Spatially-explicit prediction of low-density peri-urban development: comparison between urban and rural scenarios in the Moreton Bay Region in South East Queensland,Australia

Abstract

Mainstream urban modelling literature focuses on urban expansion featured by a relatively fast urbanisation process, but relatively less research is available to understand and model the slow-paced urban and rural land development in the low-density peri-urban context. This study aims to address this knowledge gap by simulating the urban and rural land development in the Moreton Bay Region in South East Queensland (SEQ), Australia using two cellular automata (CA) models that are coupled with a generalised simulated annealing (GSA) algorithm. With the total land available for development estimated using a Markov Chain model, the GSA-CA urban and rural models were developed, respectively, to simulate urban and rural land development from 1991 to 2011, and then to predict their future development to 2041 following vigorous model calibrations. The modelling results illustrate three snapshots of the predicted spatial patterns of urban and rural development in 2021, 2031 and 2041, with moderate growth in both the urban and rural areas over time, but with urban development occurring in a more compact form than rural development. The GSA-CA modelling approach is capable of optimising the CA transition rules and has the potential to be applied to other geographical contexts to support regional planning, decision-making and scenario designation for future land development in cities that have entered the saturation phase of urbanisation.

Keywords

Low-density land development cellular automata generalised simulated annealing algorithm scenario prediction peri-urban areas Australia

Introduction

Urbanisation, along with its causes and consequences, has been studied extensively in the field of urban and regional sciences (McGranahan and Satterthwaite, 2014). The level of urbanisation varies across regions and countries, being high in general in developed countries (e.g. above 80% in the U.S., Canada, Australia and European countries) and relatively low in developing countries (e.g. 40%–50% in China, Brazil and India; below 30% in some African countries) (The World Bank, 2020). Globally, the migration of rural population to urban living has been increasing rapidly over the past three decades; however, this rapid growth is slowing down in some regions in Asia (e.g. China), Africa (e.g. South Africa) and in some more developed countries (e.g. United States and Australia) over the past decade (The World Bank, 2020). Theoretically, the process of urban development has been recognised to undergo three phases – the initial, acceleration and saturation phases (Northam, 1979). At the initial phase of urban development, the majority of population are engaged in subsistence farming in a rural context. The primitive and relatively stagnant state of agriculture restricts the migration from rural to urban areas to a certain degree. At the acceleration phase, industrialisation and industrial upgrading trigger rapid urbanisation and attract population to move from rural to urban areas. It is usually accompanied by long-term land use change, such as the conversion of agricultural and rural land to urban space (Foley et al., 2005; Neumann et al., 2015). At the saturation phase, urbanisation slows down when more people chose to live outside the inner-city centres and move towards outer suburbs and low-density areas surrounding the metropolitan cities – often defined as peri-urban areas (Healy and Hillman, 2008). In Australia, the peri-urban areas surrounding major capital cities house a total of 1.85 million people in 2016, an 18% increase from 1.57 million in 2006 (Australian Bureau of Statistics (ABS), 2016a). Peri-urban areas are often in the transition of diverse land uses (e.g. shifting from water catchments, forestry, mineral extraction, tourism, recreation and productive farming to urban construction), and offer a unique ambiance to people loving a country-side lifestyle with easy access to urban facilities.

However, peri-urban areas are distinct to urban areas in terms of the demographic composition and socioeconomic status of residents and the configurations of public facilities and services. The relative low-density development in peri-urban areas is often associated with some well-recognised problems such as high infrastructure costs, car dependency, the loss of agriculture and open space, less educational and employment opportunities and limited access to public services and resources (Bandias and Vemuri, 2005; Healy and Hillman, 2008; Newton, 2012). Understanding the different processes of peri-urban growth (e.g. the initial stage versus well-developed stage) needs to take into consideration the characteristics of the region along with urban configuration, resource distribution, infrastructure provision and environmental pressures. These critical components are often subject to planning conflicts and dilemmas evident in the metropolitan areas (Freeman and Hancock, 2017; Park, 2017).

Scholars commonly utilise urban modelling techniques to predict different land development scenarios for urban planning, infrastructure configuration, city governance and environmental protection (e.g. Aburas et al., 2016; Lagarias, 2015; Tong and Feng, 2019; Wahyudi et al., 2019). Taking the historical and current land use patterns into consideration, scenario prediction has the capability to simulate and assess feasible options for future land use change by setting up possibilities for different development pathways (Liu, 2008). However, most urban models are developed and applied to simulate urban expansion featured by urbanisation at the initial or acceleration phase (Aburas et al., 2016) where the change of urban land over time is substantial and easy to be captured by the models. Less research is available to simulate urban development at the saturation phase featured by a slow growth pace in the low-density peri-urban context, where the change of urban and rural land over time may be too minor and subtle to be captured using traditional urban models without using advanced algorithms – a knowledge deficit that this study aims to fulfil.

Models of various types have been developed in the current scholarship to study land use change and predict urban growth, including cellular automata (CA) as a well-recognised approach for urban simulation (De Almeida et al., 2003; Li and Yeh, 2002). There are general two types of approaches in urban simulation – the top-down approach referring to that modelling work starts from the general to the specific, and the bottom-up approach referring to that modelling work starts at the specific and moves to the general (Liu, et al., 2019). Given this, CA modelling is always considered as a bottom-up approach in which cities are considered as complex adaptive systems and the outcomes of urban change dynamics are simulated through the actions of individual cells at local scale (Clarke and Gaydos, 1998). Therefore, unlike the top-down approach which develop models based on general rule/s that are applied globally to a system and works down to local specifics, the outcomes of a CA model are more complex and uncertain (Batty, 2007). Over time, a series of CA-derived models have been developed for specific modelling purposes, such as the Slope, Land use, Exclusion, Urban extent, Transportation and Hillshade (SLEUTH) model (Clarke et al., 1997; Jantz et al., 2004; Mahiny and Clarke, 2012), the Future Land Use Simulation (FLUS) model (Liang et al., 2018; Liu et al., 2017), the California Urban Futures (CUF) model (Landis and Zhang, 1998), the Conversion of Land Use and its Effects (CLUE-S) model (Verburg et al., 2002), the GIS-based land use change (GEOMOD) model (Pontius and Chen, 2006) and the FUTure Urban-Regional Environment Simulation (FUTURES) model (Meentemeyer et al., 2013). More recently, agent-based models (ABM) have also been developed to account for the impact of human decisions on urban development (Li and Liu, 2008; Wahyudi et al., 2019; Zhang et al., 2011). In an ABM, an urban system is modelled as a collection of autonomous decision-making parties termed agents. Each agent assesses its situation individually and makes decisions on the basis of information available to the agent. Collectively, the decision made by groups of agents affects the probability of land transition from rural to urban (Li and Liu, 2008). By deriving a series of spatial and non-spatial factors that drive or constrain urban growth, CA and/or CA-ABM models can be used to generate plausible predictions of future urban development (Batty, 1997; Liu et al., 2019). While most of these models perform well to simulate urban growth in specific contexts (e.g. with rapidly urbanisation rate in developing countries), there are challenges to apply such models to simulate urban growth in other geographical contexts, or to simulate other urban change processes such as urban densification and consolidation, rural development or the low-density peri-urban growth (Liu, et al., 2019), the latter of which is the subject of this paper.

