Habitat International xxx (2017) 1e15
Contents lists available at ScienceDirect
Habitat International journal homepage: www.elsevier.com/locate/habitatint
Urban geosimulations with the Logic Scoring of Preference method for agent-based decision-making Suzana Dragi cevi c*, Kristofer Hatch Spatial Analysis and Modeling Research Laboratory, Department of Geography, Simon Fraser University, 8888 University Drive, Burnaby, BC, V5A 1S6, Canada
a r t i c l e i n f o
a b s t r a c t
Article history: Received 17 September 2016 Received in revised form 28 June 2017 Accepted 18 September 2017 Available online xxx
The Logic Scoring of Preference (LSP) method is a generalized multicriteria evaluation (MCE) decision making approach with origins in soft computing. The method can model a wide range of aggregators to suit various evaluation objectives that are close to human reasoning. The LSP method can aggregate an unlimited number of inputs without loss of significance. The main objective of this study is to develop and implement an integrated approach using the LSP method to represent the decision-making process of actors influencing urban residential development represented within an agent-based model (ABM). Geospatial data for the Clayton-Cloverdale neighborhood of the City of Surrey, Canada, was used to incorporate the LSP agent-based model to simulate land-use change at the cadastral level. The geosimulations incorporated the decision-making process and interactions of agents as residents, developers, and city planners known as the main stakeholders with separate and sometimes conflicting priorities. The simulation results indicate a higher number of residents tend to choose mid-rise to highrise buildings over single residential dwellings for a longer period of the time. This can be attributed to the lack affordability and developable land for housing in the future. The LSP method captured different agent decision-making reasoning that is closer to actual human logic which has resulted in the modeling outcomes of urban residential land-use to be in accordance to long term city plans. © 2017 Elsevier Ltd. All rights reserved.
Keywords: Logic Scoring of Preference (LSP) method Agent-based model Human decision-making process Geosimulations Urban land use change Soft computing
1. Introduction One of the dominant paradigms in land-use change research has been to describe cities as complex systems (Batty, 2005). Cities are self-organizing entities with individual interconnected components including transportation, land use, demographics, and topography. These components interact and exhibit behaviors and properties associated with complex systems. The local interactions are often viewed as bottom-up that can produce complex patterns at larger spatial scales (Albeverio, Andrey, Giordano, & Vancheri, 2010). For the last two decades and more, geographic automata (GA) emerged as the predominant modeling paradigm used to simulate urban dynamics and residential land-use change (Torrens & Benenson, 2007). The GA contain cellular automata and agentbased modeling approaches. The CA component consists of a grid of cells with a finite number of states that change due to local
* Corresponding author. E-mail addresses:
[email protected] (S. Dragi cevi c),
[email protected] (K. Hatch).
neighborhood interactions based on well-defined transition rules. Despite their simple framework, CA models of urban systems can generate complex behaviors observed in urban systems (White, Engelen, & Uljee, 2015, p. 344). However, the framework of CA models imposes some limitations. CA models cannot properly represent complex interactions between individuals as transitions occur on static cell states. The more advanced generation of geographic automata called agent-based models (ABMs) can address the limitations of CA models by incorporating autonomous, heterogeneous agents into the developed geospatial model (Heppenstall, Malleson, & Crooks, 2016). Agents interact through cooperation and competition among each another and within their spatial environment. ABMs operate on a local scale and the decisions and interactions amongst agents create patterns at local and coarser spatial scales (Parker, Manson, Janssen, Hoffmann, & Deadman, 2003; Fontaine & Rounsevell, 2009). Another limitation of CA models is their use of the commonly available raster data model to represent geographic features. At coarse spatial scales, features such as houses, parks, commercial facilities, and schools are rarely adequately represented through
https://doi.org/10.1016/j.habitatint.2017.09.006 0197-3975/© 2017 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Dragicevic, S., & Hatch, K., Urban geosimulations with the Logic Scoring of Preference method for agent-based decision-making, Habitat International (2017), https://doi.org/10.1016/j.habitatint.2017.09.006
2
S. Dragicevic, K. Hatch / Habitat International xxx (2017) 1e15
assignment into cells on a square lattice. Instead, their shapes and sizes are often irregular requiring the use of vector-based GIS that can represent spatial features such as cadastral lots as irregular spatial tessellations (Stevens & Dragicevic, 2007). ABM approaches can use both raster and vector GIS datasets although some are not based on geospatial data. In order to simulate urban growth within ABMs, agents are used to represent residents, city planners, developers, environmentalists, farmers and any other parties or stakeholders that can influence decisions or affect land-use change (Jjumba and Dragi cevi c, 2011; Ligtenberg, Bregt, & Van Lammeren, 2001; Evans & Kelley, 2004; Valbuena, Verburg, Bregt, & Ligtenberg, 2010; Irwin & Bockstael, 2002; Lim, Deadman, Moran, Brondizio, & McCracken, 2002). The decision-making algorithms for these agents are based on sets of criteria relevant to each group of agents. For example, urban residents decide to occupy homes based on criteria such as the home value, amenities, and location. The ability of agents to make choices based on some decision rules brings the geosimulation models closer to human reasoning and decision making (Bommel, Becu, Le Page, & Bousquet, 2016; Groeneveld et al., 2017; Rounsevell, Robinson, & Murray-Rust, 2012). A multi-criteria evaluation (MCE) approach can be used to combine different criteria and make the agent decisions easier or optimal (Li & Liu, 2007). Fuzzy reasoning can also be used to evaluate and standardize the variables that agents consider in their decision making (Graniero & Robinson, 2006). Moreover, Bayesian Networks (Kocabas & Dragicevic, 2013) and reinforcement learning (Bone, Dragicevic, & White, 2011) have been used to model different stakeholder decision-making in the case of land use change models. Ghavami, Taleaia, and Arentze (2016) proposed the multi-actor heuristic mechanism to model multi agent exchange of opinion and negotiations that lead to decisions about land-use change while Kelley and Evans (2011) explored the influences of simulated land-owner decisions on the landscape heterogeneity. The MCE methods can assist in the creation of decision-making algorithms of agent reasoning within geographic automata and particularly ABMs. For example, the model developed by Li and Liu (2007) and Arsanjani, Helbich, and Vaz (2013) used MCE to define a utility function for the decision-making algorithms of urban resident agents. This utility function allowed residential agents to assess the value of a potential residential site with regards to price, surrounding environment, accessibility, provision of general facilities, and educational benefits (Huang, Parker, Filatova, & Sun, 2014). Typical MCE-based algorithms for agent decision-making rely on weighted linear combination (WLC) based methods for MCE analysis (Malczewski, 2006). Typically WLC takes a set of input factors and combines them together using weights of preference as each factor is given some preference between 0 and 1 inclusive, and the sum of the factor weights must add to 1. However, WLC has significant limitations when used in GIS-based MCE and as a basis for agent decision-making, especially pertaining to the capability to observe and simulate the logic of human decision-making. Several properties of MCE methods have been identified to characterise the MCE methods in the way how the factors satisfy , & Dragicevic, 2009). Some of given objectives (Dujmovic, De Tre these properties include the ability to: combine any number of inputs, combine objective and subjective inputs, combine absolute and relative criteria, develop flexible attributes, and retain and accurately represent the human decision-making logic (Dujmovic , 2011). MCE using the WLC method is not able to satisfy & De Tre all these properties. In particular, WLC cannot represent the full range of human decision-making logic, nor incorporate a large number of factors in the MCE analysis without loss of significance on any one individual factor. The Logic Scoring of Preference (LSP) method is capable of satisfying all fundamental MCE properties
