RESEARCH ARTICLE

Open Access

Coupling cellular automata and What If? models for residential expansion simulation: A case study of Southwest Sydney, Australia

Corresponding Author

Yi Lu

[email protected]

orcid.org/0000-0002-2090-9057

City Futures Research Centre, School of Built Environment, University of New South Wales, Sydney, New South Wales, Australia

Correspondence

Yi Lu, City Futures Research Centre, School of Built Environment, University of New South Wales, Sydney, NSW, Australia.

Email: [email protected]

Search for more papers by this author

Shawn Laffan,

Shawn Laffan

orcid.org/0000-0002-5996-0570

Earth and Sustainability Science Research Centre, School of Biological, Earth and Environmental Science, Faculty of Science, University of New South Wales, Sydney, New South Wales, Australia

Search for more papers by this author

Christopher Pettit,

Christopher Pettit

City Futures Research Centre, School of Built Environment, University of New South Wales, Sydney, New South Wales, Australia

Search for more papers by this author

Yi Lu,

Corresponding Author

Yi Lu

[email protected]

orcid.org/0000-0002-2090-9057

City Futures Research Centre, School of Built Environment, University of New South Wales, Sydney, New South Wales, Australia

Correspondence

Yi Lu, City Futures Research Centre, School of Built Environment, University of New South Wales, Sydney, NSW, Australia.

Email: [email protected]

Search for more papers by this author

Shawn Laffan,

Shawn Laffan

orcid.org/0000-0002-5996-0570

Earth and Sustainability Science Research Centre, School of Biological, Earth and Environmental Science, Faculty of Science, University of New South Wales, Sydney, New South Wales, Australia

Search for more papers by this author

Christopher Pettit,

Christopher Pettit

City Futures Research Centre, School of Built Environment, University of New South Wales, Sydney, New South Wales, Australia

Search for more papers by this author

First published: 14 June 2024

https://doi.org/10.1111/tgis.13198

Share a link

Email
Wechat
Bluesky

Abstract

The impact of urban expansion on achieving sustainable development goals (SDGs) has become a significant research topic in the field of geographic information science. In this article, we describe a coupled cellular automata (CA)—-What If? model to explore SDG11 “Sustainable cities and communities.” The model calculates overall residential land use demand based on historical data archives using the What If? planning support system (PSS), and then allocates it using a CA model that incorporates variables related to SDG11.2.1 and 11.7.1. Historical datasets for years 2016 and 2021 from Southwest Sydney, Australia were used to assess model accuracy, after which two residential expansion scenarios (years 2021 and 2026) were generated. Based on the modeling results, the SDG-related spatial variables can improve the overall accuracy of CA sub-models using an XGBoost machine learning training methodology. The simulation results of these scenarios confirm the effectiveness of the coupled CA-What If? model, which has the potential to generate more reliable scenario results than the standalone What If? PSS for modeling urban growth of cities across Australia and internationally.

1 INTRODUCTION

The global population living in urban settlements reached 55% in 2018, and is projected to increase to 68% by mid-century (United Nations, 2018). This rapid urbanization presents both challenges and opportunities for policymakers and urban planners, who need to ensure there is sufficient housing stock and infrastructure facilities to accommodate the increasing population migrating to cities. With the expansion of metropolitan areas, the contrast between escalating demands for construction and the scarcity of available land resources becomes increasingly pronounced. Without robust urban planning instruments, rapid urban expansion can lead to a series of environmental issues, including air pollution (Hien et al., 2020), biodiversity decline (Huang et al., 2018), heat island effects (Zhu et al., 2020), and natural habitat loss (Tang et al., 2021). The increasing urbanization of the world's population has resulted in cities shouldering the responsibility to provide affordable housing resources to accommodate the influx of people (Han et al., 2021; Yates, 2016). However, significant carbon emissions are attributed to urban living, an issue that has raised the need for a “Net Zero” commitment (Hausfather & Moore, 2022). Likewise, many cities are experiencing increased traffic congestion and a declining quality of life for local residents (Truelove & Ruszczyk, 2022). In response to these issues, the United Nations has adopted the seventeen sustainable development goals (SDGs) (United Nations, 2016; Pizzi et al., 2020).

The latest version of the SDGs encompasses 169 specific goals and 231 indicators, establishing the sustainable development concept of “integrating economic development, social progress, and environmental improvement.” They require a holistic approach due to their complex, sometimes mutually reinforcing or conflicting nature, and achieving them demands methodical planning and action considering their interrelationships (Fu et al., 2019; Icsu, 2015). Since the inception of the SDGs, researchers have developed a range of urban simulation models incorporating SDGs-related variables or constraints, reflecting a growing interest in linking urban expansion and development with global sustainability goals (Cao, Tian, et al., 2023; Wang et al., 2021; Zhou et al., 2022). These models leverage geographic information systems (GIS), remote sensing, and machine learning techniques to simulate urban growth patterns, estimate future urban land use changes, and evaluate the implications on the SDGs.

Integrating SDG indicators into urban development strategies is an important approach for guiding cities toward enhanced sustainability and resilience. As urbanization accelerates, it becomes critically important to review our urban planning schemes and realign them to focus on SDGs, addressing a wide array of multifaceted challenges. For example, the conflict between urban expansion and agriculture highlights how economic policies can exacerbate poverty and environmental degradation (addressed in SDG 1) (Acheampong et al., 2018) and food production (addressed in SDG 2) (Barthel et al., 2019). This conflict underscores the need to control urban sprawl using certain rules. Furthermore, gender-responsive urban adaptation strategies (Susan Solomon et al., 2021) are closely related to sustainable urban planning and social–ecological–infrastructural systems of cities. Additionally, sustainable urbanization has been identified as a crucial component in the protection of intellectual property rights (Gao, Zhu, et al., 2022), and in assessing city metabolism (Musango et al., 2020), aligning with the framework of SDGs 9 and 11. In summary, there is a clear necessity for strategic and sustainable urban planning in line with specific SDGs, thereby fostering the development of sustainable and resilient cities at a global scale and address these challenges.

Among all goals, SDG 11 “Make cities and human settlements inclusive, safe, resilient, and sustainable” highlights the importance of solving key challenges of urban sustainability. With the ongoing trend of global urbanization, cities that adopt the SDG 11-related strategies are better positioned to balance a growing population by ensuring equitable access to both resources and infrastructure. Given the significance of SDG 11, specific targets under this goal have been applied in numerous urban modeling and analytical studies. Examples include access to housing (SDG 11.1) (Li, El-Askary, et al., 2020), access to transport (SDG 11.2) (Chen et al., 2019), and urban growth models for SDG 11.3 (Ghazaryan et al., 2021; Mithun et al., 2022). Key aspects of SDG 11.5, such as earthquakes (Takagi & Wada, 2019), flooding (Echendu, 2020), fire (Wei et al., 2021), and urban heat island changes (Meftahi et al., 2022; Zhu et al., 2020), have also been reported.

Cellular automata (CA) modeling is a well-established approach for modeling urban expansion and development (He et al., 2006; Lu et al., 2022a; Xu et al., 2019; Yang et al., 2023; Zhai et al., 2020). A classical CA model utilizes a lattice of equally sized cells, along with a set of attribute states to represent geographic features across urban spaces (Batty et al., 1999). By using state transition rules, the CA model can model changes in cell attributes at a micro-level, thereby collectively modeling the dynamic spatial–temporal evolution of the research area (Batty, 2009; Chen et al., 2014). This approach reflects the core concept of complexity science in that complex systems arise from the interaction of simple subsystems (Li, Yeh, et al., 2020). Many aspects of CA models have been explored, including cell forms (Liang et al., 2021; Lu et al., 2015; Yang et al., 2023; Zhu et al., 2021), the discovery of transition rules (Cao et al., 2019; Ding et al., 2022; Momeni & Antipova, 2020), delimitation of neighbourhoods (Barreira-González & Barros, 2017; Zhai et al., 2021), as well as constraint and parameter sensitivity (Li et al., 2021; Wu et al., 2019; Yang et al., 2022).

Some researchers have experimented with integrating SDG indicators and CA modeling. For instance, the CA-Markov model, a commonly used modeling framework, integrates economy (SDG2.3.1 and SDG8.1.1), social (SDG3.c.1, SDG4.1.2, SDG5.b.1, SDG9.c.1, SDG 9.1.2, SDG11.2.1, and SDG11.7.1) and environmental (SDG6.3.1 and SDG11.6.2) related SDG indicators as variables for land use prediction, spatial allocation, and as evaluation metrics (SDG11.3.1) for land use efficiency. Operating through macro-level Markov processes and micro-level CA-based simulation, these models have been applied in the Niger Delta region (Musa et al., 2019), Yangtze River Delta region (Cao et al., 2022), and Tianjin metropolitan area (Lu, Qureshi, et al., 2022). These coupled SDG-Markov-CA models have been demonstrated to provide a comprehensive framework for achieving specific targets in simulating urban expansion and development processes. However, the integration of SDG indicators with the CA models is still underexplored, as evidenced by the limited number of case studies.