Another distinctive feature of this study is the utilisation of a generalised simulated annealing (GSA) algorithm in optimising the transition rules in a CA model to predict urban and rural land use change. In the current scholarship, various statistical methods (e.g. principal component analysis and logistic regression) have been employed to define CA transition rules (Feng and Liu, 2013), but those methods have the drawback of not sufficiently quantifying the effects of spatial variables on urban growth due to the spatial correlation amongst the variables (Al-Ahmadi et al., 2009; Li and Yeh, 2002). To address this drawback, optimisation methods have been increasingly applied in CA modelling, including the artificial immune system (Liu et al., 2010), ant colony optimisation (Yang et al., 2012), bee colony optimisation (Yang et al., 2013), particle swarm optimisation (Feng and Liu, 2013), cuckoo search (Cao et al., 2015) and support vector machine (Yang et al., 2008). These methods have the advantages of quantifying the effects of spatial variables on the probability of land use conversion via weight-definitions and diminishing the effects caused by the potential multi-collinearity among the spatial variables (Tong and Feng, 2019). Among these optimisation methods, the GSA algorithm has gain increasing attention as a heuristic search method to retrieve the spatial parameters of a CA model through a combinatorial optimisation process. Originally proposed by Van Laarhoven and Aarts (1987) as a computational algorithm to measure the solid-annealing process, the GSA algorithm has been applied to resolve a wide range of optimisation problems in landscape management (Baskent and Jordan, 2002) and land use allocation (Santé-Riveira et al., 2008). Earlier work by Feng and Liu (2013) and Feng et al. (2018) have compared the GSA and other optimisation methods in urban CA modelling and demonstrated that the GSA algorithm is more effective to improve modelling performance and accuracy. Thus, this study employs the GSA algorithm to optimise the land transition rules to limit the extent of mismatch between the simulated and actual land use maps and increase the simulation accuracies.

This paper aims to simulate the urban and rural land development in a low-density peri-urban area using two GSA-optimised CA models (i.e. the urban and rural GSA-CA models) in the Moreton Bay Region in South East Queensland (SEQ), Australia, under two sets of regional planning policies. These two policies are urban-oriented development, which focuses on urban consolidation, infill and higher density urban development and rural-oriented development, which allows vacant and/or agricultural land to be converted to low-density residential land via rezoning. Through the scenario prediction for urban and rural land development from 2011 to 2041, the two GSA-CA models capture the subtle differences of urban and rural development over time. The findings provide insights and policy implications for regional planning and land use management.

Study context and data

Study context

Our study area is the Moreton Bay Regional Council (MBR) (Figure S1 in Data and study context in the Supplementary Materials), a local government area located to the north of the state capital city of Brisbane in SEQ. South East Queensland is one of the fast-growing regions in Australia, comprising of 71% of the population in Queensland and a population growth rate of around 2.2% per annum since 2017 (Queensland Government, 2020). Growth management policies and plans in SEQ, such as the latest gazetted regional plan by Queensland Government – ShapingSEQ: SEQ Regional Plan 2017 (Queensland Government, 2016), hereinafter termed as ShapingSEQ – and the local government growth management strategies have sought to manage urban sprawl by delimiting growth boundaries, implementing urban consolidation policies in order to limit urban expansions. Moreton Bay Regional sits in the northern growth corridor of SEQ and is approximately 60 km to the Brisbane central business district. It comprises an area of 2037 km², with a population of 476,639 in 2020 (Moreton Bay Regional Council, 2020). According to the regional plan in ShapingSEQ (Queensland Government, 2017), MBR is projected to have a total population of 656,000 in 2041. To align with this planning timeframe in ShapingSEQ, we configured our model using 2041 as the ending year. Moreton Bay Regional serves as a local tourism area linking Brisbane and the Sunshine Coast, with Moreton Bay Marine Park providing a variety of activities such as recreational and commercial fisheries to nearly three million local residents (Queensland Government, 2019). Moreton Bay Regional is a relatively low-density region with a small urban core, several major towns including Narangba, Burpengary, Morayfield, Caboolture and Elimbah that are mainly distributed along the interstate highway M1 and the railway line, and a large agricultural and rural hinterland; it has great potential to attract incoming migrants in future (Moreton Bay Regional Council, 2020). The low-density region is characterised by dwellings that are predominately with one or no more than two floors, and a population density of around 234 persons per square km in 2020, much lower than that of 346 persons per square km in Brisbane (ABS, 2021). As such, this region is an ideal testbed for modelling low-density peri-urban land use change to better understand its future development trajectories.

Data collection and manipulation

We collected spatial data at two time points: 1991 and 2011 (Table S1 in the section Data and Study Context in the Supplementary Material), including: (1) land use maps, originally in seven categories (residential, commercial, industry, vacant, conservation and parks, water bodies and agricultural land, shown in the section Data and Study Context in Figure S2), were reclassified into two types of land uses for CA modelling. These two types are non-urban (including rural residential, agriculture, vacant, conservation and parks) and urban (including urban residential, industrial and commercial) while water bodies were excluded from modelling; (2) digital elevation model (DEM) with a spatial resolution at 30 m and a vertical resolution at 10 cm computed from Light Detection and Ranging (LIDAR) data; (3) locations of major cities and townships; (4) transportation networks including major roads, railway lines and stations; and (5) a coastal hazard map produced by Department of Environment and Heritage Protection, Queensland Government (2013) which delineate areas of land that are subject to coastal erosion, storm tide inundation and include a projected 0.8 m sea level rise by 2100 due to the impact of climate change. All spatial datasets were converted to raster data in GeoTiff format at 30 × 30 m cell size for modelling. The ‘maximum area’ encoding method was used for data conversion in ArcGIS Pro 2.8; this encoding method assigns the land use type of areas with the largest areal size in a cell as the land use type of that cell when converting polygon to raster.