, & Weghe, 2010). The use of variable ANDness (Dujmovi c, Tre (known as simultaneity in the LSP approach) and ORness (known as replaceability) among inputs in the LSP method allows LSP to better expresses human reasoning compared to the existing GIS-based &Van de Weghe, 2008). LSP also allows MCEs (Dujmovi c, De Tre for the inclusion of an infinite number of inputs in MCE analysis, without loss of significance on any individual input. It is important to explore the use of LSP as an alternative method to WLC (and other existing MCEs) and to explore the option to model the decision-making processes of agents and their reasoning logic. The use of LSP can improve the representation of a wider range of human decision-making logic and provide realistic algorithms for determining agents' decision-making in geosimulation models. The main objective of this study is to implement the Logic Scoring of Preference (LSP) as a method to represent the human decision-making process of agents in an ABM of land-use change. The proposed LSP-ABM method simulates residential land-use change at the cadastral level. Various stakeholder types including residents, developers, and city planners are integrated as agents in the geosimulation model. The LSP-ABM was implemented on geospatial datasets at the cadastral level for the Clayton-Cloverdale neighborhood of the City of Surrey, Canada and with two different scenarios of urban growth. 2. Theoretical background 2.1. GeoSpatial agent-based models Parallels between complex systems and city dynamics necessitate the use of computational complex system models such as ABMs to represent and analyse urban growth patterns. Using agent-based models to simulate real urban systems with geospatial datasets is not yet fully resolved, but has gained a significant amount of attention in the past decade (Matthews, Gilbert, Roach, Polhill, & Gotts, 2007). Agent-based models were used for the simulation of micro-scale city dynamics (Benenson, 1998). Models such as UrbanSim (Waddell et al., 2003), and PUMA (Ettema, de Jong, Timmermans, & Bakema, 2007) were some of the first agent-based simulations developed to support planning and analysis of urban development. The recent increase in the use of ABMs for modeling urban land-use change is largely due to a standardization of the modeling process through the ODD (overview, design concepts, and detailed) protocol (Grimm et al., 2010), as well as due to the benefits of exploring policies related to urban planning (Ligmann-Zielinska & Jankowski, 2007). Most ABMs consist of two components: the static - representing the environment within which agents act, and; the agent layer representing autonomous, decision-making entities known as agents (Benenson, 1998; Bonabeau, 2002) that interact and generate changes within the model. Various types of agents comprise the second layer. For example, in an ABM of land-use change, agents can include: resident agents: those that move or relocate within the static environment, government/planning agents: those that zone or establish land suitable for development, and developer agents: those that build houses and subdivisions for the resident agents. These agents communicate, cooperate, and compete with one another to change the environment. It is difficult to define an agent, but according to Crooks, Castle, and Batty (2008) their main characteristics are: autonomy - agents can exchange information among one another and make independent decisions; heterogeneity and activity - agents can apply independent control in a situation. Other features are: pro-activity; agents are goal directed; reactive and perceptive; characterized by bounded rationality - meaning they can be restricted to only partial access to information; and they are interactive, mobile, and adaptive.
Please cite this article in press as: Dragicevic, S., & Hatch, K., Urban geosimulations with the Logic Scoring of Preference method for agent-based decision-making, Habitat International (2017), https://doi.org/10.1016/j.habitatint.2017.09.006
S. Dragicevic, K. Hatch / Habitat International xxx (2017) 1e15
The ABM framework focuses on agent behaviors and dynamics. Agents are provided with a set of rules that define their interactions both with their surrounding environment and among each another. Similar to CA, ABMs are a bottom-up approach capable of representing how systems evolve over time as many individuals interact. They are often based on hypothetical datasets to provide simulations (Crooks, et al., 2008; Ligtenberg et al., 2001; Shan & Zhu, 2007). Integrating real geospatial data increases the utility of models for decision-making purposes, and allows for better representation of agent's reasoning (Cabrera, Deadman, Brondízio, & Pinedo-Vasquez, 2010; Evans & Kelley, 2004; Kocabas & Dragicevic, 2013). Moreover, an agent based model capable of partitioning available land for urban development, and generating subdivisions with the use of geospatial datasets on a cadastral scale was developed by Jjumba and Dragi cevi c (2011) but the agent reasoning was limited. As a result, incorporating human decisions within agent-based modeling can improve the capability of the buil, & Hardy, agent for decision-making (An, 2012; Bousquet, Tre 2005; Villamor, van Noordwijk, Troitzsch, & Vlek, 2012). Existing studies have incorporated simple GIS-based MCEs, based on weighted linear combination (WLC) as methods for evaluating the agent decision-making processes within land use modeling (Li & Liu, 2007; Ligtenberg et al., 2001). However, methods about the agents' ability for decision-making should be expanded. The ability of the LSP method to represent a wide range of human decisionmaking and if coupled with agent-based model can represent one step further towards the incorporation of reasoning closer to human logic. 2.2. The Logic Scoring of Preference method The LSP method was developed as an approach to combine criteria with the aim to retain the logic of human decision-making. Human decision-making is represented through the inclusion of a continuous scale of simultaneity and replaceability used when combining criteria, features not available in other common GISbased MCE. The LSP method has been used for applications in computer science such as: windowed environment software evaluation (Dujmovic & Bayucan, 1997), evaluation of Java IDEs (Dujmovic & Nagashima, 2006), and comparison of search engines (Dujmovic & Bai, 2006). The LSP has been applied to spatial applications mostly using hypothetical datasets. In particular, Dujmovi c et al. (2008) proposed the concept of LSP aggregated geographic suitability maps, or s-maps. Suitability maps assign a degree of suitability to a set of spatial locations on a continuous surface for a specific purpose such as: suitability for industrial development, agriculture, housing, education, or recreation. Recently geospatial datasets have been used to apply the LSP method for suitability analysis for urban (Dragicevic, Dujmovic, & Minardi, 2018; Hatch, Dragicevic, & Dujmovi c, 2014) and agricultural (Montgomery, Dragicevic, Dujmovic, & Schmidt, 2016) land uses as well as assessing groundwater vulnerability (Rebolledo, Gil, Flotas, & Sanchez, 2016). In order to implement the LSP method, the following sequential steps are needed: 1. Establishment of set of input criteria: Criteria are objectivespecific and must be standardized and evaluated as mandatory or optional. 2. Development of the LSP attribute tree: The attribute tree provides an organized structure for the criteria established, defining the step-wise combination of criteria until one overall output representing the combination of all criteria is attained. 3. Development of the LSP aggregation structure: Whenever two or more criteria are combined in the LSP attribute tree, an
3
LSP aggregator must be applied, defining a degree of simultaneity or replaceability among the criteria. The set of LSP aggregators used form the LSP aggregation structure. 4. Calculation of overall LSP suitability output: The output is a score representing suitability for a particular spatial location with respect to a particular objective.
2.2.1. Input criteria Inputs are a set of factors that need to be combined to satisfy some objective or decision. Inputs must be chosen and categorized based on similarity where similar inputs are combined first in the LSP aggregation structure. In order to combine inputs they must first be standardized: each input must be transformed from their existing measurement units onto a standardized scale representing suitability where suitability is reflective of the objective under consideration. Inputs must be expressed as either mandatory (denoted by a ‘þ’ sign) or optional (denoted by a ‘-’ sign). If a mandatory input is completely unsuitable, having a suitability score of zero, then the overall LSP suitability score (the output of the entire LSP method) will also have a value of zero. Optional inputs do not have this requirement, however there can be a penalty or reward based on their input suitability values (Dujmovi c, 1979). After the input criteria have been chosen, standardized, and categorized, the LSP attribute tree can be developed. 2.2.2. LSP attribute tree The LSP attribute tree organizes the decision problem and contains all relevant attributes and parameters. It takes all input criteria, and determines the order in which they will be combined together, up until and including the point at which all input criteria have been combined together. Fig. 1 shows an example of a simple attribute tree with four inputs. In this example, the input criteria are grouped into either category A or category B. The assumption is that inputs A1 and A2 are more similar to each other than the inputs B1 or B2, and vice versa. Nodes A and B represent the combinations of inputs A1 and A2, and B1 and B2 respectively while node AB represents the combination of all four inputs. In LSP aggregation structures, each node (A, B, and AB) has their own associated LSP aggregator. 2.2.3. LSP aggregation structure The attribute tree allows for the design of an aggregation structure comprising a series of LSP aggregators describing the parameterization and step-wise combination of inputs based on logical requirements and weighting parameters (Dujmovi c & Scheer, 2010). The LSP aggregators express mandatory, (þ), and optional, (), requirements associated with the input criteria. Each requirement is represented in a spectrum of conditions ranging from full disjunction (D) to full conjunction (C) (Table 1). Each LSP aggregator used reflects the degree of simultaneity,
Fig. 1. Example of a simple attribute tree for combining four inputs A1, A2, B1, and B2.