CA models enable an understanding of urban expansion and simulation of development. Nevertheless, accurate projections of urban land patterns also require established scenarios that represent possible future socioeconomic and environmental characteristics (Chen et al., 2020; Debnath, Pettit, Soundararaj, et al., 2023). The What If? planning support system (PSS) is a bottom-up model, featuring standalone suitability evaluation, demand projection, and spatial allocation functions. It is a scenario-based tool that uses GIS data for land suitability analysis, land use projection, and evaluation of policy impacts on urban development (Klosterman, 1999, 2011; Pettit et al., 2015). In this research, a CA model is integrated with the What If? PSS to identify future urban development patterns more comprehensively and reliably across a set of scenarios.

The article is divided into five sections. Following this introduction, the fundamental structure, components, and methodologies of the model. Subsequently, a detailed case study from the Southwest of the Greater Sydney Region, Australia is given in Section 3. The simulation results and predicted scenarios are then analyzed in Section 4, while Section 5 provides a summary of key findings and outlines future research directions.

2 METHODOLOGY

2.1 The general CA-What If? modeling framework

The general CA-What If? modeling framework can be divided into three key stages (Figure 1). The first stage involves input data, which comprises historical land use change maps and spatial variables, including biophysical, environmental, socioeconomic, and SDG-related indicators. Socioeconomic-related variables, such as historical population, the number of dwellings, vacancy rate, and average household size, are utilized for land use demand projection at the macro level. The biophysical, environmental, and SDG-related spatial variables, associated with land use change maps serve as the driving factors of land use change at the macro level. The processed land use datasets are then randomly selected as separate training and testing samples. In the second stage, the What If? demand sub-model derives past trends in housing supply and population growth, and then projects future land use demand. Simultaneously, key parameters of the CA sub-model are fine-tuned via hyperparameter adjustments employing a decision tree-based regression methodology. This process is pivotal for discovering the transition rules of the CA sub-model. Afterward, the CA sub-model integrates typical spatial variables and SDG-related indicators to assess whether the inclusion of SDG indicators enhances the overall performance of land use allocation. Quantitative evaluation and validation, including the use of producer's spatial accuracy metrics and kappa coefficients, are integral to the model validation process. In the third stage, the optimized CA-What If? framework is then utilized to simulate future land use scenarios, applying various strategies and growth patterns to further analyze the model's outputs. The final outputs are the land use change layouts under each of the potential scenario, and are then further evaluated and visualized for decision-making support.

Details are in the caption following the image — **FIGURE 1**
Open in figure viewer PowerPoint

The diagram of CA-What If? modelling framework.

2.2 Land use demand calculation using What If?

The land use demand component uses the population and employment growth projections defined according to historical census data and estimates the amount of residential land required to accommodate the projected household growth. Future household numbers in the study area can be estimated as (Klosterman, 2008; Pettit et al., 2015):

H_{X 2} = H_{X 1} \times {(1 + R_{h})}^{n},

()

where

H_{X 1}

and

H_{X 2}

are the number of households in current and projected years, respectively; n is time gap between projected and current years; R_h is rate of household growth; R_h is derived from historical census data:

R_{h} = \frac{H_{Y 2} - H_{Y 1}}{Y 2 - Y 1},

()

where

H_{Y 1}

and

H_{Y 2}

are the total number of households in historical years Y₁ and Y₂.

The estimated demand for residential land,

{Demand}_{resi}

, can then be calculated as:

{Demand}_{resi} = \sum_{i} \frac{B_{i} \times (1 - {IR}_{i}) \times (\frac{P_{f}}{(1 - {VR}_{i}) \times {AHS}_{f}}) - {Count}_{i}}{{Den}_{f}},

()

where i is particular type of residential housing, B_i is future breakdown percentage, IR_i is future infill rate, VR_i is future vacancy rate of residential housing i, AHS_f is future average household size, P_f is the predicted future population, Count_i is total number of residential housing i in current year, and Den_f is future density of residential housing i.

2.3 CA model calibration using XGBoost

XGBoost (“Extreme Gradient Boosting”) is an advanced implementation of gradient boosting algorithms. It calibrates a series of decision trees, aggregating their outputs to enhance predictive accuracy and manage overfitting more effectively than a single decision tree (Chen & Guestrin, 2016). It is a widely used machine learning algorithm for supervised learning tasks, with demonstrated high performance and scalability across a range of predictive urban modeling applications (Gao, Shi, et al., 2022; Lin et al., 2022; Qu et al., 2019; Zhao et al., 2021). XGBoost is primarily oriented to supervised learning problems, where input variables x_i are used to predict a response (dependent) variable y_i. The prediction score from each individual tree is then aggregated to obtain the final score, which is assessed using N additive functions to predict the output (Putatunda & Rama, 2018). Specifically, the XGBoost regressor can be described as:

{\hat{y}}_{i} = \sum_{k = 1}^{N} f_{k} (x_{i}), f_{k} \in F,

()

where N is the number of trees, F is the functional space of regression trees and f_k is a function in the functional space.

Subsequently, the CA model allocates projected demand based on the results of a suitability evaluation. The transfer probability of cell_ij is described as:

P_{ij}^{t} = S_{c} \times Ω_{ij}^{t} \times {con}_{ij}^{t} \times Rand,

()

where

P_{ij}^{t}

is transfer probability of Cell_ij, S_c is suitability of its current location (which is derived from a constructed scenario),

Ω_{ij}^{t}

is neighbourhood configuration,

{con}_{ij}^{t}

is whether a cell to be converted is situated within a location where specific constraint is applied, Rand is the stochastic perturbation during the real urban development process.

3 CASE STUDY

3.1 Study area and data processing

Southwest Sydney is located within the Greater Sydney region in Australia, featuring Cabramatta and Liverpool as two of its key urban centers. The bounding polygon used here is the “Sydney–South West” Statistical Area Level 4, as defined by the Australian Statistical Geography Standard (ABS, 2021b), with an area of 540.42 km². In the most recent population census, the region accommodated 155,782 private dwellings (households), and a total population of 474,430 (ABS, 2021a). In terms of population growth, it is one of the fastest-growing regions in Australia, reflecting the continued immigration trend seen throughout the Greater Sydney region, with consistently high demand for residential properties (Lu et al., 2023). The official projection from the NSW Planning Institute indicates that the population of the entire Greater Sydney region is expected to surpass 6.1 million by 2041, an increase of over 1 million people from the current population (NSW Government, 2022). Moreover, the predominant land use change within Southwest Sydney is toward newly developed residential areas (Figure 2). With an increasing population and evident trend in residential development, Southwest Sydney has been identified as an ideal place for the verification of the proposed CA-What If? modeling framework.

On the basis of previous research and data availability, 10 spatial variables are used in this research. These variables have also been converted to a raster format with a spatial resolution of 60 m. They are categorized into three groups (Table 1):

Proximity (Figure 3a–h). Proximity variables measure accessibility and are commonly employed in land use change modeling. Proximity in this study is measured as the Euclidean distances from the Sydney CBD (D_CBD), the town center of Southwest Sydney (D_Cen), shopping centers (D_Shop), public hospitals (D_Hosp), universities (D_Uni), main roads (D_Road), train stations (D_Rail), and parklands (D_Park). The latter two variables serve as localised indicators of SDG 11.2.1 and SDG 11.7.1, relating to public space accessibility for all, inclusive of gender, age, and persons with disabilities.
Slope (Figure 3i). The topographical slope gradient significantly influences the feasibility and cost of building construction. The slope of the study area was calculated using a 1-second Digital Elevation Model (DEM) provided by Geoscience Australia (2011), which has been processed to represent ground surface topography. The highest slope values are in the western and southern parts of the study area, reaching a maximum of 40.6°.
Constraint. Any land use transformation is excluded within listed reserve areas as outlined in Table 1.

TABLE 1. Spatial variables of CA sub-model.

Variable name	Definition	Data source
D _CBD	Distance to the Sydney CBD	OpenStreetMap
D _cen	Distance to the town center of Southwest Sydney	OpenStreetMap
D _Shop	Distance to the nearest shopping center	OpenStreetMap
D _Hosp	Distance to the nearest public hospital	OpenStreetMap
D _Uni	Distance to the Western Sydney University—Liverpool Campus	OpenStreetMap
D _Road	Distance to the nearest main roads	Geoscience Australia
Slope	The slope of the candidate cell	Geoscience Australia
Limit_Reserve	Spatial information on areas reserved under the NP&W Act 1974. Areas include National Parks, Nature Reserves, Regional Parks, State Conservation Areas, Aboriginal Areas, Historic Sites and Karst Conservation Reserves	NSW Government—The NSW National Parks and Wildlife Service (NPWS) Estate database
D_Rail (SDG11.2.1-related)	Distance to the nearest railway stations within Southwest Sydney	Geoscape Australia
D_Park (SDG11.7.1-related)	Distance to the nearest parkland	Australian Bureau of Statistics

Note: All the spatial variables (except for Limit_Reserve) have been normalized to the value range of [0, 1] to exclude the impacts of differentiated units.