Methods

Using Markov Chain model to define the land quantity for future development

A Markov Chain is a stochastic model describing a sequence of possible events in which the probability of each event depends on the state attained in the previous event (Wu et al., 2019). It has been commonly applied in the prediction of geographical events and used in urban modelling and simulation (Gidey et al., 2017). In the context of modelling land use change, the Markov Chain not only explains the quantification of the conversion between land use types but also reveals the transfer rate among different land use types. The prediction of land use changes is calculated using equation (1) (Sang et al., 2011)

S_{i} (t + 1) = P_{i} \times S_{i} (t)

(1)

where

S_{i} (t)

and

S_{i} (t + 1)

are the status of cell

i

at time

t

and

t + 1

, respectively;

P_{i}

is the land transition probability of cell

i

which is generated using the urban CA model as we introduce next. Herein, we use Markov Chain to define the quantity of land needed for future development, which is used to cease the modelling process when the number of urban cells reaches the predicted land quantity.

The urban CA model

A key step for building an urban CA model is to define the transition rules, which determine the state of a cell in subsequent time-steps as a result of the collective effect of the current state of the cell and its neighbouring cells. A conceptual formula of the transition rules can be written as a combinational function of the driving factors, neighbourhood effects, as well as constraints of future development (Feng and Tong, 2018)

O T P_{i, t} = \frac{(P T P_{i} \cdot {(1 + σ)}^{t - 1} + E N_{i} \cdot φ) ·Con (S_{i})}{2}

(2)

where

O T P_{i, t}

is the overall transition probability of cell

i

at time

t

;

Con (S_{i})

is the constraint of urban growth;

P T P_{i}

denotes the partial transition probability of cell

i

defined in equation (3);

E N_{i}

is the effect of neighbourhood defined in equation (4);

P T P_{i}

and

E N_{i} \cdot φ)

are controlled by the scaling factor

σ

and

φ

. Here,

σ

is a time-dependent parameter which adjusts the decaying effect of the land transition probability over the simulation period; its value is set through extensive simulation experiments and usually ranges from 0.0 to 0.1, with a higher value indicating a stronger scaling effect (Feng and Tong, 2020).

φ

is a local parameter that adjusts the extent of neighbourhood effects on the target cell; its value usually ranges from 0.5 to 1.0, with a higher value indicating a stronger neighbourhood effect. These two scaling factors are used in the model to better capture the interaction between the target cell and its nearby cells in order to improve the modelling accuracy (Feng and Tong, 2020). The values of these two scaling factors in our models were set to 0.01 and 0.8, respectively, following extensive experiments in our study context. These values can be calibrated according to specific research context to represent more effectively the impact of different driving factors and neighbourhood effects on urban change.

The partial transition probability $(P T P_{i})$ of cell $i$ is expressed as (Feng and Liu, 2013)

PT P_{i} = \frac{e x p (β_{0} + \sum_{k = 1}^{l} β_{k} F_{k})}{1 + e x p (β_{0} + \sum_{k = 1}^{l} β_{k} F_{k})}

(3)

where

F_{k} (k = 1, ..., l)

is a set of spatial driving factors of urban growth (e.g. distance to urban centres, main roads and other service facilities). The parameters

β = [β_{k} (k = 0, ..., l)]

define the magnitude of how the driving factors

F_{k}

affect urban growth (Zhao et al., 2016). In our study, a GSA algorithm (defined in equations (6)–(8)) is used to automatically retrieve and optimise the parameters

(β_{k})

to reflect the effect of each driving factor on urban growth;

β_{0}

is a random value defined at the initial stage of GSA process that is optimised and fixed by the GSA algorithm. The reason of using exp function is because the relationship between the driving factors and the land transition probability is not linear but follows an exponential form; this function has been employed in previous studies (e.g. Feng and Liu, 2013; Feng and Tong, 2020).

The effect of neighbourhood $(E N_{i})$ on land conversion is defined as the impact of cells within a square neighbourhood surrounding the central cell $i$ , indicating how much the neighbourhoods drive the central cell $i$ towards an urban state. The effect of neighbourhood can be written as (Dahal and Chow, 2015)

E N_{i} = \frac{\sum_{n \times n} C e n t r a l C e l l_{i} (S t a t e = U r b a n, 1; S t a t e = N o n U r b a n, 0)}{n \times n - 1}

(4)

where

n

is the side length of the neighbourhood; herein we adopted a 5 by 5 neighbourhood (n = 5) that has been commonly applied in the CA literature (e.g. Feng and Liu, 2013).

The physical or landscape constraints $Con (S_{i})$ include rivers and broad water bodies, natural conservation and coastal hazard areas that restrict the land transition from non-urban to urban. $Con (S_{i})$ is assigned to 1 if cell $i$ is not constrained, or 0 otherwise (Wu, 2002).

Collectively, ${OTP}_{i}$ represents the conversion probability of cell $i$ at the current time t; its state at a subsequent time t + 1 can be determined by comparing the ${OTP}_{i}$ with a specified threshold $R_{i}$ , defined by experiments and testing of existing studies in different geographic contexts in the previous studies and normally ranges from 0 to 1 (e.g. Feng and Tong, 2020; Wu 2002). In our study, $R_{i}$ is set to 0.278; details on the selection of this value is provided in the section Threshold Detection in the Supplementary Materials. The decision of non-urban and urban conversion is shown as (Wu, 2002)

S_{i}^{t + 1} = {\begin{matrix} C o n v e r t f r o m N o n - u r b a n t o t o U r b a n, O T P_{i}^{t} > R_{i} and S_{i}^{t} = = Non - urban \\ {\begin{matrix} R e m a i n a s U r b a n, S_{i}^{t} = = Urban o r \\ R e m a i n a s N o n - u r b a n, O T P_{i}^{t} \leq R_{i} \end{matrix} \end{matrix}

(5)

where

S_{i}^{t + 1}

is the state of cell i at the subsequent time step (t + 1). If

O T P_{i}^{t}

is larger than

R_{i}

at time t, a non-urban cell

i

will be converted to urban cell at time t + 1; otherwise, its state remains as non-urban.