Please cite this article in press as: Dragicevic, S., & Hatch, K., Urban geosimulations with the Logic Scoring of Preference method for agent-based decision-making, Habitat International (2017), https://doi.org/10.1016/j.habitatint.2017.09.006
S. Dragicevic, K. Hatch / Habitat International xxx (2017) 1e15
4
output value.
Table 1 LSP aggregators and associated exponent r. Simultaneity Symbol
C
Cþþ
Cþ
r
∞
9.06
3.51
Cþ
CA
Cþ
C
C
A
1.655
0.72
0.148
0.261
0.619
1.0
Replaceability Symbol
D
Dþþ
Dþ
Dþ
DA
Dþ
D
D
A
r
∞
20.63
9.521
5.802
3.929
2.792
2.018
1.449
1.0
neutrality, or replaceability desired to be expressed between the inputs considered. The further along the spectrum from neutral (A) to full conjunction (C) (Table 1) the aggregator is located, the stronger and more restrictive is the degree of simultaneity. The further in the other direction, from neutral (A) to full disjunction (D), the stronger is the replaceability among inputs. Neutral (A) expresses neither simultaneity nor replaceability. For example, in Fig. 1, if the satisfaction of input A1 negated the need for input A2 to be satisfied, and vice versa, then an LSP aggregator with a high degree of replaceability (such as DA, Dþ-, etc.) is assigned to node A. If inputs B1 and B2 both need to be satisfied to achieve a high satisfaction when combined, then an LSP aggregator with a high degree of simultaneity is applied to node B, such as CA or Cþ-. LSP aggregators can be grouped into one of seven aggregator types. These include: Full Conjunction (LSP aggregator C in Table 1), Hard Partial Conjunction (using aggregators such as Cþþ, Cþ, Cþ-), Soft Partial Conjunction (C-, Ce), Neutrality (A), Soft Partial Disjunction (De, D-, D-þ, DA), Hard Partial Disjunction (DA, Dþ-, Dþ, Dþþ) and Full Disjunction (D). The choice of LSP aggregator used is determined by the desired level of simultaneity or replaceability between inputs that the decision maker wants to express. A Hard Partial Conjunction (HPC) operator is often used to express the combination of mandatory inputs, whereas a Soft Partial Conjunction operator is less restrictive, and is appropriate for the combination of optional inputs. The analogue is true for Hard Partial Disjunction and Soft Partial Disjunction operators. LSP aggregators combine inputs of the same type (mandatory with mandatory, or optional with optional) using a generalized conjunction disjunction function outlined in Dujmovic et al. (2009). Given input parameters X1, …,Xn, the generalized conjunction disjunction (GCD) is computed by the weighted power mean: GCD(X1, …, Xn) ¼ [W1Xr1 þ … þ WnXrn]1/r
(1)
where GCD(X1, …,Xn) is the suitability for input parameters X1, …,Xn; W1, …,Wn express the relative importance of usefulness and inexpensiveness of inputs X1, …,Xn, and r expresses the degree of simultaneity and replaceability (Table 1) among the inputs X1, …,Xn. When combining inputs of different types (mandatory inputs with optional inputs), the conjunction partial absorption (CPA) function is used (Dujmovi c, 1979). Given a mandatory input (or set of inputs) X, and an optional input (or set of inputs) Y, there are two variants for CPA:
n o1=r2 r =r CPAðX; YÞ ¼ ð1 aÞbxr1 þ ð1 bÞyr1 2 1 þ ax
(2)
where either r1 < 1, r2 1, or r1 1, r2 < 1, and a ¼ W1, b ¼ W2 for either r1 < 1, r2 1 (CD-variant), a ¼ W2, b ¼ W1, for r1 1, r2 < 1, (DC-variant). The CPA function uses a penalty/reward scheme (Dujmovi c, 1979). Lower suitability values for optional inputs apply penalty to the CPA output value, and higher values apply reward to the CPA
2.2.4. LSP suitability output Once all inputs have been combined in the LSP attribute tree, and LSP aggregators (as well as their weights of preference) have been assigned, the overall suitability output is obtained in form of suitability score. Implementing LSP within GIS framework allows generation of results in relation to the inputs, logic aggregators, and weights of the chosen relative importance. Unlike traditional GISbased multicriteria decision-making methods, the LSP allows for the step-wise logic aggregation through the LSP structure which also facilitates flexibility based on continuous logic that is represented in terms of simultaneity and replaceability (Dujmovi c et al., 2010). LSP also allows for the inclusion of large numbers of inputs in its structure, without loss of significance due to the way in which the logic is expressed and closer to natural and gradual human reasoning. In contrary, traditional GIS-MCE methods cannot accommodate more than about ten to twelve factors, and rely on the additive scoring related to standardization of factors, while LSP method is based on the non-linear standardization (Dujmovic et al., 2009; Montgomery & Dragicevic, 2016). The common additive scoring reduces the possible spectrum of decision outcomes, and as such the LSP method produces a wider spectrum of suitability scores. 3. Methodology: LSP agent-based model The proposed LSP agent-based model consists of various agents that represent stakeholders and actors responsible for urban landuse change. The conceptual flowchart of the model structure is presented in Fig. 2. The model input are geospatial datasets stored in a GIS database and accordingly the model output is spatial, producing simulation output that are the generated maps and graphs. Various GIS operators were used for the LSP structures and combined with programs to handle agent interactions. In this study, three groups of agents making decisions were given consideration and they were the city planner agents, developer agents, and resident agents that are further composed of an additional three types of agents. These three main agent groups interact and cooperate amongst each another to shape urban land-use change across the study area over time. The city planner agent act to make sure policies and planning guidelines are followed. They also decide on land subdivisions into lots to make them available for the developer and resident agents. The developer agents choose new parcels of land to purchase and develop based on the expected profit that can be obtained from developing housing on specific parcels, as well as the demand from resident agents for parcels in certain locations, and of certain densities (i.e. high density, medium density, and low density). The resident agents then examine the newly developed and existing vacant lots in the study site and choose to occupy lots they deem most suitable. The LSP method is used to represent the reasoning behind the individual resident agents, and used to assign suitability scores to individual lots in the study site. Three sub-groups of resident agents considered in the model are: family, single, and senior agents each with their own set of preferences and decision-making logic, and each of whom assign their own individual suitability scores to vacant lots in the study site. There is a need to make the ABM descriptions more understandable and complete by using structured protocols such as the common ODD (Overview, Design concepts, Details) (Grimm et al., 2010) primarily designed for ecological applications. However, the ODD protocol has been found to be challenging when the structured representation of agents' decision making is needed to be described (Müller et al., 2013). Therefore, some aspects of the
Please cite this article in press as: Dragicevic, S., & Hatch, K., Urban geosimulations with the Logic Scoring of Preference method for agent-based decision-making, Habitat International (2017), https://doi.org/10.1016/j.habitatint.2017.09.006
S. Dragicevic, K. Hatch / Habitat International xxx (2017) 1e15
5
Fig. 2. Flowchart of the conceptual LSP-agent-based model (LSP-ABM) with the planning, developer, and three types of resident agents.
ODDþD (Overview, Design concepts, Details, Decision making) protocol designed primarily for ABM in social-ecological applications were used to summarize the details of the proposed LSP-ABM components and have been presented in Table 2. This is followed by the subsections describing in detail the study site and datasets used for the LSP-ABM implementation. The proposed model was a loosecoupled design with both raster and vector GIS software tools.
(City of Surrey, 2013) and demographic data from the Canada Census for years 2006 and 2011 (Statistics Canada, 2013). The landuse data is at a cadastral scale containing data on the lot value, the vacancy status, the development type, the year constructed, among other information.