In the simulation experiments, all CA sub-models utilized a 3 × 3 Moore Neighbourhood configuration, comprising a total of 100 iterations per experiment. A random disturbance variable (Equation 5) was used to simulate stochastic perturbations.

The data processing workflow comprises several distinct stages. First, land use categories are derived from the ABS Mesh Block polygons for Southwest Sydney between 2016 and 2021, and then converted to raster with a resolution of 60 × 60 m. Cells that changed the category to “Residential” land use between these two time periods were identified, with most changed cells initially being “Primary Production,” “Parkland,” and “Other.” The required variables for calculating the overall demand (Equations 1-3) were then input into the ‘What If?’ sub-model to estimate residential demand for 2026. Subsequently, the spatial variables are then spatially joined with the candidate raster cells, namely those categorized as “Primary Production,” “Parkland,” and “Other.” The values of these spatial fields are either the Euclidean distance from a candidate cell to the nearest POIs and FOIs, the slope of candidate cells, or whether a candidate cell is situated within the reserve areas (Limit_Reserve). Given the dataset size (66,356 records) and the incorporation cross-validation during the hyperparameter tuning process, the entire dataset has been randomly divided into training (30%) and testing (70%) samples. These samples were used for hyperparameter tuning to optimize the performance of the XGBoost-based CA sub-model. Upon deriving the optimal parameter combination, four groups of XGBoost-CA sub-models with different sets of spatial variables were developed to assess the impact of SDG-related variables in spatial allocation.

3.2 The land use demand of Southwest Sydney

Future residential-related land use demand was modeled using historical census data (ABS, 2021a). The population, number of dwellings, vacancy rates, and average household sizes were derived directly from the ABS census data for the years 2016 and 2021 (ABS, 2016, 2021a). The density of residential housing was estimated by dividing the number of dwellings by the total area of the Sydney Southwest polygons (Table 2).

TABLE 2. Input variables of What If? sub-model.

Year	Population	Number of dwellings	Vacancy rate (%)	Average household size (persons/household)	Density of residential housing (houses/km²)
2016 (actual)	405,962	122,954	4.9	3.3	239.75
2021 (actual)	474,430	148,543	5.6	3.2	288.34
2026 (predicted)	503,607	Not required	4.9	3.3	295.10

The projected population of the entire Greater Sydney region at the commencement of years 2021 and 2026 are 5,259,800 and 5,583,600, respectively (Australian Government, 2023). Given this growth rate of 6.15%, the predicted population of Southwest Sydney is 503,607 and the predicted density of residential housing is 306.07.

Additionally, to mitigate the impact of the abnormal housing vacancy rate in 2021, influenced by the global COVID-19 pandemic, both the vacancy rate and average household sizes for the year 2026 are set to match their 2016 values (Evans et al., 2020; Li et al., 2022). In the absence of official estimates, the breakdown and infill rates of both years are set as 1 and 0, separately. With the completion of all these settings, the ratio of land use demand between 2016–2021 and 2021–2026 can be represented as:

\begin{matrix} \frac{{Dem}_{16 - 21}}{{Dem}_{21 - 26}} & = \frac{\frac{{Pf}_{2021}}{(1 - {VR}_{2016}) \times {AHS}_{2021}} - {Count}_{2016}}{{Den}_{2021}} / \frac{\frac{{Pf}_{2026}}{(1 - {VR}_{2021}) \times {AHS}_{2026}} - {Count}_{2021}}{{Den}_{2026}} \\ = \frac{\frac{474430}{(1 - 4.9 %) \times 3.2} - 122954}{288.34} / \frac{\frac{503607}{(1 - 5.6 %) \times 3.3} - 148543}{295.10} = \frac{114.26}{44.45} \end{matrix}

There is an increase of 4331 residential cells between 2016 and 2021 (Table 3), corresponding to 15.59 km². Therefore, the predicted residential land use demand from the years 2021 to 2026 can be calculated as:

{Dem}_{21 - 26} = 15.59 {km}^{2} \times \frac{44.45}{114.26} = 6.04 {km}^{2}

TABLE 3. Land use statistics, years 2016 and 2021.

Land use category	Year 2016		Year 2021
Land use category	Area (km²)	Fraction (%)	Area (km²)	Fraction (%)
Commercial	3.70	0.68	4.99	0.92
Education	5.33	0.99	6.04	1.12
Hospital/medical	0.30	0.06	0.30	0.06
Industrial	18.39	3.40	18.36	3.40
Other	13.20	2.44	10.18	1.88
Parkland	66.89	12.38	63.89	11.82
Primary Production	158.79	29.38	144.33	26.71
Residential	272.38	50.40	287.97	53.29
Transport	0.70	0.13	3.61	0.67
Water	0.74	0.14	0.74	0.13
Total	540.42	100.00	540.42	100.00

3.3 Hyperparameter tuning

The parameters of XGBoost algorithm were refined using random search coupled with a stratified threefold cross-validation approach (‘GridSearchCV’ function in the scikit-learn Python library, version 1.4), for hyperparameter tuning purpose (Pedregosa et al., 2011). Hyperparameter tuning was conducted to enhance the XGBoost regressor's capability in predicting whether a candidate cell has been transferred to “Residential” at the end of the simulation period. Explanatory variables were “D_CBD,” “D_Cen,” “D_Shop,” “D_Hosp,” “D_Uni,” “D_Road,” “D_Rail,” “D_Park,” “Slope,” (Table 1) and “Neighbourhood”. The hyperparameters identified as most effective, as detailed in Table 4, were then used to configure an enhanced XGBoost regressor model. [Correction added on 23 July 2024, after first online publication: The Explanatory variable `Neighbourhood’ added after first online publication.]

TABLE 4. Hyperparameters, tested value ranges and result.

Hyperparameter	Definition	Values tested	Optimal parameter
n_estimator	Number of gradient-boosted trees	100, 200, 300	300
learning_rate	Learning rate	0.01, 0.1, 0.2	0.1
max_depth	Maximum depth of a tree	3, 4, 5	5
min_child_weight	Minimum sum of instance weight needed in a child	1, 2, 3	1
subsample	Subsample ratio of the training instances	0.6, 0.8, 1.0	0.8
colsample_bytree	The fraction of features to be randomly sampled for each tree	0.6, 0.8, 1.0	0.8

3.4 XGBoost-based CA sub-model training and testing

After hyperparameter tuning, four groups of spatial variables were applied to CA sub-models 1–4 for training purposes. CA sub-model 1 utilized all variables, including those related to both SDG11.2.1 (D_Rail) and SDG11.7.1 (D_Park). CA sub-models 2 and 3 each utilized all variables except for D_Rail and D_Park, respectively. CA sub-model 4 was trained exclusively with non-SDG-related variables. Additionally, each CA sub-model was trained and tested through 10 independent simulations, and the simulation process for each single CA sub-model consists of 100 iterations. Each simulation iteration can be simplified into three steps: (1) Calculate the transfer probabilities of all candidate cells; (2) Select cells with higher transfer potentials for conversion based on their overall probabilities in the current iteration; (3) Update the selected candidate cells to residential cells and update the spatial layers accordingly. After 100 iterations, the simulated distribution of residential and nonresidential cells for the year 2021 is produced for each of the CA sub-models. Finally, the 2016 testing sample is utilized as the reference for model evaluation.

An evaluation of four CA sub-models, highlighting the substantial influence of incorporating SDG-related variables (specifically D_Rail and D_Park) is given in Table 5. Specifically, producer's spatial accuracy and kappa coefficients are being applied as the metrics for CA sub-model evaluation. Sub-model 1, which includes all variables, demonstrates superior accuracy, achieving producer's spatial accuracy scores ranging from 96.50% to 97.52%, with a mean of 97.14%. Its average kappa coefficient of 0.967 also exceeds that of the other groups. This enhanced spatial accuracy and consistency underscores the effectiveness of integrating SDG-related variables. [Correction added on 23 July 2024, after first online publication: The spacial accuracy scores 97.14% corrected to 96.50% and 96.50% corrected to 97.14%.]

TABLE 5. Overall accuracies of four types of CA sub-models.

