Coupling agent-based, cellular automata and logistic regression into a hybrid urban expansion model (HUEM)

Abstract

Several methods for modeling urban expansion are available. Most of them are based on a statistical, a cellular automaton (CA) and/or an agent-based (AB) approach. Statistical and CA approaches are based on the implicit assumption that people's behavior is not likely to change over the considered time horizon. Such assumption limits the ability to simulate long-term predictions as people's behavior changes over time. An approach to consider people's behavior is the use of an AB system, in which the decision-making process of agents needs to be parameterized. Most existing studies, which make use of empirical data to define the agents' decision-making criteria, rely on intensive data collection efforts. The considerable data requirements limit the AB-system's ability to model a large study area, as the number of agents for which data on decision-making criteria is required, increases with the size of the study area. This paper presents a hybrid urban expansion model (HUEM) that integrates logistic regression (Logit), CA and AB approaches to simulate future urban development. A key feature of HUEM lies in its ability to address various people behaviors that are variable over time through AB relying on a sample approach that combines Logit and CA. Three agent sets are defined; developer agents, farmer agents and planning permission authority agent. The agents' decision-making process is parameterized using CA and Logit models. The interactions of the agents are simulated through a series of rules. To assess HUEM performance, it is calibrated for Wallonia (Belgium) to simulate urban expansion between 1990 and 2000. Calibration results are then assessed by comparing the 2000 simulated map and the actual 2000 land-use map. Furthermore, the performance of HUEM is compared to a number of typical spatial urban expansion models, i.e. Logit model, CA model and CA-Logit to assess the added-value of HUEM. The comparison shows the performance of HUEM is better than other models in terms of allocation ability.

Comparing support vector machines with logistic regression for calibrating cellular automata land use change models

“…AB models (e.g. Mustafa et al, 2017;Zhang, Zeng, Bian, & Yu, 2010) examine agents as goal-oriented entities capable of responding to their environment and taking independent actions, where these agents may represent individuals, institutions etc. In AB models, solutions have been designed to explore the emergent properties of systems with relatively simple behavioral rules representing individual agents.…”

Addressing the determinants of built-up expansion and densification processes at the regional scale

“…Feng et al, 2011;Hao et al, 2013;Hu and Lo, 2007;Liu et al, 2008;Vermeiren et al, 2012). Most studies assume that builtup/urban expansion is a binary process, contrasting two classes of cells, i.e., built-up vs nonbuilt-up cells (e.g., Mustafa et al, 2017;Vermeiren et al, 2012). Such a binary representation of built-up environment somehow disregards the fuzzy nature of urban boundaries (Ban and Ahlqvist, 2009).…”

Benefits of a multiple‐solution approach in land change models

“…Model calibration is the process of finding a best‐fit set of spatial parameters presented in Equations (1), (2), and (3). Often, the driving force weights, Equation (2), are defined with a logistic regression (e.g., Achmad, Hasyim, Dahlan, & Aulia, ; Mustafa et al, ). However, some studies used optimization research algorithms to define the weights of driving forces (e.g., Feng et al, ).…”

“…The assessment of different models depends on the quality of their outcomes, and therefore model calibration is a critical step in the construction of the model (van Vliet et al, ). Several methods have been proposed in the literature to calibrate spatial LC models, including a visual test (e.g., Clarke, Hoppen, & Gaydos, ; Ward, Murray, & Phinn, ), multi‐criteria evaluation (e.g., Mahiny & Clarke, ), reusing parameters from other studies (e.g., Mustafa, Saadi, Cools, & Teller, ), statistical analysis (e.g., García, Santé, Boullón, & Crecente, ), machine learning (e.g., Mileva, Suzana, Miloš, & Branislav, ), artificial neural networks (e.g., Basse, Omrani, Charif, Gerber, & Bódis, ; Pijanowski et al, ), and search algorithms for optimization such as genetic algorithms (e.g., Mustafa, Heppenstall et al, ), particle swarm optimization (e.g., Feng, Liu, Tong, Liu, & Deng, ), and a combination of various methods (e.g., Mustafa, Cools, Saadi, & Teller, ). Although there is no standard calibration method, statistical analysis is one of the most frequently applied calibration approaches (van Vliet et al, ).…”