3.1. Study site and datasets
3.2.1. City planner agent The city planner agent ensures a set of established city development guidelines and city bylaws for the study area are followed and they interact directly with developer agents. The city planners react by either permitting or denying the proposed urban development submitted by developer agents. The developer agents act by proposing to develop tracts of land within the study area. These tracts vary in size, but are usually on the scale of a city-block (or one side of the street of a city-block), encompassing between five and twenty single family homes, or a single low rise apartment. The city planner agent can restrict development of lots of any type of density (high, medium, or low) at a particular location requiring from the developer agent to choose a different tract or density of tract to develop. For example, if a location is zoned as industrial, commercial, forest, park or agriculture they may deny the proposal for development. The city planner agent can restrict lots for particular density development that are close to major freeways or intersections. However, if a developer agent proposal is accepted, the tract of land is then subdivided into individual cadastral lots. The
The Clayton-Cloverdale neighborhood of City of Surrey, British Columbia (BC), Canada (Fig. 3) was selected for the LSP-ABM implementation and simulations. The main reason is the City of Surrey is one of the fastest growing cities in Canada (City of Surrey, 2013; Statistics Canada, 2013), with Clayton-Cloverdale being one of its most dynamic neighborhoods. The Clayton-Cloverdale has a strong vision for community design, with an extensive Neighborhood Concept Plan (NCP) (City of Surrey, 2006). These two factors make Clayton-Cloverdale a suitable site to represent the urban growth dynamics at a fine cadastral scale. Several geospatial datasets were used for the development of the LSP-ABM. Input datasets are land-use data for year 2011 in vector GIS format were used to generate geospatial data layers and information for the following land categories: parks, commercial sites, schools, proposed park sites, proposed commercial sites, proposed school sites, and cadastral land parcels. Geospatial data sets were obtained from the City of Surrey Open Data Catalogue
3.2. Agent descriptions
Please cite this article in press as: Dragicevic, S., & Hatch, K., Urban geosimulations with the Logic Scoring of Preference method for agent-based decision-making, Habitat International (2017), https://doi.org/10.1016/j.habitatint.2017.09.006
S. Dragicevic, K. Hatch / Habitat International xxx (2017) 1e15
6
Table 2 Description of the LSP-ABM components based on some aspects of the ODDþD protocol. Structural Elements Purpose Entities, Variables and Scales Process Overview and Scheduling Design Concepts Background Decision-Making Behavior Adaptation Sensing Prediction
Heterogeneity Observation Implementation Details Initialization Input Sub-models
Simulation of urban growth at fine spatial resolution and development in the City of Surrey as part of Metro Vancouver region, using longterm land use development data. Main entities are agents called city planner, developer, and resident. Variables are subdivided into a few categories: terrain and environment, amenities, transportation, and economic; the spatial scale is at the cadastral level i.e. individual residential housing units (approximately 400 e700m2) while temporal resolution is one year. Vector and raster GIS data were used and the model is explicitly spatial. The variables' criteria are combined using the LSP method related to individual resident agent types (family, single, and senior) as potential homebuyers. Variable weightings are changed based on each resident agent type, and updated at each iteration. New agents are added to the urban environment for each iteration that represent one year. The model is theoretical but use real geospatial datasets. Resident agents make decisions based on the LSP structure that is designed for each resident agent type. Each individual resident agent has their own homebuying characteristics that vary within the LSP structure outputs indicting individual agent’ ability to choose location of future residence, i.e. cadastral lot to occupy, and behave within their resident type. Resident agents change their choices based on what cadastral lots have been occupied by other resident agents in previous model iterations. Agents made decisions based on the variables in the LSP attribute tree. The tacit predictions are based upon the values for weights in the LSP aggregation structure so the agents will not do anything outside of the norm for particular type of agent, while at the same time having a certain range of autonomous decision-making ability in choosing the cadastral lot as future residence. The three types of resident agents collectively act as different heterogenous agents. Collectively output results indicate that the choice of cadastral lot and residence type are grouped around the defined idea surrounding the agent type e seniors and individuals will mostly choose mid- to high- rise buildings, family agents will choose single dwellings. The model was implemented in combination with Repast, Idrisi and ArcGIS software. Vector based GIS data at cadastral scale and information regarding the expected urban development and population growth defined by the City Plan. Selected number of resident agents are released in each simulation iteration to comply with the planned population growth. For each resident agent type (family, single, and senior) individual agent combines attributes using their own LSP aggregation structure. The outcomes of the LSP structure has been implemented in Idrisi GIS software. Two different scenarios have been designed to accommodate for growth as usual and accelerated growth.
Fig. 3. Clayton-Cloverdale neighborhood study site, located in the Municipality of Surrey, within Metro Vancouver, British Columbia, Canada.
Please cite this article in press as: Dragicevic, S., & Hatch, K., Urban geosimulations with the Logic Scoring of Preference method for agent-based decision-making, Habitat International (2017), https://doi.org/10.1016/j.habitatint.2017.09.006
S. Dragicevic, K. Hatch / Habitat International xxx (2017) 1e15
main purpose of the city planner agent within the proposed model is to ensure that logical and realistic development principles are followed given the anticipated population growth over the next several years within the study site. 3.2.2. Developer agents The flowchart for the developer agent reasoning is presented in Fig. 4. The Developer agents make decisions on areas within the study site to purchase for future development, as well as what type of lot to develop in a study site. These decisions are influenced by three factors. First, the number of resident agents expected to be searching for a new home. This is based on the number of agents that searched for a home in the previous year, multiplied by a predicted growth factor. Second, the type of resident agents anticipated to search for homes in the study area, which is also based on data from the previous year. Different resident agent types have different preferences, especially when considering which housing type they prefer within high, medium, low density land use. Developer agents attempt to fill the market demand by developing new sites in areas that fit the resident agent land-use preferences. Third, the developer agents focus on cost. Cost is represented as the sum of the lot cost as well as the anticipated cost to provide improvements to the land. For example, a low density single family lot is cheaper to develop than a high density apartment building. 3.2.2.1. Developer agent decision making algorithm. The developer agents have a primary goal of profit. This comes from building on desirable land parcels to maximize profit. Each iteration, the developer attempts to fill the demand for housing in two steps. Firstly, the proposed model determines an adequate number of parcels Ln for the developer agents to develop: Ln ¼ GfamRfam þ GsinRsin þ GsenRsen þ Cfam þ Csin þ Csen
(3)
where Rfam, Rsin, Rsen are the number of resident agents interested in family, single, and senior housing respectively; Gfam, Gsin, Gsen are growth factors for each of the three type of resident agents, and; Cfam, Csin, Csen are the number of resident agents of each type from the previous iteration but that are still searching for adequate housing.
7
Secondly, the developer agents determine the total number of new high density Hn, medium density Mn, and low density Ln land parcels to develop, each defined as:
Hn ¼
aHfam Gfam Rfam þ aHsin Gsin Rsin þ aHsen Gsen Rsen bHfam Gfam Rfam þ bHsin Gsin Rsin þ bHsen Gsen Rsen
Mn ¼
aMfam Gfam Rfam þ aMsin Gsin Rsin þ aMsen Gsen Rsen bMfam Gfam Rfam þ bMsin Gsin Rsin þ bMsen Gsen Rsen (4)
Ln ¼
aLfam Gfam Rfam þ aLsin Gsin Rsin þ aLsen Gsen Rsen bLfam Gfam Rfam þ bLsin Gsin Rsin þ bLsen Gsen Rsen
where ai and bj are coefficients corresponding to the suitability for high, medium, and low density housing for each of the three resident agent types. Once the land parcel number and type have been established for the current iteration, the developer agent searches for new candidate land parcels for the next development. Developers choose among the available parcels for development, and decide which areas would be most suitable for high, medium, and low density, based upon which parcels would give the most total profit. Each land parcel has a cost value (measured in dollars) as well as an associated cost of providing amenities to that parcel for a particular density (low, medium, or high). Each land parcel is then assigned a development cost value Cost dev as follows:
Costdev ¼
Total Cost þ Cost of Providing Amenities Total number of occupants that can live on one parcel (5)
The elements of stochasticity associated with the developer's lot choice are incorporated within the Costdev as it stochastically fluctuate in value, but developer agents are more inclined to develop lots that provide the lowest cost per occupant so it is set to choose the lowest Costdev. Once a candidate site for a particular land parcel type (high, medium, low density) is chosen, that parcel is made visible in the environment, and then made available for resident
Fig. 4. Flowchart of developer agent decision-making process.