	Variable	Max (%)	Min (%)	Mean (%)	Max	Min	Mean
	Variable	Producer's spatial accuracy			Kappa coefficient
Sub-model 1	D_CBD, D_cen, D_Shop, D_Hosp, D_Uni, D_Road, D_Rail, D_Park, Slope, Neighbourhood	97.52	96.50	97.14	0.971	0.959	0.967
Sub-model 2	D_CBD, D_cen, D_Shop, D_Hosp, D_Uni, D_Road, D_Rail, Slope, Neighbourhood	97.13	96.24	96.59	0.967	0.956	0.960
Sub-model 3	D_CBD, D_cen, D_Shop, D_Hosp, D_Uni, D_Road, D_Park, Slope, Neighbourhood	97.46	96.37	96.87	0.971	0.958	0.964
Sub-model 4	D_CBD, D_cen, D_Shop, D_Hosp, D_Uni, D_Road, Slope, Neighbourhood	96.31	92.87	95.49	0.957	0.917	0.948

The D_Uni variable appears to be the most significant variable across all sub-models (Table 6). The SDG11.7.1-related variable, D_Park, is also important, particularly in sub-models 1 and 3 for which it records values of 14.0% and 16.2%, ranking as the second highest in feature importance among all variables. The D_CBD variable is identified as the third-highest in feature importance across all models except for sub-model 2. Additionally, the SDG 11.2.1-related variable D_Rail demonstrates feature importance values of 7.8% and 11.0%, ranking 5th and 4th highest in sub-models 1 and 2, respectively. These findings affirm the pivotal impact of specific SDG-related variables on the simulation accuracy of the model. [Correction added on 23 July 2024, after first online publication: The sentence `The DCBD variable is identified as the third-highest in feature importance across all models’ corrected to `The DCBD variable is identified as the third-highest in feature importance across all models except for sub-model 2’]

TABLE 6. Feature importance of all spatial variables in different sub-models.

Variable name	Average feature importance (%)
Variable name	Sub-model 1	Sub-model 2	Sub-model 3	Sub-model 4
D _CBD	12.4 (3)	13.6 (2)	12.5 (3)	13.2 (3)
D _cen	6.5 (6)	8.2 (5)	8.8 (5)	13.2 (3)
D _Shop	5.9 (8)	7.8 (6)	7.6 (6)	8.0 (5)
D _Hosp	8.6 (4)	11.8 (3)	9.9 (4)	16.6 (2)
D _Uni	33.1 (1)	34.2 (1)	33.3 (1)	31.0 (1)
D _Road	6.4 (7)	7.0 (7)	5.9 (7)	6.8 (7)
D_Rail (SDG11.2.1 related)	7.8 (5)	11.0 (4)	n.a.	n.a.
D_Park (SDG11.7.1 related)	14.0 (2)	n.a.	16.2 (2)	n.a.
Slope	1.2 (10)	3.2 (8)	2.6 (9)	5.2 (8)
Neighbourhood	4.0 (9)	3.1 (9)	3.2 (8)	6.9 (6)

Note: The average feature importance is calculated based on the mean value of 10 separate operations of each type of CA sub-model, the numbers in the brackets indicate the relative importance of these features in sub-models.

3.5 Scenario planning outcomes using a CA-What If? model

The importance of SDG-related variables in CA sub-modeling has been shown in the preceding section. Consequently, CA sub-model 1, which uses all spatial variables listed in Table 6, is selected for allocating the overall land use demand from the What If? sub-model.

Scenario planning was initially introduced by Royal Dutch/Shell in the late 1960s to early 1970s for generating and evaluating strategic options (Wack, 1985). As awareness of urban growth and sustainable development grew, scenario planning began to be applied for forecasting and analyzing urban land use changes (Chakraborty & McMillan, 2015; Pettit et al., 2020; Wang et al., 2022). Figure 4 illustrates the spatial distribution of newly developed residential land under two scenarios: “Business as Usual” and “Sustainable growth.” Additionally, Table 7 shows the proportions of newly transformed residential cells in every SA3. In the “Business as usual” scenario, the transformation rules and the types of land eligible for a from 2016 to 2021. Here, newly added residential cells for 2021–2026 area selected from the “Primary Production,” “Parkland” and “Other” categories. In contrast, the “Sustainable growth” scenario, while maintaining the identical transition rules for land use demand allocation, reduces the types of available categories to “Primary Production” and “Other.”

TABLE 7. Proportion of newly transformed residential cells in every SA3.

SA3 name	Scenario 1. Business as usual		Scenario 2. Sustainable growth
SA3 name	Number of new residential cells	Proportion (%)	Number of new residential cells	Proportion (%)
Bringelly—Green Valley	1015	60.49	1260	75.09
Fairfield	162	9.65	257	15.32
Liverpool	501	29.86	161	9.59

To evaluate the accuracy of future land use predictions, three spatial layers are used: biodiversity value, bushfire-prone areas, and proposed future residential growth areas (Table 8). The actual future land use is obviously unknown, making a calculation of the overall accuracy or Figure of Merit (FoM) in the traditional sense infeasible.

TABLE 8. Evaluation of spatial layers of future scenario modeling outcomes.

Name	Description	Data source and year	Link
Biodiversity Values Map	The Biodiversity Values Map (BV Map) identifies land with high biodiversity value that is particularly sensitive to impacts from development and clearing. The BV Map is one of the triggers for determining whether the Biodiversity Offset Scheme (BOS) applies to a clearing or development proposal	NSW Government (2018)	https://datasets.seed.nsw.gov.au/dataset/biodiversity-values-map
NSW Bushfire-Prone Land	Bushfire-Prone Land is mapped within a local government area, which becomes the trigger for planning for bushfire protection. Bushfire-Prone Land mapping is intended to designate areas of the State that are considered to be higher bushfire risk for development control purposes	NSW Government (2020)	https://datasets.seed.nsw.gov.au/dataset/bush-fire-prone-land
Growth centers	The proposed areas of growth centers outlined in in State Environmental Planning Policy (Precincts–Western Parkland City)	NSW Government (2021)	https://prod.planning-nsw.links.com.au/opendata/dataset/state-environmental-planning-policy-precincts-western-parkland-city-2021

4 DISCUSSION

Between 2016 and 2021, the overall proportion of “Residential” land in the study area increased from 50.40% to 53.29%, with an associated decrease in “Primary Production,” “Parkland,” and “Other” categories. This reflects a general trend of urban residential expansion in order to satisfy population growth and related housing demand in the entire Greater Sydney region over the past decades. It corresponds to the latest version of the Greater Sydney region Plan (NSW Government, 2018), which anticipates an increased demand and preference for housing to meet the needs of evolving communities. Thus, it can be inferred that the conversion of nonresidential to residential land is likely to remain the dominant land use trend in the Greater Sydney region for the upcoming future.

To explore and simulate the spatial distribution of prospective land use change in the uncertain future, this article proposes a coupled CA-What If? modeling framework which simulates urban residential expansion under various scenarios. The What If? sub-model predicts the overall land demand in the study area between 2021 and 2026 at a macro-level scale, drawing on the historical Australian national census data (years 2016 and 2021), along with the manual setting of the vacancy rate, average household size and density of residential housing in the year 2026 by taking past trend as reference. It is suggested by this forecast that from 2021 to 2026, an additional of 6.04 km² residential land will be required in Southwest Sydney, in comparison with the previous 15.59 km² change of residential land from 2016 to 2021. This trend suggests changes in the compact city form and increasing densification of Southwest Sydney, characterized by smaller land parcels for single detached housing alongside a rise in apartments and higher density developments (Easthope et al., 2022; Kleeman et al., 2022).

Regarding our CA sub-model, the effectiveness of two SDG-related spatial variables is validated in terms of their impacts on the spatial allocation accuracy. Afterward, the CA sub-model with selected spatial variables, which generate the most accurate outcome, is used for future scenario planning. Different combinations of spatial variables affect the CA sub-models' accuracy (Table 5), even in a relatively small area like Southwest Sydney. Incorporating two SDG-related variables resulted in an average producer's accuracy of 97.14% from 10 independent simulations, higher than when only a single SDG factor was considered (96.59% and 96.87%). The sub-model without SDG factors had the lowest spatial allocation accuracy, ranging between 92.87% and 96.31%. Furthermore, the importance of the SDG-related variables was also evident in the spatial rule extraction results based on the XGBoost method. The spatial variable D_Park (Related to SDG 10.7.1) ranked second in feature importance in both sub-models 1 and 3, while D_Rail (Related to SDG 11.2.1) ranked fifth and fourth in sub-models 1 and 2, respectively (Table 6). These findings underscore the significance of SDG spatial variable in the overall accuracy of CA sub-models.