The optimised GSA-CA model

We optimised the CA model using a GSA algorithm, given its capacity of improving the performance and accuracy of the CA model has been demonstrated in previous study (Feng and Liu, 2013; Lin et al., 2001; Li et al., 2021). Hereinafter, we term our model as GSA-CA model. Initially referring to the thermodynamic process of solid cooling (Van Laarhoven and Aarts, 1987), the term ‘annealing’ means a solid-cooling process (e.g. metal or glass) through which the solid atoms change from a high to low level of energy by losing thermal mobility (Kirkpatrick et al., 1983: 671). The slow cooling or optimisation process can be achieved through a control parameter, T (representing ‘temperature’ in the thermodynamics system), and an annealing schedule that drives the optimisation process. With the advantage of solving large-scale optimisation problems and reaching a global optimum in a general system, the GSA process is accompanied by an annealing schedule that drives the optimisation process. The procedure of using GSA to optimise a set of parameters $(β_{0}, \dots β_{k})$ for the transition rules of a CA model is shown in Figure 1, with the five major steps described below.

Figure 1.

The GSA optimisation procedures used to retrieve CA parameters, revised from Feng and Liu (2013: 453). Note: SA = spatial annealing algorithm.

The first step of the GSA optimising process is to construct an objective function $O F_{(β)}$ , with the purpose of mathematically minimising the CA modelling residuals between a simulated result and an actual value, which can be written as (Feng and Liu, 2013)

O F_{(β)} = \sqrt{\sum_{i = 1}^{μ} {[P T P_{i} (β) - O B S_{i}]}^{2}}

(6)

where

β

is a set of CA parameters;

P T P_{i} (β)

is computed according to equation (3) representing the partial transition probability of cell

i

;

O B S_{i}

denotes the observed conversion state for cell

i

, with a value of either 0 (no change between 1991 and 2011: urban to urban, or non-urban to non-urban) or 1 (change between 1991 and 2011: non-urban to urban);

μ

is the number of sample cells. Herein, equation (6) is designated referring to the Least Squares Estimates – calculating the residuals between

P T P_{i}

and

O B S_{i}

in a square form and taking the square root of the sum up of residuals in each observation. In this way, it best captures the difference between a simulated result and the actual value compared to the simple subtraction from

P T P_{i}

O B S_{i}

which may introduce significant data bias. Ideally, using

O T P

in equation (6) is optimal but it involves heavy computational iterations. Using

P T P

largely reduces the computation time and generates results that are also acceptable according to other experimental results reported in existing studies (e.g. Feng and Liu, 2013). In addition, we used the sample cells we retrieved initially from the land use maps and driving factor maps referring to the experiences of advanced modelling techniques (e.g. machine learning models, Batista et al., 2004). This sampling method has the advantage to improve the generalisation of our model and its modelling performance for prediction, and to avoid the overfitting of the trained model.

The second step is to assign an initial set of parameters for a CA model, based on which the GSA algorithm can be applied to search for potential solutions to improve the model performance. The initial solution in our model was generated using a logistic regression method, as was used in previous studies (Liu and Feng, 2012; Wu, 2002). The initial solution was then extended to a value range to enable the GSA to test and select the optimal candidate solution (Table S5). If the initial parameter is smaller than zero, the upper bound of the value range is set to zero and the lower bound is set as double the initial value of the parameter; if the initial parameter is larger than zero, the lower bound of the value range is set to zero and the upper bound is double the initial value of the parameter (see Table S5 notes for a full explanation).

The third step is to produce a candidate solution randomly within the neighbourhood of the current solution through the following algorithms (Aarts et al., 2005)

P {E (β)} = \frac{1}{Z (T_{k})} e x p (- \frac{E (β)}{k_{B} T_{k}})

(7)

Z (T_{k}) = \sum_{β ϵ D} e x p (- \frac{E (β)}{k_{B} T_{k}})

(8)

where

E (β)

is an indicator of the land use status (e.g. urban or non-urban);

k_{B}

is an constant that is larger than 0;

T_{k}

is a control parameter at iteration

k

;

P {E (β)}

represents the probability of a cell remaining in the status of

E (β)

Accordingly, the probability of accepting a cell changing its state from the current to the new solution is

p = {\begin{matrix} 1 i f E (β_{n e w}) < E (β_{c u r}) o r Δ E [E (β_{n e w}) - E (β_{c u r})] < 0 \\ e x p (- \frac{Δ E}{T_{k}}) i f E (β_{n e w}) \geq E (β_{c u r}) o r Δ E [E (β_{n e w}) - E (β_{c u r})] \geq 0 \end{matrix}

(9)

where

β_{c u r}

is the current solution;

β_{n e w}

is a new solution as a candidate for land use change.

If $Δ E < 0,$ the candidate solution will be accepted; otherwise, there is a probability of exp $(- \frac{Δ E}{T_{k}})$ to accept the candidate solution. The advantage of GSA is to allow the model to accept any candidate solutions at the early iterations but with the decreasing probability of exp $(- \frac{Δ E}{T_{k}})$ , fewer solutions are accepted at later iterations. Ultimately, only one optimal candidate solution will be accepted when the control parameter $T_{k}$ is approaching zero (Tsallis and Stariolo, 1996).

The final step is to define the initial and final control parameters $T_{0}$ and $T_{k}$ ( $k$ is the number of iterations) that commences and ends the GSA optimisation process, respectively. In our model, the default settings include an initial control parameter $T_{0}$ being assigned a value of 10 and the stopping criteria set to a maximum number of 5000 iterations.

Model implementation

We built two GSA-CA models (Figure 2) representing two different urban development scenarios in peri-urban regions: (1) an urban model prioritising urban development and protecting rural land and (2) a rural model prioritising rural development and restricting the expansion of existing urban.

Figure 2.

Modelling workflow with key methods and procedures.

These two models are inter-connected, using the same set of driving factors (detailed in the section Driving Factors to Land Development in the Supplementary Materials) in the calibration process given the effects of public facilities and transport (e.g. bus and train stations), and topographic factors (e.g. elevation and slope) on urban growth remained relatively stable over time. But they used different constraints in the prediction (detailed in the section Urban and Rural Constraints in the Supplementary Materials) with the urban model constraining rural development and the rural model constraining urban development. We commenced with a systematic sampling method to draw 10,000 sampling cells that were evenly distributed in a full rectangular area covering the MBRC boundaries, and then we select a total of 5295 that fall within the boundary of the study area. This sample size accounts for 0.25% of the total cells in the study area) from the 1991 land use map, and the corresponding cells in the 2011 land use map and the driving factor maps. The systematic sampling method ensures that all cells sampled were evenly distributed in the entire study region; it has been applied in existing CA modelling work to reduce the computational workload (e.g. Feng and Liu, 2013; Feng et al., 2019; Wu, 2002), and the samples were then used to retrieve the CA parameters and build the CA model. The calibration process of the modelling commenced with input of land use data from 1991, nine driving factors and constraints, with each iteration representing 1 year, and stops after 20 iterations to 2011. The GSA-derived CA parameters were incorporated into the transition rules to generate a land transition probability map which was then used in model prediction. The modelling accuracy was assessed by comparing the simulated land use map with the actual land use map in 2011 (detailed in the section Model Assessment in the Supplementary Materials). For model prediction, we used the 2011 land use map as input to define the initial state of the cells, and we took on board the land transition probability map as well as the two sets of constrains (for urban and rural development, respectively) to generate future urban and rural development in three ten-year periods, 2011–2021, 2021–2031 and 2031–2041. Both the urban and rural GSA-CA models were implemented using the UrbanCA software program developed byFeng and Tong (2018), and the modelling outputs were analysed in ArcMap 10.8.

