heatwaves + cyberattacks

Hospitals and Power Systems in Short-term Recovery

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

23

24 of 48

Power Systems are extremely vulnerable to disasters

Hurricane Ian (2022)

2.6 M customers without power at peak electricity loss

Fort Myers Beach and Sanibel without power almost a month after the event

Systems is centralized and power lines vulnerable

Source: PowerOutage.us; Erin Davis/Axios’s visualizations; Eva Uzcategui/Bloomberg

Hurricanes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

24

1) Hurricane Maria caused the total loss of power in multiple major cities in Puerto Rico in 2017, and in California, (longest and largest power outage in modern history by NASA). In January, only half of PR had access to power.

2) the medium-size Napa earthquake left 70,000 residents completely without power in 2014.

3) Hurricane Sandy 2012, most of New York city and the north-east part of NJ without power for a week

90% of the power outages have their roots in the distribution system.

Battling the Extreme: A Study on the Power System Resilience

This paper presents a load restoration framework based on distribution automation technologies. By Zhaohong Bie, Yanling lin, gengfeng li, and furong li

Distributed generation, storage, demand response and energy efficiency as alternatives to grid capacity enhancement Rahmatallah Poudineh, Tooraj Jamasb n Durham University Business School, Mill Hill Lane, Durham DH1 3LB, UK

25 of 48

Distributed Energy Resources (DERs) can enhance the resilience of vulnerable power systems

DERs’ features for resilience:

Decentralization: energy avoids old and vulnerable power lines
~55% US generation will be from solar by 2050 (US EIA, 2022)

Objectives:

Quantify the resilience value of solar panels

Earthquake case also studied

Patel et al. Earthq. Spectra 37, 2638-2661 (2021)

Ceferino et al. (2023). ASCE Natural Hazards Review 24 (4)

Hurricanes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

25

26 of 48

Hurricane Katrina (2005)

Irradiance decay during hurricanes

Irradiance at 3 pm (UTC) on 08/29/2006

Irradiance Decay during Hurricanes

Hurricanes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

26

27 of 48

Data Analysis

Best track: Past hurricanes
NREL: Solar irradiance (~100GB)

Good predictors

Distance to the eye
Category

Irradiance Decay during Hurricanes

Hurricanes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

27

28 of 48

Proposed Model for Irradiance Decay during Hurricanes

Time-dependent

Space and time-dependent

Ceferino et al., Stoch. Environ. Res. Risk Assess. 36, 2681–2693 (2022)

Hurricanes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

28

29 of 48

Using the Proposed model with Synthetic Storms

Generation can decrease beyond 70% in large regions during a category-4 hurricane even if the solar infrastructure is undamaged

Large GHI decay

Hurricanes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

29

30 of 48

Solar Panels’ Structural Vulnerabilities

Unique dataset on panel vulnerability

Structural performance in the Caribbean after Hurricane Irma and Maria in 2017 and Dorian in 2019

46 rooftop installations and 14 ground-mounted installations

Common failures: clip and racking failures; rooftop attachment failures in rooftop panels

Rooftop Panels

Ground-mounted panels

Wind fields (Chavas et al., 2015)

Ceferino et al. Reliab. Eng. Syst. Saf. 229, 108896 (2023)

Hurricanes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

30

31 of 48

Bayesian Fragility functions for Solar Panels

Rooftop

Ground-mounted

Fort Myers, Florida (50 m/s)

Likelihood

Prior

Hurricanes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

31

32 of 48

Solar Generation during hurricanes in Miami, Florida

Generation during hurricane

Counterfactual

Clouds

Wind damage

Clouds

Wind damage

Hurricanes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

32

33 of 48

Capacity: Cumulative generation at day 4 after landfall relative to counterfactual

10-year return period

1000-year return period

100-year return period

Ceferino et al., ASCE Nat. Haz. Review 24, 04023029 (2023)

Hurricanes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

33

34 of 48

Outage Models for the Traditional Grid

Data (2,300 cities)

Ida (2021): 1.2 M
Isaias (2020): 3.8 M
Michael (2018): 1.6 M
Harvey (2017): 1.3 M

Risk to Existing Grids

Inputs

Probabilistic Models

(Power Outage Data for Historical Hurricanes)

Power Outage Modeling at city level

Hurricane Winds
Precipitation
Soil Moisture
System Variables
Population Density
Land Cover
Root Zone Depth
Treed Area
Elevation