After fine-tuning and validation, the CA sub-model is then applied for the allocation of the overall land use demand for Southwest Sydney, with 100-iteratons in each of the proposed simulation experiments. Scenario 2 “Sustainable growth” better integrates natural risk management with urban development planning than Scenario 1 “Business as usual” (Figure 5). Specifically, there is a total of 5.61 km² newly transformed residential cells outside biodiversity value zones in Scenario 2, compared with 4.83 km² cells in Scenario 1. This enhancement signifies a more ecologically considerate approach to urban residential expansion, aiming to minimize impacts on biodiversity value. In addition, there is also a slight rise in the number of cells within proposed growth centers in Scenario 2 (4.41 km²) compared with Scenario 1 (4.18 km²), reflecting a targeted commitment to development within planned growth boundaries. Moreover, the identical figures for newly transformed cells outside bushfire-prone land (5.97 km² in both Scenarios 1 and 2) demonstrate a consistent emphasis on preventing bushfire risks while balancing developmental ambitions. These evaluation standards from Figure 5 prove that Scenario 2 has advanced the reconciliation of State Environmental Planning Policy (Precincts—Western Parkland City) in our case study. It is also concluded that in comparison with large-scale urban land expansion scenarios (Chen et al., 2020), even within a relatively short simulation period (5 years) and a relatively small study area (540.42 km²), small adjustment of land use conversion rules in proposed scenarios can lead to significant differences in the simulation results of the study area. These findings are also consistent with previous CA model or What If?-based scenario planning (Daniel & Pettit, 2022; Debnath, Pettit, & Leao, 2023; Feng et al., 2019; Liang et al., 2018).

5 CONCLUSIONS

This research represents an exploration of integrating CA and What If? models, where the What If? sub-model is utilized for land use demand prediction, and the CA sub-model is used to allocate overall demand to specific raster cells. Specifically, the What If? sub-model predicts that the conversion from nonresidential to residential land will continue as the primary trend in urban development in Southwest Sydney. Less newly developed residential land is expected between 2021 and 2026 (6.04 km²) compared with the period 2016–2021 (15.59 km²). In comparison with its initial version (Lu et al., 2022b), the CA sub-model in this study is calibrated using the XGBoost machine learning algorithm, is capable of discerning complex and nonlinear landscape change patterns in this region, a finding echoed by other researchers in similar studies in Khulna city, Bangladesh (Islam et al., 2021), Yancheng City, China (Hao et al., 2022), Seoul, Korea (Kim et al., 2023), separately. In conclusion, the modeling outcome reveals that Scenario 2 ‘Sustainable Growth’ is more effective in balancing residential expansion needs with reduced bushfire risk compared with Scenario 1 “Business as Usual.” Furthermore, it aligns more closely with the growth centers proposed by the NSW Department of Planning.

Overall, the coupled CA-What If? model is not only capable of capturing the regulations of historical urban residential expansion and SDG-related indicators, but also to predicting the future residential land use demands at a macro-level, and then allocating these demands at a micro-level. However, there is still potential for further improvement in its framework. For instance, further categorizing residential land into types, such as low, medium, and high-density, corresponding to diverse housing types, could enhance the framework's realism, particularly with regard to cell types in the CA sub-model. Furthermore, the What If? sub-model could consider both constraints on land transformation and land use change priority in specific areas, as demonstrated for a different region by Pettit et al. (2015). It could also incorporate the synergies and trade-offs among complex SDG indicators, which is crucial to achieving long-term sustainability goals (Cao, Chen, et al., 2023; Hegre et al., 2020; Kuc-Czarnecka et al., 2023). Finally, validating the coupled CA-What If? model's applicability in larger metropolitan areas, such as the entire Greater Sydney or other metropolitan regions would be the next step in testing the generality of the model. This would assess the impact of spatial heterogeneity rules on the model's performance across different subregions in a further step.

ACKNOWLEDGMENTS

This research was enabled through the Australia Research Data Commons (ARDC) and Australian Urban Research Infrastructure Network (AURIN) funded – Australian Housing Data Analytics Platform (RG203395). The authors are also grateful for the data provided by the Australian Bureau of Statistics (ABS) and OpenStreetMap (OSM). Open access publishing facilitated by University of New South Wales, as part of the Wiley - University of New South Wales agreement via the Council of Australian University Librarians.

CONFLICT OF INTEREST STATEMENT

The authors declare no potential conflicts of interest with respect to the research, authorship, and publication of this paper.

Open Research

DATA AVAILABILITY STATEMENT

The ABS Mesh Block polygon dataset used in this study is available at https://www.abs.gov.au/statistics/standards/australian-statistical-geography-standard-asgs-edition-3/jul2021-jun2026/access-and-downloads/digital-boundary-files.