Please cite this article in press as: Dragicevic, S., & Hatch, K., Urban geosimulations with the Logic Scoring of Preference method for agent-based decision-making, Habitat International (2017), https://doi.org/10.1016/j.habitatint.2017.09.006
8
S. Dragicevic, K. Hatch / Habitat International xxx (2017) 1e15
agents to potentially subdivide and later occupy. 3.2.3. Resident agents The urban residential environment is developed in part due to the actions of the resident agents. Resident agents look to occupy potential housing property (cadastral) lots built within the study site. As a result of their relocation into the study area, resident agents drive the demand for new lots as well as densification type of lots. Cadastral lots are designated as either high density (such as a low rise apartment), medium density (such as townhouses), and low density (such as a single family home), with density calculated as the number of occupants per square meter. Resident agents are added at each simulation iteration and are trying to find suitable housing by querying the set of vacant cadastral lots in the study area. Fig. 2 presents the number of steps of the resident agent decision-making process. First, resident agents are represented by three types: family, single, and senior resident, each with their own set of preferences and properties as the three most dominant groups of individuals to inhabit the study area. In the proposed LSPABM model, several variables are chosen that affect the resident agents demand for a particular cadastral lot. These variables each contains several criteria, and are incorporated in the LSP component of the model as follows: Terrain and Environment: Residential properties near parks and green spaces are at a premium. Natural and constructed amenities such as parks are valuable attributes in housing demand (Cho et al., 2006) thus distance to parks is considered. Moreover, the aspect in terms of direction that a lot is facing plays a role in determining house choice. The direction that a home is facing affects the amount of sunlight the home receives throughout the day. Generally, a south facing home receive the most sun during the day. In addition, north facing homes in the study area have views of the local mountains and are also preferred. Amenities: Local amenities have an impact in explaining demand for housing as well as housing prices, with proximity to parks and highways or commercial outlets and all play an important role in the growth of cities (McLeod, 1984; Vaz, 2016). Families with children seeking suburban housing are concerned about adequate schools nearby and are more likely to buy suburban homes with greater accessibility to primary and secondary schooling (Varady, 1995). In the suburban Cloverdale neighborhood of Surrey, at least 65% of the population resides in a household of 3 or more persons (City of Surrey, 2006), making accessibility to schools and other educational facilities very important factors. Therefore, distance to various amenities such as commercial outlets, schools or educational facilities are taken into consideration. Transportation: It is widely accepted that transportation corridors have an impact on urban sprawl (Ewing, Pendall, & Chen, 2003) consequently distance to highways, distance to transit lines, and distance to major collector roads are taken into consideration. Residents of the greater region of the study area, Surrey, BC, spend an average of 31 min commuting to work, with only 12.8% of those using public transportation (National Household Survey, 2013). For this reason, distance to highways and major collector roads play an important role in housing choice. In addition, there is a proposed high-speed transit station to be built in the study area and this will provide influence on potential homebuyers. Economic: Housing consumers overwhelmingly prefer singlefamily suburban homes to any other residential alternative (Glaeser, Gyourko, & Saks, 2006; Myers & Gearin, 2001). Therefore, information on year built, and type of lot (apartment, single family occupancy, senior housing facility, etc.) have been considered in this study. Moreover, future planning regimes and demographic changes may lead to a greater demand of medium and high density
housing. This is incorporated in the model by different resident agent types having differing preferences for low density (single family homes), medium density (townhouses), and high density (apartment complexes, low rise, and high rise buildings). The criteria (distance to roads, schools, aspect, etc.) and urban residential density type (low, medium, or high) are evaluated by resident agent for each vacant cadastral lot. These criteria are presented with suitability functions in a standardize manner (Fig. 5). Each resident agent type then combines the set of the standardized criteria using their own unique LSP aggregation structure in order to make final decision on the choice of housing lot depending on their own preferences. The LSP structures for each type family, single, and senior resident agent are presented on Figs. 6e8. The resident agents assign each cadastral lot a LSP suitability score that is an output of the LSP structure. The resident agents then query the entire set of vacant lots and occupy the most desirable lot - the lot with the highest LSP suitability score with respect to that individual agent type. If the LSP suitability scores of all of the unoccupied lots are below a threshold value, the agents leave the study site and seek housing elsewhere outside the study site. The LSP aggregators combine inputs by using the GCD and CPA (formulae (1) and (2) respectively) associated with LSP. For each agent, the individual weights of preference when combining two standardized inputs can vary ±0.1 from the values in Figs. 6e8; and the LSP aggregators can vary by one degree of simultaneity more or one degree of replaceability more than given in the aggregation structures. LSP aggregators for each presented aggregation structure are chosen to develop a canonical aggregation structure , 2011), one where the level of simultaneity (Dujmovi c & De Tre increases as the number of inputs combined in an individual aggregator increases. The weights used when combining inputs were determined based on weighting schemes used in previous LSP studies (Dujmovic et al., 2009), on the desired level of influence each input has on the overall suitability for each scenario, and on the desired level of influence on each input when combined with other inputs. When all input criteria are combined in the LSP aggregation structures the LSP suitability score for a particular resident agent type is obtained with respect to a particular cadastral lot. The LSP method used has the goal of determining suitable locations for residential development based on the combination of the number of given criteria and LPS aggregation structure thus represent human reasoning. Cadastral lots are assigned a suitability value reflecting their appropriateness for occupation with respect to a choice of the individual resident agent, based on their output from that particular resident agent type LSP aggregation structure. The resident agents iterate over the entire set of vacant lots and choose the lot with the highest LSP suitability score. If two or more resident agents choose the same lot, one is randomly assigned that lot and the other agents then having to choose from the remaining most suitable lots. 4. Results 4.1. LSP-ABM implementation The ABM was developed within the Recursive Porous Agent Simulation Toolkit (REPAST), Simphony edition, which is an objectoriented programming framework capable of integrating real geospatial data (North et al., 2013). The model consists of numerous components (Fig. 2), each acting independently, performing a specific task in the model, with all the components integrated to form the entire LSP-ABM. The LSP method was implemented in the Idrisi Selva GIS software and it was loose-coupled with the Esri ArcGIS 10 and REPAST software. This enabled the LSP-ABM model to operate on geospatial data at the cadastral-scale neighborhood,
Please cite this article in press as: Dragicevic, S., & Hatch, K., Urban geosimulations with the Logic Scoring of Preference method for agent-based decision-making, Habitat International (2017), https://doi.org/10.1016/j.habitatint.2017.09.006
S. Dragicevic, K. Hatch / Habitat International xxx (2017) 1e15
Fig. 5. Criteria functions for LSP aggregation structures.
9
Please cite this article in press as: Dragicevic, S., & Hatch, K., Urban geosimulations with the Logic Scoring of Preference method for agent-based decision-making, Habitat International (2017), https://doi.org/10.1016/j.habitatint.2017.09.006
Fig. 6. LSP aggregation structure for decision-making of a family resident agent.
Fig. 7. The LSP aggregation structure for decision-making of a single resident agent.