Results

The map of land use transit probability and the evaluation of modelling performance and calibration are presented in the sections Model Assessment and Model Calibration in the Supplementary Materials, indicating a reasonable soundness of the urban and rural CA models we built. We use these two models to predict the urban growth in 2021–2041.

Figure 3 shows the differences of predicted urban development by the urban and rural model, respectively, over different years while Figure 4 shows such differences in each year. Overall, the differences of urban development in 2021, 2031 and 2041 predicted by these two models were not substantial given the low-density in MBR and the low-urbanising rate in Australia (Figure 3). The majority of newly developed urban areas are predicted to distribute around the existing urban pattern in 2011 with several clusters of urban areas in the centre, south and northwest of MBR. When we enlarge into the urban clusters, the subtle changes of urban areas across three ten-year periods are distinctive to see (Figure 3 U1-8 and R1-8). Results from the urban model show that substantial urban growth will appear in the 2011–2021 period (areas in orange colour), and most of the growth will appear in the west and north of Caboolture (Area 2 and 3), west of Morayfield (Area 4), in Burpengary (Areas 5 and 6) and Strathpine (Area 7 and 8). Urban growth in the 2021–2031 period (areas in green) is less noticeable, mainly appearing in the fringe of the urbanised area in 2021 in Bellthorpe, North Lakes and Strathpine (Area 1, 2, 6 and 8). However, urban growth in the 2031–2041 period (areas in purple) becomes more noticeable, extending from the urban patterns in 2031. Table 1 summarises the areas of urban development predicted by the two models. The urban model generates a higher urban increase rate in the 2011–2021 period compared to the rural model (9.19% for the urban model vs. 7.60% for the rural model), but a slightly lower rate in the 2021–2031 period (5.19% vs. 5.66%) and 2031–2041 period (5.83% vs. 6.24%). It reflects the common trend predicted by both models that urban area grows faster in 2011–2021, but such growth slows down afterwards up to 2041. It might be partially explained by the estimation of the reduced urbanising rate in Australia (The World Bank, 2020) and partially due to the increasing effect of the constraints applied in the model prediction on restricting urban growth with the accumulation of yearly iterations.

Figure 3.

Urban patterns from 2011 to 2041 predicted by the urban and rural mode (enlarged maps on side).

Figure 4.

Urban patterns predicted by the urban and rural model in 2021 (A), 2031 (B) and 2041 (B), with enlarged segments (D) and a conceptualised pattern of urban/rural land development (E). Note: The enlarged segments take the urban pattern in 2041 as an example given the urban patterns predicted by the two CA models in other years are similar.

Table 1.

Comparison of urban development by two CA models.

	Urban model (sqkm)			Rural model (sqkm)
	Rural	Urban	Urban increase*	Rural	Urban	Urban increase*
2011	1710.76	171.64	—	1707.73	174.66	—
2021	1694.99	187.40	15.77 (9.19%)	1694.46	187.94	13.28 (7.60%)
2031	1685.28	197.12	9.72 (5.19%)	1683.82	198.57	10.63 (5.66%)
2041	1673.78	208.62	11.50 (5.83%)	1671.43	210.97	12.39 (6.24%)

Note: * the percentage in the bracket is the urban increase rate calculated as the urban increased from Year 1 to Year 2 over the urban area in Year 1

Figure 4 further unveils the urban patterns in 2021, 2031 and 2041, expanding from the edge of the existing urban in 2011 towards rural areas and mainly concentrating in the centre of MBR (Caboolture and Burpengary), in the north (Woodford), in the southwest (Dayboro) and in the south (North Lakes and Strathpine). The enlarged segments of land development predicted by both models in 2041 (Figure 4(d)) show that the new urban predicted by the rural model surrounds that by the urban model. In other words, the urban model predicts the urban development up to 2041 in a more compact way compared to the rural model, possibly due to that the urban model prioritising urban development has more restrict constraints in rural areas to limit urban expansion and protect rural land to some degree. The new urban predicted by the urban model appears in the vacant or agricultural land mixed with the residential clusters internally; while the new urban predicted by the rural model sprawls more obviously, occurring externally on the edge of previous urban land (Figure 4(e)).

Discussion and conclusion

This study simulates and predicts the slow-paced land development in the low-density peri-urban region by adopting an urban and a rural GSA-CA model to represent two regional planning policies (urban-oriented vs. rural-oriented development). By integrating the GSA algorithm into a typical logistic regression-based CA model, our GSA-CA models are capable of optimising the CA transition rules. Through the scenario prediction for urban and rural land development from 2011 to 2041, the two models capture the subtle differences of urban and rural development over time. Our findings provide new insights of peri-urban development to enrich the current urban theories that mainly explain the growth of major urban areas, and offer evidence for policy implications in regional/urban planning and land use management.

Complementing to the limited work of CA modelling in Australia, our study contributes to the current literature with an optimised GSA-CA model. Our study enriches the existing urban theory with a practical application in simulating low-density urban growth in the slow-paced peri-urban areas which has been largely under-explored in the current literature. The urban and rural model, representing urban-oriented and rural-oriented regional planning policies, unveils some common trends in future urban development, that is, the predicted urban growth is faster in 2011–2021 but slows down afterwards up to 2041. This is in alignment with the Australia’s declining urbanisation rate which that has been also observed in other developed countries such as the United States and Canada (The World Bank, 2020). The urbanisation in Australia has entered the saturation phase (Maheshwari and Connellan, 2015), accompanied by the process of suburbanisation during which people chose to live outside the inner-city centres and move towards outer suburbs and low-density peri-urban areas surrounding the metropolitan cities (Healy and Hillman, 2008). As one of the major migrants-receiving countries, Australia is estimated to receive 235,000 overseas migrants by 2031 (Australian Government, Centre for Population, 2021). Population growth has become spatially concentrated in the peri-urban areas as the backyard of major urban centres according to the experience in developed countries (Fisher, 2003); thus, it is reasonable to specular that the increase of migrants in Australia will settle in or drive more urban dwellers towards urban fringe and peri-urban areas where they can enjoy country-living style to realise the Australian dream (Daley et al., 2018). The future urban patterns predicted by our models show that the most significant urban growth will appear in Burpengary, Morayfield, North Lakes and Strathpine along the M1 highway, which provides new insights on the pathway of urban growth in low-density peri-urban regions to guide through future urban configuration and residential development.