Arora & Ceferino, Nat. Haz. Earth Sys. Sci. 23, 1665-1683 (2023)

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

34

35 of 48

Risk to Existing Grids

A new Quasi-binomial power outage model to address current limitations

Proposed quasi-binomial model

Negative-binomial

Random forest

Arora & Ceferino, ASCE-ASME J. Risk Uncertain. Eng. Syst. Part A Civ. Eng. 10 (2), 04024027 (2023)

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

35

36 of 48

Risk to Existing Grids

Test on Hurricane Ian (2022): 3.1 M outages

Proposed model

Random Forest

Negative binomial

Actual outages

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

36

37 of 48

Going back to the community scale: Demand vs. Capacity

Hurricane Isaias (2020)

~1.3 M without power in New Jersey (~800k from JCPL)
Winds of ~37 m/s

Outage Data

At the building level from JCPL
At the census tract level for storms (Michael, Harvey)

Arora & Ceferino, Nat. Haz. Earth Sys. Sci. 23, 1665-1683 (2023)

Hurricanes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

37

38 of 48

Could rooftop solar panels and storage have enhanced the electricity resilience during Hurricane Isaias (2020)?

Marlboro Township, New Jersey

~9k (98%) outages
90% recovery took ~4 days

What-if analysis:

Widespread adoption of rooftop panels and behind-the-meter batteries
Demands: Consumption of 100% and 50% of pre-disaster levels

Hurricanes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

38

39 of 48

Solar Resilience Scenario: Marlboro Township during Hurricane Isaias (2020)

Outage

Non-Outage

Hurricanes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

39

40 of 48

Conclusions

Power system is extremely vulnerable

DERs provide an opportunity to increase its resilience

Novel models and procedures that evaluate solar generation during storms

Irradiance decay due to hurricane cloud conditions
Structural panel failures due to high winds

Hurricane Application to shows:

Cloud conditions generate transient generation losses (>70% for category four events)
Ground-mounted and residential panels had structural reliability indexes below code.
Rooftop panels and behind-the-meter batteries could have significantly enhanced electricity access after Hurricane Isaias, especially if consumption is intentionally reduced

We are currently studying hurricanes in the future climates

Hurricanes

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

40

41 of 48

Earthquakes

Hurricanes

Heatwaves + cyberattacks

Hospitals and Power Systems in Short-term Recovery

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

41

42 of 48

Compounding natural events

Heavy uncertainties in event’s timing and location

Maria

(2017)

M 6.4 EQ

(2019-20)

Fiona

(2022)

Compounding natural event and cyberattack

Natural Hazard

Cyber-attack

Nat. Haz. 1

Nat. Haz. 2

Compounding actions

Nat. Haz. 1

Cyberattack

Compounding actions

Avraam et al., App. Energy 349, 121577 (2023)

Cyberattacks are deliberate and targeted (time and location)

Heatwaves/Cyberattacks 42

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

42

1) Hurricane Maria caused the total loss of power in multiple major cities in Puerto Rico in 2017, and in California, (longest and largest power outage in modern history by NASA). In January, only half of PR had access to power.

2) the medium-size Napa earthquake left 70,000 residents completely without power in 2014.

3) Hurricane Sandy 2012, most of New York city and the north-east part of NJ without power for a week

90% of the power outages have their roots in the distribution system.

Battling the Extreme: A Study on the Power System Resilience

This paper presents a load restoration framework based on distribution automation technologies. By Zhaohong Bie, Yanling lin, gengfeng li, and furong li

Distributed generation, storage, demand response and energy efficiency as alternatives to grid capacity enhancement Rahmatallah Poudineh, Tooraj Jamasb n Durham University Business School, Mill Hill Lane, Durham DH1 3LB, UK

43 of 48

Motivation

Power systems are heavily disrupted by heatwaves, hurricanes, wildfires, earthquakes
Power systems are also exposed to cyberattacks

Objectives

Create a framework to quantify critical infrastructure’s disruptions due to these compounding effects
Model NYISO (~20M customers) under:

Heatwaves
Cyberattacks

Compounding Natural Hazards and Cyberattacks on Power Systems

250k people without power

2015

2022

Heatwaves/Cyberattacks 43

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

43

f) The objective function of the cyberattacker in (6a) is a measure of the target disruption C˜(·), while (6b) enforces that all lower-level variables are solutions to the DC-OPF. Equations (6c)-(6e) capture that compromised generation and transmission capacity, and limits on voltage angle differences can not exceed the respective physical capacities and limits. However, a cyberattacker can target intelligent electronic devices of substations, e.g., transmission line protective relays [38], and generators, e.g., manipulating the automatic generation control signals [3], to constrain power flows and generation. Equation (6f) describes the capabilities of the cyberattacker in each hour h of each season

Since generation, transmission, and end-use components use Supervisory Control And Data Acquisition (SCADA) devices and Programmable Logic Controller (PLC) interfaces [2], the cyberattacker can modify SCADA and PLC firmware [40, 72] and hardware [73, 74] to execute unauthorized code. In our framework, the cyberattacker can inject spoof measurements and modify control algorithms and messages exhanged SCADA-SCADA, SCADA-PLC, PLC-PLC communications to constrain and manipulate regional generation, flow and angle difference values and limits

44 of 48

Regional Impact

NYISO

Demand vs. Capacity

Bilevel Optimization

Heatwave

Cyberattack

Disruption

Propagation

Baseline Electricity Demand + 9%

Framework for Modeling Compounding Natural Hazards and Cyberattacks

Avraam et al., App. Energy 349, 121577 (2023)

Heatwaves/Cyberattacks 44

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

44

First, we model the impact of an extreme weather event on energy infrastructures. In our case the extreme weather event is a heatwave, which increases regional demand in all zones of the New York Independent System Operator.��Then, we generate compound cyber-physical threat scenarios using bilevel optimization: on the upper level, the cyberattacker targets generation and transmission components to maximize unserved load. The response of the NYISO is modeled as an optimal power flow problem in the lower level. ��The compound cyber-physical threat limits electricity supply and the disruption propagates across all economic sectors. We assess the propagation of the disruption to all sectors using a Computable General Equilibrium, which models economic interactions between all industries, services, types of consumers and the government.��Across space + across time + across sectors …

45 of 48

Compounding Effects increase disruptions by a factor of four

Outages: 20k (cyberattack) versus 80k (cyberattack + heatwave)

Due to higher cyber vulnerability, generation is compromised before transmission capacity

Local effects

Most of the outages are in Long Island (13% unserved load) since both generation and transmission are almost at maximum capacity due to the heatwave

Heatwaves/Cyberattacks 45

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

45

We first calibrated a moderate case, where a cyber attack can compromise 20k customers (10% than in the case in Ukraine) due to cyber-vulnerabilities in generation and transmission.

We doung that in this case coumpding effects have a factor of four, increasing outages to 80 k when it is timed with a heatwave.

This happens mostly in Long Island, where transmission and generation were almost at capacity during the heatwave.

We assess the potential of compound cyber-physical threats on urban electricity infrastructures by simulating a cyberattack during a Sandy-like hurricane in NYC. The hurricane damaged the substation in Lower West Manhattan, leading to load curtailment in all of Lower Manhattan.�

Under normal operating conditions, a cyberattack that constrains power flow from New Jersey to Manhattan barely affects demand. However, when the same cyberattack happens during or immediately after the hurricane, it can compound unserved demand due to the hurricane in Manhattan by 60%.

46 of 48

Extreme outages: 5M

Many transmission lines and generation nodes are compromised

Local Effects

Attacks to transmission lines leave NYC isolated and exposed to larger outages

Compounding effects are more important for moderate cyber-vulnerabilities

Heatwaves/Cyberattacks 46

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

46

47 of 48

The deliberate “nature” of cyberattacks exposes us to new compounding threats (during or after natural extreme events)

We proposed a two-stage methodology for compound cyber-physical threat scenario generation

NYISO can experience outages that could affect from thousands of customers to millions

Active lines of inquiry

Natural hazards that damage power infrastructure (floods)
Benchmarking against natural compounding events
Expanding the models to respond to the cyberattack

Conclusions

Heatwaves/Cyberattacks 47

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

47

48 of 48

Thank you

Acknowledgments

Prateek Arora
Riccardo Negri
SoungEil Houng
Azin Ghaffary
Charalampos Avraam

2022 Hurricane Ian

2016 Ecuador Earthquake

2023 Turkey-Syria Earthquake

Introduction

Earthquakes

Hurricanes

Heatwaves/Cyberattacks

48