Please cite this article in press as: Dragicevic, S., & Hatch, K., Urban geosimulations with the Logic Scoring of Preference method for agent-based decision-making, Habitat International (2017), https://doi.org/10.1016/j.habitatint.2017.09.006
S. Dragicevic, K. Hatch / Habitat International xxx (2017) 1e15
11
Fig. 8. The LSP aggregation structure for decision-making of a senior resident agent.
with each cadastral parcel having its own individual properties and land-use type, such as: school, park, commercial, low density residential, medium density residential, or high density residential. The model runs iteratively, with land parcels changing from a vacant state to occupied state as residential agents choose to move in. At each iteration, more resident agents are added to the model using demographic data from the City of Surrey (City of Surrey Open Data Catalogue, 2013). Cadastral land parcels remain static throughout the simulation. Each cadastral parcel obtains the value for the resident agents to base their decisions on. Cadastral lots are assigned as either vacant or occupied. Once occupied, cadastral lots remain occupied throughout the simulation. Over time throughout the simulation process, resident agents choose desirable vacant cadastral lots, and the chosen lots then become occupied. Cadastral lots are also designated as either high density (such as a low-rise apartment), medium density (such as townhouses), and low density (such as a single family home), with density established based on the number of occupants per square meter. The LSP-ABM model verification was applied to ensure the correctness of the model logic and that the simulation outcomes are meaningful, for example the choice of urban housing happens on urban cadastral lots of specific designation and not on polygons representing roads or industrial areas. The model calibration has been accomplished comparing existing data to the long-range development defined by the City Plan (City of Surrey, 2013, 2006) to ensure it followed the expected growth density in the region over the long-term. Further, the calibration process helped determine optimal number of agents selected for each simulation iteration and that the growth of number of agents is at reasonable rate over time that is in accordance with the City Plan projected
population growth. Sensitivity analysis was performed and the results indicated, similar to other MCE methods, that the LSP method outputs are sensitive to changes in the choice of weights and aggregator factors - for example, for change from a “full conjunction” C to a “neutral” A. However, the sensitivity analysis indicates if the aggregation weights and factors are kept within a reasonable range plus or minus one factor: from a CA to a C-þ or Cþ- the model did not produce unreasonably different output results. Given aggregations and weights should be determined by the stakeholders involved the land development, this could potential help perform full model validation. 4.2. Simulation results For the purpose of model simulation two different scenarios were developed to reflect separate population trends in the study site. Scenario 1 follows projected growth rates established by the City of Surrey for the Cloverdale town center, a superset of the Clayton-Cloverdale neighborhood. Scenario 2 represent an accelerated growth of new residents and doubles the projected growth rates from the Scenario 1. Both scenarios involve perturbations to the number of agents that are added into the model at each iteration. Population statistics for the Clayton-Cloverdale neighborhood and the Cloverdale neighborhood projected growth rates (City of Surrey Open Data Catalogue, 2013) are as follows: 2013e2016 ¼ 2.28%, 2017e2021 ¼ 2.41%, 2022e2026 ¼ 2.32%, 2027e2031 ¼ 3.21%, 2032e2036 ¼ 2.48%, 2037e2041 ¼ 1.15%. These growth rates are used in Scenario 1 to help determine the number of resident agents to add in each iteration. Given a total number of agents, demographic data provided by the City of Surrey
Please cite this article in press as: Dragicevic, S., & Hatch, K., Urban geosimulations with the Logic Scoring of Preference method for agent-based decision-making, Habitat International (2017), https://doi.org/10.1016/j.habitatint.2017.09.006
12
S. Dragicevic, K. Hatch / Habitat International xxx (2017) 1e15
Fig. 9. Spatial distribution of cadastral lots and developed parcels for Scenario 1 at five year snapshots.
for the Cloverdale region is then used to determine the number of each of the three resident agent types to add into the model at each iteration. As of 2013, Cloverdale consists of approximately 75.6% families (households consisting of three or more people), and 5%
total seniors, with the remaining 19.3% considered single adults, or married without children. These three percentages are used to partition the total number of agents into each of the three resident agent categories: family, single, and senior. The temporal resolution
Fig. 10. Spatial distribution of cadastral lots and developed parcels for Scenario 2 at five year snapshots.
Please cite this article in press as: Dragicevic, S., & Hatch, K., Urban geosimulations with the Logic Scoring of Preference method for agent-based decision-making, Habitat International (2017), https://doi.org/10.1016/j.habitatint.2017.09.006
S. Dragicevic, K. Hatch / Habitat International xxx (2017) 1e15
of each simulation is one year, with simulations beginning in the year 2012, and new resident agents added at the beginning in year 2013. Figs. 9 and 10 present the simulations outcomes for both scenarios at different temporal snapshots. Each of the scenarios only perturb a certain number of agents added to the model at each iteration - there is not an extreme amount of disparity in the spatial distributions of land parcels built. For example, the high density land parcels built first in Scenario 1 are for the most part the same as the high density land parcels built first in Scenario 2. This is due to the developer agent logic being the same for both scenarios - they attempt to develop the most profitable parcels of land given the number and type of resident agents added into the simulation. However, differences are seen in the two scenarios in the rate of development of land parcels as well as in the rate of occupation of cadastral lots by resident agents. Some common features can be observed in the spatial distribution of development in both scenarios. First, there are planning constraints that prevent the northwest region of the study area from being developed. As a result, the low density lots in the northwest region of the study site remain unchanged throughout the simulations. Additionally, a limited amount of new low density housing is permitted in the study region, and its primary location of development is in the far northwest region of the study area, where “low density occupied” lots are developed. The reasoning behind low density development in this area is due that is far from major freeways, intersections, and existing transportation lines, while the areas where higher density developments are more desirable for planners and profitable for developers. The western boundary of the study site is delimited by the Fraser highway, a major freeway that funnels traffic into the central business district of Vancouver. One implication of this is that most of the major transportation routes throughout the study area eventually funnel out into downtown Vancouver by means of the Fraser highway. As a result, a large proportion of high density housing is located on the western edge of the study site. Newly developed medium density housing is located mostly in the middle of the study site due to many of the major streets running east to west, providing ideal candidate sites for medium density development. Fig. 11 shows the total number of newly developed low, medium and high-density land parcels, i.e. cadastral lots, created by developer agents after each iteration. For each Scenario, the number of low density land parcels reaches a maximum at some point between iteration 8 (2021) and iteration 12 (2024); reflecting city planner agent only allowing a small amount of newly zoned land to be used for low density lots, such as single family homes to accommodate future anticipated growth. Due to Scenario 2 simulating rapid growth, and thus many more resident agents added in each iteration, many more medium and high density lots are developed each iteration in comparison to Scenario 1. For Scenario 1, 113 tracts of land (where tracts are usually the size of one side of the street on an average city-block in the study area) are developed after 25 years, allowing for a total new occupancy of 9238 resident agents. Scenario 2 has 153 tracts developed after 25 years, allowing for 14309 resident agents. Each resident agent is an individual household (such as a family, or a couple) occupying one lot. The average household size for the Cloverdale neighborhood is 2.9 (City of Surrey Open Data Catalogue, 2013). Assuming this size remains constant, it implies 26,790 new persons are added to the study site in Scenario 1 after 25 years and 41,496 in Scenario 2. With the area where new development occurs being approximately 4 km2, this leads to a population density of approximately 6500 persons per km2 in Scenario 1 and 10,000 persons per km2 in Scenario 2. Comparing this to densities of 11,577 persons per km2 in the city of Vancouver, these results are within the realm of possibility for providing increased density into the study area.
13
Fig. 11. Total number of low, medium and high density parcels developed after each iteration (year i, expressed on the x-axis) for Scenarios 1 and 2.
5. Conclusions The developed LSP agent-based model simulates urban land use changes at the cadastral scale. It placed primary emphasis on residents as agents who drive land change as well as involving developers and city planners as agents in the simulation process. The resident agents have their decision-making logic simulated using the LSP method, wherein a set of preferences for each resident agent were aggregated together, and an overall suitability score for each individual land parcel was obtained. The proposed LSP-ABM as a geosimulation model, was implemented on the real datasets datasets representing controlled and fast urban expansion of West Clayton neighborhood that follow guidelines stipulated by the development plan of The City of Surrey. The LSP method is a generalized MCE method and represent in a novel way for modeling agent reasoning that is closer to human-decision making.