Our models and the analytical framework benefit policy making in regional and urban planning as follows. First, the models provide scenario predictions to assess the regional and urban planning strategies with adjustable parameters to balance the urban and rural development. The input of constraints in the models (e.g. the ratio of agriculture/vacant land as the partially conversable land to urban) can be adjusted based on different policies to generate land demand for urban growth as a critical factor that is typically considered by urban planners and developers. In the policy scenario advocating rural land protection, the ratio of the agriculture to vacant land conversion can be reduced to be very small (e.g. to protect most of the agriculture land and vacant land remained as minimum as possible) in that way rural agriculture land will be largely protected and urban invasion to rural will be controlled to some degree. Conversely, in the policy scenario facilitating the rural-to-urban conversion, the parameter of the ratio of agriculture/vacant land can be increased to be reasonably large to accelerate the transition from rural to urban land. Second, the predicted location of new urban growth will help to speculate the distribution of population growth/control in MBR with a total estimated population increase of nearly 100,000 population from 2017 to 2031 according to the regional plan, ShapingSEQ (Queensland Government, 2016). The distribution of population growth will further facilitate the planning of urban infrastructure, open space, public resources and transportation services together with the allocation of more educational and employment opportunities (Bandias and Vemuri, 2005; Newton, 2012). Third, the analytical framework presented in our study can be augmented to a monitoring system for land use management, for example, for evaluating how much agriculture, conservation and vacant land should be preserved as parts of the rural protection strategy (Holmes, 2006), or delineating the locations and spatial patterns of predicted urban growth and future land use change. That is particularly important for making planning regulations, for example, reconciling residential development, the preservation of green space and conservation protection. Moreover, it can guide through the development of urban infrastructures and the configuration of public facilities, for example, delineating areas predicted to have the most significant urban growth where the construction of infrastructures (e.g. water pipelines and street lights) and the delivery of public services (e.g. health and social security services) are most needed. In a sum, our CA modelling procedure provides the projections of future land use patterns to support urban planners and policy makers in resource distribution, infrastructure provision, service configuration and addressing conflicts and dilemmas of land development in the peri-urban areas.

Our study has a number of limitations that set up directions for future research. First, our urban and rural model used a fixed set of driving factors in the calibration and prediction process on the assumption that driving factors influencing urban growth in the past will function in the same way in the future. Further efforts can be put to investigate a way of integrating data-driven methods with knowledge-driven methods and making the CA modelling more realistic by involving dynamic input of driving factors over different phases of prediction. Second, it is worth involving more advanced automatic calibration methods (e.g. geospatial artificial intelligence approaches including machine and deep learning algorithms) to complement the knowledge-based extraction methods, to improve the modelling performance of CA models to be more reliable and realistic. Third, the current CA model can be improved by applying finer resolution datasets to improve the modelling accuracy for more nuanced prediction of urban growth. High-resolution satellite images (e.g. PlanetScope satellite imageries available at the 3-m resolution) would capture more detailed land use patterns which are particularly important for detecting the land change in peri-urban regions slow in urbanisation.

To conclude, this study presents an application of a GSA-CA model to simulate the slow-paced land development. With the combination of optimised CA parameters by the GSA algorithm, various driving and constraining factors, the GSA-CA urban and rural models capture the subtle land use change in a slow urbanisation rate context and generate land development scenarios to assist planning decision and policy making for future land development. Australian cities have been entering into an era of slow urbanisation. Although an increasing share of growth is happening through urban infill and consolidation in the major urban areas, suburbanisation in peri-urban areas is an inevitable trend due to the Australians’ housing and living style. In the face of continuous population growth driven by immigration, the simulating land development in the low-density peri-urban areas is needed. The GSA-CA model in our study meets this need and demonstrates the value of the CA technique in studying urban dynamics. It also provides a tool and a workflow that can be applied to other geographical contexts charactered by high urbanisation levels.

Data availability statement

The following data sources that we used in this paper are publicly available and provided by different governments:

Queensland spatial catalogue – QSpatial: http://qldspatial.information.qld.gov.au/catalogue/custom/search.page?

q = coastal + hazard + ma.

ShapingSEQ: SEQ Regional Plan 2017: https://www.data.qld.gov.au/dataset/south-eastqueensland- regional-plan-2017-shapingseq-series.

Digital cadastral database: https://data.qld.gov.au/dataset/cadastral-data-queensland-series.

Light Detection and Ranging remote sensing imageries: https://data.qld.gov.au/dataset/digital-elevation-model-3-second-queensland.

Coastal hazard technical guide – Determining coastal hazard areas: https://www.qld.gov.au/__data/assets/pdf_file/0025/67462/hazards-guideline.pdf

The software (UrbanCA) is available in the public repository at 10.6084/m9. figshare.14391902.

Supplemental Material

sj-pdf-1-epb-10.1177_23998083211069382 – Supplemental Material for Spatially-explicit prediction of low-density peri-urban development: comparison between urban and rural scenarios in the Moreton Bay Region in South East Queensland, Australia

Supplemental Material, sj-pdf-1-epb-10.1177_23998083211069382 for Spatially-explicit prediction of low-density peri-urban development: comparison between urban and rural scenarios in the Moreton Bay Region in South East Queensland, Australia by Siqin Wang, Yan Liu, Yongjiu Feng and Zhenkun Lei in Journal of Environment and Planning B: Urban Analytics and City Science

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The study is funded by an Australian Research Council Discovery Grant [DP170104235].

ORCID iDs

Siqin Wang

Yan Liu

Supplemental material

Supplemental material for this article is available online.

References

Aarts

Korst

Michiels

(2005) Simulated annealing. In: Burk

Kandall

(eds), Search Methodologies: Introductory Tutorials in Optimization and Decision Support Techniques. New York: Springer, 187–210.