REFERENCES

ABS. (2016). 2016 census all persons QuickStats: Sydney—South west. Australian Bureau of Statistics. https://www.abs.gov.au/census/find-census-data/quickstats/2016/127
Google Scholar
ABS. (2021a). Census all persons QuickStats: Sydney—South west. Australian Bureau of Statistics. https://www.abs.gov.au/census/find-census-data/quickstats/2021/127
Google Scholar
ABS. (2021b). Australian statistical geography standard (ASGS) ( 3rd ed.). Australian Bureau of Statistics. https://www.abs.gov.au/statistics/standards/australian-statistical-geography-standard-asgs-edition-3/jul2021-jun2026
Google Scholar
Acheampong, M., Yu, Q., Enomah, L. D., Anchang, J., & Eduful, M. (2018). Land use/cover change in Ghana's oil city: Assessing the impact of neoliberal economic policies and implications for sustainable development goal number one–a remote sensing and GIS approach. Land Use Policy, 73, 373–384. https://doi.org/10.1016/j.landusepol.2018.02.019
10.1016/j.landusepol.2018.02.019
Web of Science® Google Scholar
Australian Government. (2023). Capital city and rest-of-state projections, 2021–22 to 2032–33. https://population.gov.au/data-and-forecasts/projections/capital-city-and-rest-state-projections-2021-22-2032-33
Google Scholar
Barreira-González, P., & Barros, J. (2017). Configuring the neighbourhood effect in irregular cellular automata based models. International Journal of Geographical Information Science, 31(3), 617–636. https://doi.org/10.1080/13658816.2016.1219035
10.1080/13658816.2016.1219035
Web of Science® Google Scholar
Barthel, S., Isendahl, C., Vis, B. N., Drescher, A., Evans, D. L., & van Timmeren, A. (2019). Global urbanization and food production in direct competition for land: Leverage places to mitigate impacts on SDG2 and on the earth system. The Anthropocene Review, 6(1–2), 71–97. https://doi.org/10.1177/2053019619856672
10.1177/2053019619856672
Web of Science® Google Scholar
Batty, M. (2009). Cities as Complex Systems: Scaling, Interaction, Networks, Dynamics and Urban Morphologies. In R Meyers (Ed.), Encyclopedia of Complexity and Systems Science. Springer. https://doi.org/10.1007/978-0-387-30440-3_69
10.1007/978-0-387-30440-3_69
Google Scholar
Batty, M., Xie, Y., & Sun, Z. (1999). Modeling urban dynamics through GIS-based cellular automata. Computers, Environment and Urban Systems, 23(3), 205–233. https://doi.org/10.1016/S0198-9715(99)00015-0
10.1016/S0198-9715(99)00015-0
Google Scholar
Cao, M., Chang, L., Ma, S., Zhao, Z., Wu, K., Hu, X., Gu, Q., Lü, G., & Chen, M. (2022). Multi-scenario simulation of land use for sustainable development goals. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 15, 2119–2127. https://doi.org/10.1109/JSTARS.2022.3152904
10.1109/JSTARS.2022.3152904
Web of Science® Google Scholar
Cao, M., Chen, M., Zhang, J., Pradhan, P., Guo, H., Fu, B., Li, Y., Bai, Y., Chang, L., & Chen, Y. (2023). Spatio-temporal changes in the causal interactions among sustainable development goals in China. Humanities and Social Sciences Communications, 10(1), 1–9. https://doi.org/10.1057/s41599-023-01952-z
10.1057/s41599-023-01952-z
CAS Google Scholar
Cao, M., Huang, M., Xu, R., Lü, G., & Chen, M. (2019). A grey wolf optimizer–cellular automata integrated model for urban growth simulation and optimization. Transactions in GIS, 23(4), 672–687. https://doi.org/10.1111/tgis.12517
10.1111/tgis.12517
Web of Science® Google Scholar
Cao, M., Tian, Y., Wu, K., Chen, M., Chen, Y., Hu, X., Sun, Z., Zuo, L., Lin, J., & Luo, L. (2023). Future land-use change and its impact on terrestrial ecosystem carbon pool evolution along the silk road under SDG scenarios. Science Bulletin, 68(7), 740–749. https://doi.org/10.1016/j.scib.2023.03.012
10.1016/j.scib.2023.03.012
PubMed Web of Science® Google Scholar
Chakraborty, A., & McMillan, A. (2015). Scenario planning for urban planners: Toward a practitioner's guide. Journal of the American Planning Association, 81(1), 18–29. https://doi.org/10.1080/01944363.2015.1038576
10.1080/01944363.2015.1038576
Web of Science® Google Scholar
Chen, G., Li, X., Liu, X., Chen, Y., Liang, X., Leng, J., Xu, X., Liao, W., Qiu, Y. A., & Wu, Q. (2020). Global projections of future urban land expansion under shared socioeconomic pathways. Nature Communications, 11(1), 537. https://doi.org/10.1038/s41467-020-14386-x
10.1038/s41467-020-14386-x
CAS PubMed Web of Science® Google Scholar
Chen, T., & Guestrin, C. (2016). Xgboost: A scalable tree boosting system. 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York, Unites States. http://doi.org/10.1145/2939672.2939785
10.1145/2939672.2939785
Google Scholar
Chen, Y., Bouferguene, A., Shen, Y., & Al-Hussein, M. (2019). Assessing accessibility-based service effectiveness (ABSEV) and social equity for urban bus transit: A sustainability perspective. Sustainable Cities and Society, 44, 499–510. https://doi.org/10.1016/j.scs.2018.10.003
10.1016/j.scs.2018.10.003
Web of Science® Google Scholar
Chen, Y., Li, X., Liu, X., & Ai, B. (2014). Modeling urban land-use dynamics in a fast developing city using the modified logistic cellular automaton with a patch-based simulation strategy. International Journal of Geographical Information Science, 28(2), 234–255. https://doi.org/10.1080/13658816.2013.831868
10.1080/13658816.2013.831868
Web of Science® Google Scholar
Daniel, C., & Pettit, C. (2022). Charting the past and possible futures of planning support systems: Results of a citation network analysis. Environment and Planning B: Urban Analytics and City Science, 49(7), 1875–1892. https://doi.org/10.1177/23998083211072866
10.1177/23998083211072866
Google Scholar
Debnath, R., Pettit, C., & Leao, S. Z. (2023). Opportunities and limitations of integrating computational and collaborative approaches to scenario planning. Journal of Urban Management, 12, 314–326. https://doi.org/10.1016/j.jum.2023.07.002
10.1016/j.jum.2023.07.002
Web of Science® Google Scholar
Debnath, R., Pettit, C., Soundararaj, B., Shirowzhan, S., & Jayasekare, A. S. (2023). Usefulness of an urban growth model in creating scenarios for city resilience planning: An end-user perspective. ISPRS International Journal of Geo-Information, 12(8), 311. https://doi.org/10.3390/ijgi12080311
10.3390/ijgi12080311
Web of Science® Google Scholar
Ding, Y., Cao, K., Qiao, W., Shao, H., Yang, Y., & Li, H. (2022). A whale optimization algorithm–based cellular automata model for urban expansion simulation. International Journal of Applied Earth Observation and Geoinformation, 115, 103093. https://doi.org/10.1016/j.jag.2022.103093
10.1016/j.jag.2022.103093
Web of Science® Google Scholar
Easthope, H., Crommelin, L., Kerr, S.-M., Troy, L., van den Nouwelant, R., & Davison, G. (2022). Planning for lower-income households in privately developed high-density Neighbourhoods in Sydney, Australia. Urban Planning, 7(4), 213–228. https://doi.org/10.17645/up.v7i4.5699
10.17645/up.v7i4.5699
Web of Science® Google Scholar
Echendu, A. J. (2020). The impact of flooding on Nigeria's sustainable development goals (SDGs). Ecosystem Health and Sustainability, 6(1), 1791735. https://doi.org/10.1080/20964129.2020.1791735
10.1080/20964129.2020.1791735
Web of Science® Google Scholar
Evans, R., Rosewall, T., & Wong, A. (2020). The rental market and COVID-19. 1. 1 Managing the Risks of Holding Self-Securitisations as Collateral 2. 11 Government Bond Market Functioning and COVID-19 3. The Economic Effects of Low Interest Rates and Unconventional 21 Monetary Policy 4. Retail Central Bank Digital Currency: Design Considerations, Rationales, 75.
Google Scholar
Feng, Y., Wang, J., Tong, X., Shafizadeh-Moghadam, H., Cai, Z., Chen, S., Lei, Z., & Gao, C. (2019). Urban expansion simulation and scenario prediction using cellular automata: Comparison between individual and multiple influencing factors. Environmental Monitoring and Assessment, 191, 1–20. https://doi.org/10.1007/s10661-019-7451-y
10.1007/s10661-019-7451-y
Web of Science® Google Scholar
Fu, B., Wang, S., Zhang, J., Hou, Z., & Li, J. (2019). Unravelling the complexity in achieving the 17 sustainable-development goals. National Science Review, 6(3), 386–388. https://doi.org/10.1093/nsr/nwz038
10.1093/nsr/nwz038
PubMed Web of Science® Google Scholar
Gao, Q., Shi, V., Pettit, C., & Han, H. (2022). Property valuation using machine learning algorithms on statistical areas in greater Sydney, Australia. Land Use Policy, 123, 106409. https://doi.org/10.1016/j.landusepol.2022.106409
10.1016/j.landusepol.2022.106409
Web of Science® Google Scholar
Gao, X., Zhu, J., & He, B. J. (2022). The linkage between sustainable development goals 9 and 11: Examining the association between sustainable urbanization and intellectual property rights protection. Advanced Sustainable Systems, 6(3), 2100283. https://doi.org/10.1002/adsu.202100283
10.1002/adsu.202100283
Web of Science® Google Scholar
Geoscience Australia. (2011). 1 second SRTM digital elevation model (DEM), bioregional assessment source dataset.
Google Scholar
Ghazaryan, G., Rienow, A., Oldenburg, C., Thonfeld, F., Trampnau, B., Sticksel, S., & Jürgens, C. (2021). Monitoring of urban sprawl and densification processes in western Germany in the light of SDG indicator 11.3. 1 based on an automated retrospective classification approach. Remote Sensing, 13(9), 1694. https://doi.org/10.3390/rs13091694
10.3390/rs13091694