Please cite this article in press as: Dragicevic, S., & Hatch, K., Urban geosimulations with the Logic Scoring of Preference method for agent-based decision-making, Habitat International (2017), https://doi.org/10.1016/j.habitatint.2017.09.006
14
S. Dragicevic, K. Hatch / Habitat International xxx (2017) 1e15
This is established through the non-linear scoring and stepwise aggregation logic that provide a continuous variety of logical conditions and express more complex operators that are closer to actual human reasoning. However, the complete model validation and verification of agents' reasoning is the next logical step to enhancing the proposed geosimulation methodology. Additional sensitivity analysis on the suitability functions, LSP aggregation structures, as well as developer rules could further improve the model. Participatory agentbased modeling approaches (Becu, Neef, Schreinemachers, & Sangkapitux, 2008; Castella, Trung, & Boissau, 2005; Guyot & Shinichi, 2006; Zellner et al., 2012), although some not spatial, have been implemented in communities and with direct involvement of stakeholders for collective decision-making. Therefore, the involvement of actual stakeholders to build the LSP structures and choose weights for confirmation of agents' reasoning and behavior is one way to perform model validation. Moreover, urban planners would be able to directly involve various stakeholders to provide inputs into the model and observe various scenarios of outcomes of possible future urban growth and potential collective decision or policy making. The simulations could benefit from improved developer decision making analysis, such as giving developers the ability to develop in un-zoned and potentially less costly areas near the study site to see what effects it would have on the plans laid out by the City of Surrey for the Clayton-Cloverdale neighborhood. Linking information on land prices and residents' economic abilities would also benefit the enhancement of the proposed model. The LSP-ABM can be transposed to other study areas with different and larger spatial extents, and by developing additional LSP structures to accommodate reasoning of other parties involved in the process could be of assistance to urban planners to advance regional development guidelines. Acknowledgments This study was funded by the Natural Sciences and Engineering Research Council (NSERC) of Canada Discovery Grant (No. 3282242012) awarded to S. Dragicevic. The authors thank to two anonymous reviewers for constructive feedback on the manuscript. In addition, we thank Dr. Jozo Dujmovic for his valuable comments in early stage of this research and Mr. Stuart Jones, Senior Planner of City of Surrey for providing several geospatial datasets used in this study. References Albeverio, S., Andrey, D., Giordano, P., & Vancheri, A. (2010). The dynamics of complex urban Systems: An interdisciplinary approach. Physica-Verlag HD. An, L. (2012). Modeling human decisions in coupled human and natural systems: Review of agent-based models. Ecological Modelling, 229(24), 25e26. Arsanjani, J. J., Helbich, M., & Vaz, E. (2013). Spatiotemporal simulation of urban growth patterns using agent-based modeling: The case of Tehran. Cities, 32, 33e42. Batty, M. (2005). Agents, cells, and cities: New representational models for simulating multiscale urban dynamics. Environment and Planning A, 37, 1373e1394. Becu, N., Neef, A., Schreinemachers, P., & Sangkapitux, C. (2008). Participatory computer simulation to support collective decision-making: Potential and limits of stakeholder involvement. Land Use Policy, 25(4), 498e509. Benenson, I. (1998). Multi-agent simulations of residential dynamics in the city. Computers, Environment and Urban Systems, 22(1), 25e42. Bommel, P., Becu, N., Le Page, C., & Bousquet, F. (2016). Cormas: An agent-based simulation platform for coupling human decisions with computerized dynamics. In T. Kaneda, H. Kanegae, Y. Toyoda, & P. Rizzi (Eds.), Simulation and gaming in the network society. Translational systems sciences (Vol. 9, pp. 387e410). Singapore: Springer. Bonabeau, E. (2002). Agent-based modeling: Methods and techniques for simulating human systems. Proceedings of the National Academy of Sciences of the United States of America, 99(Suppl 3), 7280e7287. Bone, C., Dragicevic, S., & White, R. (2011). Modeling-in-the-middle: Bridging the gap between agent-based modeling and multi-objective decision-making for
land use change. International Journal of Geographical Information Science, 25(5), 717e737. buil, G., & Hardy, B. (Eds.). (2005). Companion modelling and multi Bousquet, F., Tre agent systems for integrated natural resource management in Asia (p. 360). Los Banos (Philippines): IRRI Press. Cabrera, A. R., Deadman, P. J., Brondízio, E. S., & Pinedo-Vasquez, M. (2010). Exploring the choice of decision making method in an agent based model of land use change. In Proceedings of 5th international congress on environmental modelling and software, Ottawa, Canada, July 2010. http://scholarsarchive.byu. edu/cgi/viewcontent.cgi?article¼2418&context¼iemssconference. Castella, J. C., Trung, T. N., & Boissau, S. (2005). Participatory simulation of land-use changes in the northern mountains of Vietnam: The combined use of an agentbased model, a role-playing game, and a geographic information system. Ecology and Society, 10(1). Cho, S. H., Bowker, J. M., & Park, W. M. (2006). Measuring the contribution of water and green space amenities to housing values: An application and comparison of spatially weighted hedonic models. Journal of Agricultural and Resource Economics, 485e507. City of Surrey. (2006). Community demographic profiles. Retrieved September 14, 2013 from: http://www.surrey.ca/business-economic-development/1417.aspx. City of Surrey. (2013). Official community plan bylaw No 18020, amended April 27, 2015. Planning and development, Surrey, Canada. Retrieved from: http://www. surrey.ca. Crooks, A., Castle, C., & Batty, M. (2008). Key challenges in agent-based modelling for geo-spatial simulation. Computers, Environment and Urban Systems, 32(6), 417e430. Dragicevic, S., Dujmovic, J., & Minardi, R. (2018). Modeling urban land-use suitability with soft computing: The GIS-LSP method. In J.-C. Thill, & S. Dragicevic (Eds.), GeoComputational analysis and modeling of regional systems (pp. 257e275). Springer. Dujmovi c, J. (1979). Partial absorption function. Journal of the University of Belgrade, EE Dept., Series Mathematics and Physics, 659, 156e163. Dujmovi c, J., & Bai, H. (2006). Evaluation and comparison of search engines using the LSP method. Computer Science and Information Systems, 3(2), 31e56. Dujmovi c, J. J., & Bayucan, A. R. (1997). A Quantitative method for software evaluation and its application in evaluating windowed environments. In Proceedings of software engineering conference, IASTED, San Francisco. , G. (2011). Multicriteria methods and logic aggregation in Dujmovi c, J., & De Tre suitability maps. International Journal of Intelligent Systems, 26(10), 971e1001. , G., & Dragicevic, S. (2009). Comparison of multicriteria Dujmovic, J., De Tre methods for land-use suitability assessment. In Proceedings of 2009 IFSA world congress/EUSFLAT conference (pp. 1404e1409). European Society for Fuzzy Logic and Technology (EUSFLAT). , G., & Van de Weghe, N. (2008). Suitability maps based on the Dujmovi c, J. J., De Tre LSP method. Modeling Decisions for Artificial Intelligence, 15e25. Dujmovi c, J. J., & Nagashima, H. (2006). LSP method and its use for evaluation of Java IDEs. International Journal of Approximate Reasoning, 41(1), 3e22. Dujmovi c, J., & Scheer, D. (2010). Logic aggregation of suitability maps. In Proceedings of fuzzy systems (FUZZ), 2010 IEEE international conference (pp. 1e8). , G., & Weghe, N. (2010). LSP suitability maps. Soft Computing, 14(5), Dujmovi c, J., Tre 421e434. Ettema, D., de Jong, K., Timmermans, H., & Bakema, A. (2007). PUMA: Multi-agent modelling of urban systems. In Modelling land-use change (pp. 237e258). Springer Netherlands. Evans, T. P., & Kelley, H. (2004). Multi-scale analysis of a household level agentbased model of landcover change. Journal of Environmental Management, 72(1), 57e72. Ewing, R., Pendall, R., & Chen, D. (2003). Measuring sprawl and its transportation impacts. Transportation Research Record: Journal of the Transportation Research Board, 1831(1), 175e183. Fontaine, C. M., & Rounsevell, M. D. a (2009). An agent-based approach to model future residential pressure on a regional landscape. Landscape Ecology, 24(9), 1237e1254. Ghavami, S. M., Taleaia, M., & Arentze, T. (2016). Socially rational agents in spatial land use planning: A heuristic proposal based negotiation mechanism. Computers, Environment and Urban Systems, 60, 67e78. Glaeser, E. L., Gyourko, J., & Saks, R. E. (2006). Urban growth and housing supply. Journal of Economic Geography, 6(1), 71e89. Graniero, P. A., & Robinson, V. B. (2006). A probe mechanism to couple spatially explicit agents and landscape models in an integrated modeling framework. International Journal of Geographical Information Science, 20(9), 965e990. Grimm, V., Berger, U., DeAngelis, D. L., Polhill, J. G., Giske, J., & Railsback, S. F. (2010). The ODD protocol: A review and first update. Ecological Modelling, 221(23), 2760e2768. Groeneveld, J., Müller, B., Buchmann, C. M., Dressler, G., Guo, C., Hase, N., et al. (2017). Theoretical foundations of human decision-making in agent-based land use models e a review. Environmental Modelling & Software, 8, 39e48. Guyot, P., & Shinichi, H. (2006). Agent-based participatory Simulations: Merging multi-agent systems and role-playing games. Journal of Artificial Societies and Social Simulation, 9(4). Hatch, K., Dragicevic, S., & Dujmovi c, J. (2014). Logic Scoring of Preference and spatial multicriteria evaluation for urban residential land use analysis. Lecture Notes in Computer Science, 8728, 64e80. Heppenstall, A., Malleson, N., & Crooks, A. (2016). Space, the final Frontier”: How good are agent-based models at simulating individuals and space in cities?