Australian Bureau of Statistics (2016a) 3218.0 - Regional Population Growth, Australia, 2017-18. Available at: https://www.abs.gov.au/ausstats/abs@.nsf/mf/3218.0 (Accessed 18 November 2019).

Australian Bureau of Statistics (2021) Regional population. Available at: https://www.abs.gov.au/statistics/people/population/regional-population/latest-release.

Australian Government, Centre for Population (2021) Population statement, national projections. Available at: https://population.gov.au/data-and-forecasts/data-and-forecasts-dashboard-statement-national.html (Accessed 18 August 2021).

Aburas

Ramli

, et al. (2016) The simulation and prediction of spatio-temporal urban growth trends using cellular automata models: a review. International Journal of Applied Earth Observation and Geoinformation 52: 380–389.

Al-Ahmadi

See

Heppenstall

, et al. (2009) Calibration of a fuzzy cellular automata model of urban dynamics in Saudi Arabia. Ecological Complexity 6: 80–101.

Bandias

Vemuri

(2005) Telecommunications infrastructure facilitating sustainable development of rural and remote communities in Northern Australia. Telecommunications Policy 29(2–3): 237–249.

Batista

Prati

Monard

(2004) A study of the behavior of several methods for balancing machine learning training data. ACM SIGKDD Explorations Newsletter 6(1): 20–29.

Baskent

Jordan

(2002) Forest landscape management modeling using simulated annealing. Forest Ecology and Management 165(1–3): 29–45.

10.

Batty

(2007) Cities and Complexity: Understanding Cities with Cellular Automata, Agent-Based Models, and Fractals. Cambridge; London: The MIT Press.

11.

Batty

(1997) The computable city. International Planning Studies 2(2): 155–173.

12.

Cao

Tang

Shen

, et al. (2015) A new discovery of transition rules for cellular automata by using cuckoo search algorithm. International Journal of Geographical Information Science 29(5): 806–824.

13.

Clarke

Gaydos

(1998) Loose-coupling a cellular automaton model and GIS: long-term urban growth prediction for San Francisco and Washington/Baltimore. International Journal of Geographical Information Science 12(7): 699–714.

14.

Clarke

Hoppen

Gaydos

(1997) A self-modifying cellular automaton model of historical urbanization in the San Francisco Bay area. Environment and Planning B: Planning and Design 24(2): 247–261.

15.

Dahal

Chow

(2015) Characterization of neighborhood sensitivity of an irregular cellular automata model of urban growth. International Journal of Geographical Information Science 29(3): 475–497.

16.

Daley

Coates

Wiltshire

(2018) Housing Affordability: Re-imagining the Australian Dream. Carlton, Victoria: Grattan Institute.

17.

De Almeida

Batty

Monteiro

AMV

, et al. (2003) Stochastic cellular automata modeling of urban land use dynamics: empirical development and estimation. Computers, Environment and Urban Systems 27(5): 481–509.

18.

Feng

Liu

(2013) A heuristic cellular automata approach for modelling urban land use change based on simulated annealing. International Journal of Geographical Information Science 27(3): 449–466.

19.

Feng

Liu

Tong

(2018) Comparison of metaheuristic cellular automata models: A case study of dynamic land use simulation in the Yangtze River Delta. Computers, Environment and Urban Systems 70: 138–150.

20.

Feng

Tong

(2018) Calibration of cellular automata models using differential evolution to simulate present and future land use. Transactions in GIS 22(2): 582–601.

21.

Feng

Tong

(2020) A new cellular automata framework of urban growth modeling by incorporating statistical and heuristic methods. International Journal of Geographical Information Science 34(1): 74–97.

22.

Feng

Yang

Tong

, et al. (2019) Long-term regional environmental risk assessment and future scenario projection at Ningbo, China coupling the impact of sea level rise. Sustainability 11(6): 1560.

23.

Fisher

(2003) Differentiation of growth processes in the peri-urban region: an Australian case study. Urban Studies 40(3): 551–565.

24.

Foley

DeFries

Asner

, et al. (2005) Global consequences of land use. Science 309(5734): 570–574.

25.

Freeman

Hancock

(2017) Energy and communication infrastructure for disaster resilience in rural and regional Australia. Regional Studies 51(6): 933–944.

26.

Gidey

Dikinya

Sebego

, et al. (2017) Cellular automata and Markov Chain (CA_Markov) model-based predictions of future land use and land cover scenarios (2015-2033) in Raya, northern Ethiopia. Modeling Earth Systems and Environment 3(4): 1245–1262.

27.

Healy

Hillman

(2008) Young families migrating to non‐metropolitan areas: Are they at increased risk of social exclusion? Australian Journal of Social Issues 43(3): 479–497.

28.

Holmes

(2006) Impulses towards a multifunctional transition in rural Australia: gaps in the research agenda. Journal of Rural Studies 22(2): 142–160.

29.

Jantz

Goetz

Shelley

(2004) Using the SLEUTH urban growth model to simulate the impacts of future policy scenarios on urban land use in the Baltimore-Washington metropolitan area. Environment and Planning B: Planning and Design 31(2): 251–271.

30.

Kirkpatrick

Gelatt

Vecchi

(1983) Optimization by simulated annealing. Science 220(4598): 671–680.

31.

Lagarias

(2015) Exploring land use policy scenarios with the use of a cellular automata-based model: urban sprawl containment and sustainable development in Thessaloniki. Geocarto International 30(9): 1033–1051.

32.

Landis

Zhang

(1998) The second generation of the California urban futures model. Part 1: Model logic and theory. Environment and Planning B: Planning and Design 25(5): 657–666.

33.

Liu

(2008) Embedding sustainable development strategies in agent‐based models for use as a planning tool. International Journal of Geographical Information Science 22(1): 21–45.

34.

Tian

Wang

, et al. (2021) Optimal sampling strategy of water quality monitoring at high dynamic lakes: a remote sensing and spatial simulated annealing integrated approach. Science of The Total Environment 777: 146113.

35.

Yeh

AG-O

(2002) Neural-network-based cellular automata for simulating multiple land use changes using GIS. International Journal of Geographical Information Science 16(4): 323–343.

36.

Liang

Liu

, et al. (2018) Delineating multi-scenario urban growth boundaries with a CA-based FLUS model and morphological method. Landscape and Urban Planning 177(2018): 47–63.

37.

Lin

Chang

Teng

(2001) Characterization of soil lead by comparing sequential Gaussian simulation, simulated annealing simulation and kriging methods. Environmental Geology 41(1–2): 189–199.