Web of Science® Google Scholar
Han, H., Kim, S., Jin, M. Y., & Pettit, C. (2021). Providing affordable housing through urban renewal projects in Australia: Expert opinions on barriers and opportunities. International Review for Spatial Planning and Sustainable Development, 9(2), 41–61. https://doi.org/10.14246/irspsd.9.2_41
10.14246/irspsd.9.2_41
Web of Science® Google Scholar
Hao, L., He, S., Zhou, J., Zhao, Q., & Lu, X. (2022). Prediction of the landscape pattern of the Yancheng coastal wetland, China, based on XGBoost and the MCE-CA-Markov model. Ecological Indicators, 145, 109735. https://doi.org/10.1016/j.ecolind.2022.109735
10.1016/j.ecolind.2022.109735
Web of Science® Google Scholar
Hausfather, Z., & Moore, F. C. (2022). Net-zero commitments could limit warming to below 2°C. Nature Publishing Group UK London. https://doi.org/10.1038/d41586-022-00874-1
Google Scholar
He, C., Okada, N., Zhang, Q., Shi, P., & Zhang, J. (2006). Modeling urban expansion scenarios by coupling cellular automata model and system dynamic model in Beijing, China. Applied Geography, 26(3–4), 323–345. https://doi.org/10.1016/j.apgeog.2006.09.006
10.1016/j.apgeog.2006.09.006
Web of Science® Google Scholar
Hegre, H., Petrova, K., & Von Uexkull, N. (2020). Synergies and trade-offs in reaching the sustainable development goals. Sustainability, 12(20), 8729. https://doi.org/10.3390/su12208729
10.3390/su12208729
Web of Science® Google Scholar
Hien, P. D., Men, N., Tan, P., & Hangartner, M. (2020). Impact of urban expansion on the air pollution landscape: A case study of Hanoi, Vietnam. Science of the Total Environment, 702, 134635. https://doi.org/10.1016/j.scitotenv.2019.134635
10.1016/j.scitotenv.2019.134635
CAS PubMed Web of Science® Google Scholar
Huang, C.-W., McDonald, R. I., & Seto, K. C. (2018). The importance of land governance for biodiversity conservation in an era of global urban expansion. Landscape and Urban Planning, 173, 44–50. https://doi.org/10.1016/j.landurbplan.2018.01.011
10.1016/j.landurbplan.2018.01.011
Web of Science® Google Scholar
Icsu, I. (2015). Review of the sustainable development goals: The science perspective. International Council for Science (ICSU).
Google Scholar
Islam, M. D., Islam, K. S., & Mia, M. (2021). An XGBoost based approach for urban land use and land cover change modelling. Authorea Preprints.
10.22541/au.159646139.97656606/v2
Google Scholar
Kim, M., Kim, D., Jin, D., & Kim, G. (2023). Application of explainable artificial intelligence (XAI) in urban growth modeling: A case study of Seoul metropolitan area, Korea. Land, 12(2), 420. https://doi.org/10.3390/land12020420
10.3390/land12020420
Web of Science® Google Scholar
Kleeman, A., Hooper, P., Giles-Corti, B., & Foster, S. (2022). A new Australian dream? Exploring associations between apartment design attributes and housing satisfaction in three Australian cities. Cities, 131, 104043. https://doi.org/10.1016/j.cities.2022.104043
10.1016/j.cities.2022.104043
Web of Science® Google Scholar
Klosterman, R. E. (1999). The What If? collaborative planning support system. Environment and Planning B: Planning and Design, 26(3), 393–408. https://doi.org/10.1068/b260393
10.1068/b260393
Web of Science® Google Scholar
Klosterman, R. E. (2008). A new tool for a new planning: The What If? TM planning support system. In Planning support systems for cities and regions. Lincoln Institute of Land Policy.
Google Scholar
Klosterman, R. E. (2011). What If? 2.0 user's guide. http://www.whatifinc.biz/im/docs/UsersGuide2.0.pdf
Google Scholar
Kuc-Czarnecka, M., Markowicz, I., & Sompolska-Rzechuła, A. (2023). SDGs implementation, their synergies, and trade-offs in EU countries–sensitivity analysis-based approach. Ecological Indicators, 146, 109888. https://doi.org/10.1016/j.ecolind.2023.109888
10.1016/j.ecolind.2023.109888
Web of Science® Google Scholar
Li, T., Jing, X., Wei, O., Yinlong, L., Jinxuan, L., Yongfu, L., Li, W., Ying, J., Weipan, X., & Yaotian, M. (2022). Mobility restrictions and their implications on the rental housing market during the COVID-19 pandemic in China's large cities. Cities, 126, 103712. https://doi.org/10.1016/j.cities.2022.103712
10.1016/j.cities.2022.103712
PubMed Web of Science® Google Scholar
Li, W., El-Askary, H., Lakshmi, V., Piechota, T., & Struppa, D. (2020). Earth observation and cloud computing in support of two sustainable development goals for the river Nile watershed countries. Remote Sensing, 12(9), 1391. https://doi.org/10.3390/rs12091391
10.3390/rs12091391
Web of Science® Google Scholar
Li, X., Yeh, A., Liu, X., Li, S., & Yang, Q. (2020). Geographical simulation system: Cellular automata and spatial intelligence. Science Press.
Google Scholar
Li, X., Zhang, J., Li, Z., Hu, T., Wu, Q., Yang, J., Huang, J., Su, W., Zhao, Y., & Zhou, Y. (2021). Critical role of temporal contexts in evaluating urban cellular automata models. GIScience & Remote Sensing, 58(6), 799–811. https://doi.org/10.1080/15481603.2021.1946261
10.1080/15481603.2021.1946261
Web of Science® Google Scholar
Liang, X., Guan, Q., Clarke, K. C., Chen, G., Guo, S., & Yao, Y. (2021). Mixed-cell cellular automata: A new approach for simulating the spatio-temporal dynamics of mixed land use structures. Landscape and Urban Planning, 205, 103960. https://doi.org/10.1016/j.landurbplan.2020.103960
10.1016/j.landurbplan.2020.103960
Web of Science® Google Scholar
Liang, X., Liu, X., Li, X., Chen, Y., Tian, H., & Yao, Y. (2018). Delineating multi-scenario urban growth boundaries with a CA-based FLUS model and morphological method. Landscape and Urban Planning, 177, 47–63. https://doi.org/10.1016/j.landurbplan.2018.04.016
10.1016/j.landurbplan.2018.04.016
Web of Science® Google Scholar
Lin, L., Liang, Y., Liu, L., Zhang, Y., Xie, D., Yin, F., & Ashraf, T. (2022). Estimating PM2. 5 concentrations using the machine learning RF-XGBoost model in guanzhong urban agglomeration, China. Remote Sensing, 14(20), 5239. https://doi.org/10.3390/rs14205239
10.3390/rs14205239
Web of Science® Google Scholar
Lu, L., Qureshi, S., Li, Q., Chen, F., & Shu, L. (2022). Monitoring and projecting sustainable transitions in urban land use using remote sensing and scenario-based modelling in a coastal megacity. Ocean and Coastal Management, 224, 106201. https://doi.org/10.1016/j.ocecoaman.2022.106201
10.1016/j.ocecoaman.2022.106201
Web of Science® Google Scholar
Lu, Y., Cao, M., & Zhang, L. (2015). A vector-based cellular automata model for simulating urban land use change. Chinese Geographical Science, 25(1), 74–84. https://doi.org/10.1007/s11769-014-0719-9
10.1007/s11769-014-0719-9
CAS Web of Science® Google Scholar
Lu, Y., Laffan, S., & Pettit, C. (2022a). A geographically partitioned cellular automata model for the expansion of residential areas. Transactions in GIS, 26(3), 1548–1571. https://doi.org/10.1111/tgis.12866
10.1111/tgis.12866
Web of Science® Google Scholar
Lu, Y., Laffan, S., & Pettit, C. (2022b). The integration of cellular automata and What If? for scenario planning: Future residential expansion in the city of Ipswich. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 10, 151–158. https://doi.org/10.5194/isprs-annals-X-4-W3-2022-151-2022
10.5194/isprs-annals-X-4-W3-2022-151-2022
Google Scholar
Lu, Y., Shi, V., & Pettit, C. J. (2023). The impacts of public schools on housing prices of residential properties: A case study of greater Sydney, Australia. ISPRS International Journal of Geo-Information, 12(7), 298. https://doi.org/10.3390/ijgi12070298
10.3390/ijgi12070298
Web of Science® Google Scholar
Meftahi, M., Monavari, M., Kheirkhah Zarkesh, M., Vafaeinejad, A., & Jozi, A. (2022). Achieving sustainable development goals through the study of urban heat Island changes and its effective factors using spatio-temporal techniques: The case study (Tehran city). Natural Resources Forum, 46, 88–115. https://doi.org/10.1111/1477-8947.12245
10.1111/1477-8947.12245
Web of Science® Google Scholar
Mithun, S., Sahana, M., Chattopadhyay, S., Chatterjee, S., Islam, J., & Costache, R. (2022). Comparative framework for spatially explicit urban growth modeling for monitoring urban land-use efficiency and sustainable urban development (SDG 11.3. 1): A study on Kolkata metropolitan area, India. Geocarto International, 37(27), 17933–17970. https://doi.org/10.1080/10106049.2022.2136259
10.1080/10106049.2022.2136259
Web of Science® Google Scholar
Momeni, E., & Antipova, A. (2020). Pattern-based calibration of cellular automata by genetic algorithm and Shannon relative entropy. Transactions in GIS, 24, 1447–1463. https://doi.org/10.1111/tgis.12646
10.1111/tgis.12646
Web of Science® Google Scholar
Musa, S. I., Hashim, M., & Reba, M. N. M. (2019). Geospatial modelling of urban growth for sustainable development in The Niger Delta region, Nigeria. International Journal of Remote Sensing, 40(8), 3076–3104. https://doi.org/10.1080/01431161.2018.1539271
10.1080/01431161.2018.1539271
Web of Science® Google Scholar
Musango, J. K., Currie, P., Smit, S., & Kovacic, Z. (2020). Urban metabolism of the informal city: Probing and measuring the ‘unmeasurable’ to monitor sustainable development goal 11 indicators. Ecological Indicators, 119, 106746. https://doi.org/10.1016/j.ecolind.2020.106746
10.1016/j.ecolind.2020.106746
Web of Science® Google Scholar