Please cite this article in press as: Dragicevic, S., & Hatch, K., Urban geosimulations with the Logic Scoring of Preference method for agent-based decision-making, Habitat International (2017), https://doi.org/10.1016/j.habitatint.2017.09.006
S. Dragicevic, K. Hatch / Habitat International xxx (2017) 1e15 Systems, 4(1), 9. Huang, Q., Parker, D. C., Filatova, T., & Sun, S. (2014). A review of urban residential choice models using agent-based modeling. Environment and Planning B, 41(4), 661e689. Irwin, E. G., & Bockstael, N. E. (2002). Interacting agents, spatial externalities and the evolution of residential land use patterns. Journal of Economic Geography, 2(1), 31e54. Jjumba, A., & Dragi cevi c, S. (2011). High resolution urban land-use change modeling: Agent iCity approach. Applied Spatial Analysis and Policy, 5(4), 291e315. Kelley, H., & Evans, T. P. (2011). The relative influences of land-owner and landscape heterogeneity in an agent-based model of land-use. Ecological Economics, 70(6), 1075e1087. Kocabas, V., & Dragicevic, S. (2013). Bayesian networks and agent-based modeling approach for urban land-use and population density change: A BNAS model. Journal of Geographical Systems, 15(4), 403e426. Ligmann-Zielinska, A., & Jankowski, P. (2007). Agent-based models as laboratories for spatially explicit planning policies. Environment and Planning B: Planning and Design, 34(2), 316e335. Ligtenberg, A., Bregt, A. K., & Van Lammeren, R. (2001). Multi-actor-based land use modelling: Spatial planning using agents. Landscape and Urban Planning, 56(1e2), 21e33. Li, X., & Liu, X. (2007). Defining agents' behaviors to simulate complex residential development using multicriteria evaluation. Journal of Environmental Management, 85(4), 1063e1075. Lim, K., Deadman, P. J., Moran, E., Brondizio, E., & McCracken, S. (2002). Agent-based simulations of household decision making and land use change near Altamira, Brazil. In R. Gimblett (Ed.), Integrating geographic information systems and agentbased modeling techniques for simulating social and ecological processes (pp. 277e310). Oxford Press. Malczewski, J. (2006). GIS-based multicriteria decision analysis: A survey of the literature. International Journal of Geographical Information Science, 20(7), 249e268. Matthews, R. B., Gilbert, N. G., Roach, A., Polhill, J. G., & Gotts, N. M. (2007). Agentbased land-use models: A review of applications. Landscape Ecology, 22(10), 1447e1459. McLeod, P. B. (1984). The demand for local amenity: An hedonic price analysis. Environment and Planning A, 16(3), 389e400. Montgomery, B., & Dragicevic, S. (2016). Comparison of GIS-based Logic Scoring of Preference and multi-criteria evaluation methods: Urban land use suitability. Geographical Analysis., 48, 427e447. Montgomery, B., Dragicevic, S., Dujmovic, J., & Schmidt, M. (2016). A GIS-based Logic Scoring of Preference method for evaluation of land capability and suitability for agriculture. Computers and Electronics in Agriculture, 124, 340e353. Müller, B., Bohn, F., Dreßler, G., Groeneveld, J., Klassert, C., Martin, R., et al. (2013). Describing human decisions in agent-based models - ODDþD, an extension of the ODD protocol. Environmental Modelling & Software, 48, 37e48. Myers, D., & Gearin, E. (2001). Current preferences and future demand for denser
15
residential environments. Housing Policy Debate, 12(4), 633e659. National Household Survey. (2013). Statistics Canada. Retrieved October 28, 2013 from: http://www.statcan.gc.ca/daily-quotidien/130626/dq130626b-eng.htm. North, M. J., Collier, N. T., Ozik, J., Tatara, E. R., Macal, C. M., Bragen, M., et al. (2013). Complex adaptive systems modeling with Repast simphony. Complex Adaptive Systems Modeling, 1(1), 1e26. Open Data Catalogue. (2013). City of Surrey. Retrieved June 05, 2013 from: https:// www.surrey.ca/city-services/658.aspx. Parker, D. C., Manson, S. M., Janssen, M. A., Hoffmann, M. J., & Deadman, P. (2003). Multi-agent systems for the simulation of land-use and land-cover change: A review. Annals of the Association of American Geographers, 93(2), 314e337. Rebolledo, B., Gil, A., Flotas, Z., & Sanchez, J. A. (2016). Assessment of groundwater vulnerability to nitrates from agricultural sources using a GIS-compatible logic multicriteria model. Journal of Environmental Management, 171, 70e80. Rounsevell, M., Robinson, D. T., & Murray-Rust, D. (2012). From actors to agents in socio-ecological systems models. Philosophical Transactions of The Royal Society B, 367, 259e269. Shan, Y., & Zhu, X. (2007). Simulation of automatic addressing behavior based on urban residential land dynamics multi-agents model. In Proceedings of IEEE international conference on automation and logistics (pp. 1358e1363). Statistics Canada. (2013). Statistics Canada. Retrieved July 7, 2013, from: http:// www.statcan.gc.ca/start-debut-eng.html. Stevens, D., & Dragicevic, S. (2007). A GIS-based irregular cellular automata model of land-use change. Environment and Planning B: Planning and Design, 34(4), 708e724. Torrens, P., & Benenson, I. (2007). Geographic automata systems. International Journal of Geographical Information Science, 19(4), 385e412. Valbuena, D., Verburg, P. H., Bregt, A. K., & Ligtenberg, A. (2010). An agent-based approach to model land-use change at a regional scale. Landscape Ecology, 25(2), 185e199. Varady, D. P. (1995). Selling cities: Attracting homebuyers through schools and housing programs. Albany: State University of New York Press. Vaz, E. (2016). The future of landscapes and habitats: The regional science contribution to the understanding of geographical space. Habitat International, 51, 70e78. Villamor, G. B., van Noordwijk, M., Troitzsch, K. G., & Vlek, P. L. G. (2012). Human decision making in empirical agent-based models: Pitfalls and caveats for land€ hring, & U.Lotzmann (Eds.), Prouse/change policies. In K. G. Troitzsch, M. Mo ceedings 26th European conference on modelling and simulation (ECMS). Waddell, P., Borning, A., Noth, M., Freier, N., Becke, M., & Ulfarsson, G. (2003). Microsimulation of urban development and location choices: Design and implementation of UrbanSim. Networks and Spatial Economics, 3(1), 43e67. White, R., Engelen, G., & Uljee, I. (2015). Modeling cities and regions as complex systems: From theory to planning applications. MIT Press. Zellner, M. L., Lyons, L., Hoch, C. J., Weizeorick, J., Kunda, C., & Milz, D. (2012). Modeling, learning and planning together: An application of participatory agent-based modeling to environmental planning. URISA Journal, 24(1), 77e92.
Please cite this article in press as: Dragicevic, S., & Hatch, K., Urban geosimulations with the Logic Scoring of Preference method for agent-based decision-making, Habitat International (2017), https://doi.org/10.1016/j.habitatint.2017.09.006