Post on 22-Jun-2020
transcript
Exploring green infrastructure programscenarios through stakeholder informed agent-based simulationsPresented by: FA Montalto
Funding:
Partners:
Bartrand, T; Geldi, J; Loomis, C; McAfee, C; Montalto, FA; Riggal, G; Travaline, K; and A Waldman
Disclaimer: Research findings do not necessarily reflect any official position of PWD
Acknowledgements
Managerial question:• What GI program strategies will result in meaningful
quantities of runoff being controlled on “x” % of an urban watershed within “y” yrs?
Confounding factors:• Different land typologies• Different types of GI • Complex community dynamics (+ and -)• Hodge podge of physical, social, regulatory, legal and fiscal
factors that vary time and in space
Generalized Goals:
Addresses watershed goalsAddresses
regulatory timeframe
Managerial question:• What GI program strategies will result in meaningful
quantities of runoff being controlled on “x” % of an urban watershed within “y” yrs?
Confounding factors:• Different land typologies• Different types of GI • Complex community dynamics (+ and -)• Hodge podge of physical, social, regulatory, legal and fiscal
factors that vary time and in space
Generalized Project Goals:
A family of computational models, typically custom built, that simulate the “bottom up” actions and interactions of autonomous “agents” in a network environment
Can be used to develop insights into how agent behavior and multi-domain interactions affect system performance
Methods:agent-based modeling
Netlogo: A free multi-agent programmable modeling environment developed at the Center for Connected Learning (CCL)at Northwestern University
Study Site:Point Breeze Neighborhood – Philadelphia, PA
Neighborhood Statistics:Area: ~ 175 hectares10,363 lots18.5% of lots are vacant75% of lots are residential82% of surface imperviousPop: 21,20035% below poverty line82% Af. Am. 10% Asian
Planimetric data• Tax lot boundaries• Building footprints• Impervious areas• Canopy cover• Lengths and widths of streets, sidewalks,
alleys
Planning data• Property status (owner occupied, rented,
vacant, tax delinquent)• Designated land uses/zones• Required setbacks• Street classification (one way, two way)• Parking status (one side, both sides)• Special features
• Previously implemented pilot projects• Bus/Metro Stops • Play street designation
Demographic data• Owner type (private or public)• Owner name & address• Tenure status (owner occupied or rented• Mean household income• Last sale date
Georeferenced windshield survey data
• Downspout status (to front or back of house)• Vehicular and pedestrian traffic counts• Community assets
Other• Street reconstruction/ repaving schedule• Philadelphia real estate market conditions
Derived• Connectedness status
• Value for management of routed runoff
Physical Model Components (GIS):Spatial features & their attributes
Interfacing Netlogo with GISCustomized dashboard for advanced spatial analysis
Example #1:Rain gardens possible with setback = 10ft
Green: private lots w/ rain gardens
Brown: Raingardens not feasible
Example #2: Rain gardens possible with setback = 10ft, green roofs with minimum building footprint (roof area) = 1000 sq. ft.
Green: private lots w/ rain gardens
Brown: Raingardens not feasible
Blue: meet green roof minimum area requirement
Example #3: Rain gardens possible with setback = 10ft, connectivity between rain gardens
Green: private lots w/ rain gardens
Orange: public lots that can capture 1” rain on their own area, but also yellow lots
Yellow: cannot capture 1” on site, but can be hydraulically connected to orange lots through alleys.
Blue: meet green roof minimum area Requirement
Brown: GI not feasible
Example #4: Rain gardens possible with setback = 10ft, connectivity between rain gardens, green roofs with minimum building footprint (roof area) = 1000 sq. ft., green streets (streets & sidewalks greened), social constraints = only connectivity to rain gardens on city owned properties
Next step in model developmentAdd “agents” and rules governing their behavior
What is an “agent”?• An autonomous entity that can interact with its
environment
What are “agent types”?• Groups of agents with common sets of goals, percepts,
and possible actions
How do individuals differ from one another?• Take different actions based on unique attributes and
experiences
Methods: Empirical methods for selecting agent types and behavioral rules in concert with stakeholders
Participant-observation 20-month period > 30 meetings & events Purposive Sampling
Interviews 13 Semi-structured Countless Unstructured Purposive Sampling
Community Street Fair attended by > 40 local
residents
Questionnaires 70 residents local preferences,
concerns, and adoption of GI
Policy Official Outreach 5 local agencies 1 local councilmember 1 state representative 2 state departments 1 regional agency
Local NGOs & informal associations
Tax lots, Streets, Alleys, Sidewalks, Census blocks, & GI facilities
PWD
Issue-driven organizations & other govt
agencies
Non-resident owners &
speculators
Local institutions (churches, schoolsChew, etc)
Global agent Local agent set Reactive set
Our “Agents”
Residents, Resident Owners, Non-resident owners
1. PWD establishes a block-scale GI Program Establish goals: Amount and timing of expenditures, spatial %
green goals, adaptive or static program Action #1: Prioritizing blocks for greening Action #2: Prioritizing incentives for each block
2. Property owners decide whether or not to adopt GI
3. Information is transmitted through the agent network
4. The sequence repeats
What happens in the model?