NSW Government. (2018). Greater Sydney region plan—A metropolis of three cities. https://greatercities.au/strategic-planning/region-plans/metropolis-three-cities
Google Scholar
NSW Government (2020). NSW Bush Fire Prone Land. https://datasets.seed.nsw.gov.au/dataset/bush-fire-prone-land
Google Scholar
NSW Government (2021). State Environmental Planning Policy (Precincts – Western Parkland City) 2021. https://prod.planning-nsw.links.com.au/opendata/dataset/state-environmental-planning-policy-precincts-western-parkland-city-2021
Google Scholar
NSW Government. (2022). Population projections. https://www.planning.nsw.gov.au/research-and-demography/population-projections
Google Scholar
Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., Prettenhofer, P., Weiss, R., & Dubourg, V. (2011). Scikit-learn: Machine learning in python. The Journal of Machine Learning Research, 12, 2825–2830.
Web of Science® Google Scholar
Pettit, C., Shi, Y., Han, H., Rittenbruch, M., Foth, M., Lieske, S., van den Nouwelant, R., Mitchell, P., Leao, S., & Christensen, B. (2020). A new toolkit for land value analysis and scenario planning. Environment and Planning B: Urban Analytics and City Science, 47(8), 1490–1507. https://doi.org/10.1177/2399808320924678
10.1177/2399808320924678
Google Scholar
Pettit, C. J., Klosterman, R. E., Delaney, P., Whitehead, A. L., Kujala, H., Bromage, A., & Nino-Ruiz, M. (2015). The online What If? planning support system: A land suitability application in Western Australia. Applied Spatial Analysis and Policy, 8(2), 93–112. https://doi.org/10.1007/s12061-015-9133-7
10.1007/s12061-015-9133-7
Web of Science® Google Scholar
Pizzi, S., Caputo, A., Corvino, A., & Venturelli, A. (2020). Management research and the UN sustainable development goals (SDGs): A bibliometric investigation and systematic review. Journal of Cleaner Production, 276, 124033. https://doi.org/10.1016/j.jclepro.2020.124033
10.1016/j.jclepro.2020.124033
Web of Science® Google Scholar
Putatunda, S., & Rama, K. (2018). A comparative analysis of hyperopt as against other approaches for hyper-parameter optimization of XGBoost. International Conference on Signal Processing and Machine Learning. New York, United States. https://doi.org/10.1145/3297067.3297080
10.1145/3297067.3297080
Google Scholar
Qu, Y., Lin, Z., Li, H., & Zhang, X. (2019). Feature recognition of urban road traffic accidents based on GA-XGBoost in the context of big data. IEEE Access, 7, 170106–170115. https://doi.org/10.1109/ACCESS.2019.2952655
10.1109/ACCESS.2019.2952655
Web of Science® Google Scholar
Susan Solomon, D., Singh, C., & Islam, F. (2021). Examining the outcomes of urban adaptation interventions on gender equality using SDG 5. Climate and Development, 13(9), 830–841. https://doi.org/10.1080/17565529.2021.1939643
10.1080/17565529.2021.1939643
Web of Science® Google Scholar
Takagi, J., & Wada, A. (2019). Recent earthquakes and the need for a new philosophy for earthquake-resistant design. Soil Dynamics and Earthquake Engineering, 119, 499–507. https://doi.org/10.1016/j.soildyn.2017.11.024
10.1016/j.soildyn.2017.11.024
Web of Science® Google Scholar
Tang, L., Ke, X., Chen, Y., Wang, L., Zhou, Q., Zheng, W., & Xiao, B. (2021). Which impacts more seriously on natural habitat loss and degradation? Cropland expansion or urban expansion? Land Degradation & Development, 32(2), 946–964. https://doi.org/10.1002/ldr.3768
10.1002/ldr.3768
Web of Science® Google Scholar
Truelove, Y., & Ruszczyk, H. A. (2022). Bodies as urban infrastructure: Gender, intimate infrastructures and slow infrastructural violence. Political Geography, 92, 102492. https://doi.org/10.1016/j.polgeo.2021.102492
10.1016/j.polgeo.2021.102492
Web of Science® Google Scholar
United Nations. (2016). Sustainable Development Goals Report 2016. https://unstats.un.org/sdgs/report/2016/
Google Scholar
United Nations. (2018). World urbanization prospects: The 2018 revision. https://population.un.org/wup/Publications/Files/WUP2018-Methodology.pdf
10.18356/02486bd4-en
Google Scholar
Wack, P. (1985). Scenarios: Uncharted waters ahead. Harvard Business Review, 63(5), 72–89.
Web of Science® Google Scholar
Wang, R., Murayama, Y., & Morimoto, T. (2021). Scenario simulation studies of urban development using remote sensing and GIS. Remote Sensing Applications: Society and Environment, 22, 100474. https://doi.org/10.1016/j.rsase.2021.100474
10.1016/j.rsase.2021.100474
Google Scholar
Wang, Z., Gao, Y., Wang, X., Lin, Q., & Li, L. (2022). A new approach to land use optimization and simulation considering urban development sustainability: A case study of Bortala, China. Sustainable Cities and Society, 87, 104135. https://doi.org/10.1016/j.scs.2022.104135
10.1016/j.scs.2022.104135
Web of Science® Google Scholar
Wei, M., Zhang, Z., Long, T., He, G., & Wang, G. (2021). Monitoring Landsat based burned area as an indicator of sustainable development goals. Earth's Future, 9(6), e2020EF001960. https://doi.org/10.1029/2020EF001960
10.1029/2020EF001960
Web of Science® Google Scholar
Wu, H., Li, Z., Clarke, K. C., Shi, W., Fang, L., Lin, A., & Zhou, J. (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. https://doi.org/10.1080/13658816.2019.1568441
10.1080/13658816.2019.1568441
Web of Science® Google Scholar
Xu, T., Gao, J., & Coco, G. (2019). Simulation of urban expansion via integrating artificial neural network with Markov chain–cellular automata. International Journal of Geographical Information Science, 33(10), 1960–1983. https://doi.org/10.1080/13658816.2019.1600701
10.1080/13658816.2019.1600701
Web of Science® Google Scholar
Yang, J., Dong, J., Sun, Y., Zhu, J., Huang, Y., & Yang, S. (2022). A constraint-based approach for identifying the urban–rural fringe of polycentric cities using multi-sourced data. International Journal of Geographical Information Science, 36(1), 114–136. https://doi.org/10.1080/13658816.2021.1876236
10.1080/13658816.2021.1876236
Web of Science® Google Scholar
Yang, J., Tang, W., Gong, J., Shi, R., Zheng, M., & Dai, Y. (2023). Simulating urban expansion using cellular automata model with spatiotemporally explicit representation of urban demand. Landscape and Urban Planning, 231, 104640. https://doi.org/10.1016/j.landurbplan.2022.104640
10.1016/j.landurbplan.2022.104640
Web of Science® Google Scholar
Yates, J. (2016). Why does Australia have an affordable housing problem and what can be done about it? The Australian Economic Review, 49(3), 328–339. https://doi.org/10.1111/1467-8462.12174
10.1111/1467-8462.12174
Web of Science® Google Scholar
Zhai, S., Feng, Y., Yan, X., Wei, Y., Wang, R., & Li, P. (2021). Using spatial heterogeneity to strengthen the neighbourhood effects of urban growth simulation models. Journal of Spatial Science, 68, 319–337. https://doi.org/10.1080/14498596.2021.1982783
10.1080/14498596.2021.1982783
Web of Science® Google Scholar
Zhai, Y., Yao, Y., Guan, Q., Liang, X., Li, X., Pan, Y., Yue, H., Yuan, Z., & Zhou, J. (2020). Simulating urban land use change by integrating a convolutional neural network with vector-based cellular automata. International Journal of Geographical Information Science, 34, 1–25. https://doi.org/10.1080/13658816.2020.1711915
10.1080/13658816.2020.1711915
Web of Science® Google Scholar
Zhao, X., Xia, N., Xu, Y., Huang, X., & Li, M. (2021). Mapping population distribution based on XGBoost using multisource data. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 14, 11567–11580. https://doi.org/10.1109/JSTARS.2021.3125197
10.1109/JSTARS.2021.3125197
Web of Science® Google Scholar
Zhou, M., Ma, Y., Tu, J., & Wang, M. (2022). SDG-oriented multi-scenario sustainable land-use simulation under the background of urban expansion. Environmental Science and Pollution Research, 29(48), 72797–72818. https://doi.org/10.1007/s11356-022-20904-9
10.1007/s11356-022-20904-9
PubMed Web of Science® Google Scholar
Zhu, J., Sun, Y., Song, S., Yang, J., & Ding, H. (2021). Cellular automata for simulating land-use change with a constrained irregular space representation: A case study in Nanjing city, China. Environment and Planning B: Urban Analytics and City Science, 48(7), 1841–1859. https://doi.org/10.1177/2399808320949889
10.1177/2399808320949889
Google Scholar
Zhu, R., Guilbert, É., & Wong, M. S. (2020). Object-oriented tracking of thematic and spatial behaviors of urban heat islands. Transactions in GIS, 24(1), 85–103. https://doi.org/10.1111/tgis.12586
10.1111/tgis.12586
Web of Science® Google Scholar

Volume28, Issue6

September 2024

Pages 1465-1485

This article also appears in:

Innovative Research on SDGs Supported by Big Earth Data

Coupling cellular automata and What If? models for residential expansion simulation: A case study of Southwest Sydney, Australia

Abstract

1 INTRODUCTION

2 METHODOLOGY

2.1 The general CA-What If? modeling framework

2.2 Land use demand calculation using What If?

2.3 CA model calibration using XGBoost

3 CASE STUDY

3.1 Study area and data processing

3.2 The land use demand of Southwest Sydney

3.3 Hyperparameter tuning

3.4 XGBoost-based CA sub-model training and testing

3.5 Scenario planning outcomes using a CA-What If? model

4 DISCUSSION

5 CONCLUSIONS

ACKNOWLEDGMENTS

CONFLICT OF INTEREST STATEMENT

Open Research

DATA AVAILABILITY STATEMENT

REFERENCES

Figures

References

Information

About Wiley Online Library

Help & Support

Opportunities

Connect with Wiley