38.

Liu

(2008) Modelling Urban Development with Geographical Information Systems and Cellular Automata. New York: CRC Press.

39.

Liu

Batty

Wang

, et al. (2019) Modelling urban change with cellular automata: contemporary issues and future research directions. Progress in Human Geography 45(1): 0309132519895305.

40.

Liu

Feng

(2012) A logistic based cellular automata model for continuous urban growth simulation: a case study of the Gold Coast city, Australia. In: Heppenstall

Crooks

See

, et al. (eds), Agent Based Models of Geograhical Systems. Dordrecht: Springer, 643–662.

41.

Liu

Liang

, et al. (2017) A future land use simulation model (FLUS) for simulating multiple land use scenarios by coupling human and natural effects. Landscape and Urban Planning 168(2017): 94–116.

42.

Liu

Shi

, et al. (2010) Simulating land use dynamics under planning policies by integrating artificial immune systems with cellular automata. International Journal of Geographical Information Science 24(5): 783–802.

43.

Maheshwari

Connellan

(2015) Irrigation in peri-urban environments: coping with challenges of urbanisation in Australia. Irrigation Australia: The Official Journal of Irrigation Australia 31(2): 4–5.

44.

Mahiny

Clarke

(2012) Guiding SLEUTH land use/land-cover change modeling using multicriteria evaluation: towards dynamic sustainable land use planning. Environment and Planning B: Planning and Design 39(5): 925–944.

45.

McGranahan

Satterthwaite

(2014) Urbanisation Concepts and Trends. London: International Institute for Environment and Development, Vol. 220. Available at: https://www-jstor-org-443.web.bisu.edu.cn/stable/pdf/resrep01297.pdf?acceptTC=true&coverpage=false&addFooter=false.

46.

Meentemeyer

Tang

Dorning

, et al. (2013) FUTURES: multilevel simulations of emerging urban–rural landscape structure using a stochastic patch-growing algorithm. Annals of the Association of American Geographers 103(4): 785–807.

47.

Moreton Bay Regional Council (2020) Our Region. https://www.moretonbay.qld.gov.au/Council/Our-Region.

48.

Neumann

Vafeidis

Zimmermann

, et al. (2015) Future coastal population growth and exposure to sea-level rise and coastal flooding-a global assessment. PloS one 10(3): e0118571.

49.

Newton

(2012) Liveable and sustainable? Socio-technical challenges for twenty-first-century cities. Journal of Urban Technology 19(1): 81–102.

50.

Northam

(1979) Urban Geography. Second edition. New York: John Wiley.

51.

Park

(2017) Digital inequalities in rural Australia: a double jeopardy of remoteness and social exclusion. Journal of Rural Studies 54(2017): 399–407.

52.

Pontius

Jr Chen

(2006) GEOMOD Modeling. Clark University. Available at: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.89.4466&rep=rep1&type=pdf (Accessed 18 November 2019).

53.

Queensland Government (2013) Department of Environment and Heritage Protection, 2013, Coastal Hazard Technical Guide — Determining Coastal Hazard Areas. Available at: https://www.qld.gov.au/__data/assets/pdf_file/0025/67462/hazards-guideline.pdf.

54.

Queensland Government (2016) Projected Population, by Local Government Area, Queensland, 2016 to 2041. Available at: http://www.qgso.qld.gov.au/subjects/demography/population-projections/tables/proj-pop-lga-qld/index.php (Accessed 18 November 2019).

55.

Queensland Government (2019) Moreton Bay Marine Park. Available at: https://parks.des.qld.gov.au/parks/moreton-bay/ (Accessed 18 November 2019).

56.

Queensland Government (2020) Population Estimates. Available at: https://www.qgso.qld.gov.au/statistics/theme/population/population-estimates/regions (Accessed 18 August 2021).

57.

Sang

Zhang

Yang

, et al. (2011) Simulation of land use spatial pattern of towns and villages based on CA–Markov model. Mathematical and Computer Modelling 54(3–4): 938–943.

58.

Santé-Riveira

Boullón-Magán

Crecente-Maseda

, et al. (2008) Algorithm based on simulated annealing for land use allocation. Computers & Geosciences 34(3): 259–268.

59.

The World Bank (2020) Urban Population. Available at: https://data.worldbank.org/indicator/SP.URB.TOTL.IN.ZS.

60.

Tong

Feng

(2019) A review of assessment methods for cellular automata models of land use change and urban growth. International Journal of Geographical Information Science 2: 1–33.

61.

Tsallis

Stariolo

(1996) Generalized simulated annealing. Physica A: Statistical Mechanics and Its Applications 233(1–2): 395–406.

62.

Van Laarhoven

PJM

Aarts

EHL

(1987) Simulated annealing. In: Simulated Annealing: Theory and Applications. Dordrecht: Springer, 7–15.

63.

Verburg

Soepboer

Veldkamp

, et al. (2002) Modelling the spatial dynamics of regional land use: the CLUE-S model. Environmental Management 30(3): 391–405.

64.

Wahyudi

Liu

Corcoran

(2019) Combining landsat and landscape metrics to analyse large-scale urban land cover change: a case study in the Jakarta metropolitan area. Journal of Spatial Science 64(3): 515–534.

65.

(2002) Calibration of stochastic cellular automata: the application to rural-urban land conversions. International Journal of Geographical Information Science 16(8): 795–818.

66.

Clarke

, et al. (2019) Examining the sensitivity of spatial scale in cellular automata Markov chain simulation of land use change. International Journal of Geographical Information Science 33(5): 1040–1061.

67.

Yang

Shi

(2008) Cellular automata for simulating land use changes based on support vector machines. Computers & Geosciences 34(6): 592–602.

68.

Yang

Tang

Cao

, et al. (2013) An intelligent method to discover transition rules for cellular automata using bee colony optimisation. International Journal of Geographical Information Science 27(10): 1849–1864.

69.

Yang

Zheng

X-Q

L-N

(2012) A spatiotemporal model of land use change based on ant colony optimization, Markov chain and cellular automata. Ecological Modelling 233(2012): 11–19.

70.

Zhang

Gensler

Garcia

(2011) A study of the diffusion of alternative fuel vehicles: an agent‐based modeling approach. Journal of Product Innovation Management 28(2): 152–168.

71.

Zhao

Yuan

Zhang

(2016) A study of the system dynamics coupling model of the driving factors for multi-scale land use change. Environmental Earth Sciences 75(6): 529.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.51 MB