PWD selects streetblock based on implementation
strategy
PWD allocates funds for extant GI O&M
While budget remains, PWD funds/incentivizes GI
Private property
Streets
Schools & parks
Vacant lots
Quarter = Quarter + 1
30-year simulation complete?
GI adoption decisions made
PWD Selects GI implementation strategy, sets
solution parameters
Stopno yes
All streetblocks assessed?
Block and neighborhood attributes updated
no
yes
All blocks assessed
?
Budget exhausted
?
Among streetblocks not
yet at green target, select
streetblock with highest potential private GI green
fraction
PWD incentivizes
private GI for all eligible private
property owners
Private property owners decide
whether to adopt GI (rain garden, green
roof, flow-through planter)
Adopt GI?PWD installs GI,
updates block properties
Incentive offered to all private property owners?
Street paving?
PWD installs ETPs, porous
pavement Non-greened school , park?
PWD greens school or
parkBlock
organized ?
Covert vacant lots to
GI
Allow GI adoption on other blocks, no incentive offered
Advance to next quarter
yes
noyes
no
no
yes
yes
no
yes
no
yes
no
no
yes
Sample GI implementation strategy(prioritizing private property GI)
Collect information
Social network
Local conditions
Assess information
Values
Trust
Decide (stochastic)
Index
Constraints
Inform
Social network
Physical environment
Decision-making process:Property owners
Social network Local conditions
Block environment
Taxlot environment
PWD incentives
Global and local market conditions
Non-resident owner class (City,
Organization or Landlord
Neighbors (each with a separate social
network)
Renter (with a separate social
network)
Organizations with which non-resident owner is affiliated
Owner assesses
information, makes
decision
Owner experience
with GI
GI adoption decisions:tax lot owners
P(GI adoption) = fs * fe * fk
fs = binary spatial feasibility factor, based on lot configuration [0,1]
fe = baseline probability of adoption, modeled with a scaled logistic distribution function based on initial incentive ($108.75) as a fraction of monthly household income (MHH)
Sample adoption algorithm
108.75 Incentive includes: 1-time incentive to adopt: $100 monthly SW fee (perm.): $8.75
(based on Portland, Oregon downspout disconnect model)
Monthly Household Income derived for each lot based on
2000 US Census Tract data
50%
99%
gowner = $108.75 / (monthly household income)ownerLookup function
Scaled logistic distribution function used in model fe
P(GI adoption) = fs * fe * fk
fk = knowledge factor modeled w/ scaled logistic distribution function based on property owner▪ Exposure / experience with GI▪ Number of investment properties on block▪ Property owner membership in GI advocacy groups
Sample adoption algorithm
r = tenure status of property0 if owner occupied, 1 if rented
m = membership status of owner1 if member of NGO, 0 if not
nGI/np = fraction of block properties already retrofit w/ GI
nro/np = fraction of owners who reside on block
xlot = 1 – rlot + 2mowner + (nGI/np)block+(nro/np)block
3%
92%
Lookup function
Scaled logistic function used to represent fk
ResultsDynamic adoption of GI in Point Breeze
Ongoing work
1. Refining PWD behavioral rules and property owner adoption algorithms through more stakeholder interactions
2. Exploration of effectiveness of different GI policies, budgets, budget phasing, etc (current model exhausts budget because water utility assumed to pay for O&M)
Preliminary Conclusions
1. ABM▪ Useful tool for water utilities struggling to predict the spatiotemporal
extent of GI that may emerge in complex urban watersheds
2. Potential uses: ▪ Advanced spatial analysis of intricate GI opportunities▪ Consideration of dynamic interactions between stakeholders▪ Stimulating productive dialogue…emphasizing process over
outcome…