The rate of urbanization all over the world is quite alarming, with the proportion of the world’s population living in urban environments projected to reach 66% by 2050 [1
]. However, the understanding of urbanization is primarily based on population figures obtained from the United Nations; these statistics do not include any information on the population spatial distributions, or evolution metrics of built-up areas within urban environments [2
Urbanization, urban growth, urban sprawl and urban expansion are different concepts that have caused much confusion in multidimensional urban systems analysis. For a differentiation, it can be noted that urbanization can be viewed as a characteristic of the population as a particular kind of land use and land cover, as well as a characteristic of social and economic processes and interactions affecting both population and land [5
]. Urban growth mainly refers to an increase in urban population size, independent of rural population [7
]. Urban sprawl is treated as a process that focuses on describing pattern of land-use in an urbanized area through eight distinct dimensions: density, continuity, concentration, compactness, centrality, nuclearity, diversity, and proximity [5
]. There is no specific definition of urban expansion—this concept is commonly used to describe urban population, physical expansion, quality of urban layout, land and housing regulation and so on [9
]. In most of real world situations, these terms cannot be clearly separated, since urbanization, urban growth, urban sprawl and urban expansion are highly interlinked. However, it is important to realize that the huge growth of urban population may cause uncontrolled urban growth, resulting in urban sprawl and urban expansion. Urbanization may also result from and contribute to urban growth, urban sprawl and urban expansion [5
Although the complexity of these four terms and their ambiguous and partially overlapping meanings make it difficult reach a consensus of a distinct urban phenomenon analysis, a variety of urbanization parameters from the standpoint of the built environment have been proposed to describe urbanization trends [11
]. For instance, some of these used specific landscape spatial metrics to characterize the configuration and composition of urban areas (e.g., [15
]). While many such parameters have been identified, it is generally difficult to distinguish between those that are useful and those that are not. Some studies have measured the accessibility within each city on the basis of gravity transportation modelling (e.g., [12
]), but the acquisition of these transport network datasets is problematic. Generally, there appears to be no consensus between those investigating urban landscapes on the parameters to use for urbanization velocity evaluation. This research therefore provides below a succinct review of the most widely accepted and commonly used parameters for measuring urbanization in order to provide a descriptive framework that can be used for measuring urbanization velocity [8
A number of studies have also shown that remote sensing data and associated techniques are advantageous for classifying, monitoring and analyzing urbanization and its development over time at a range of scales, with an emphasis on mapping large areas at a time [22
]. Urbanization velocity (also called urban expansion speed) is defined as the annual growth rate of urban area within a period. It indicates the absolute differences (in terms of footprints) of urban areas within a certain time period [25
]. The measurement of urbanization using remote sensing imagery has been widely used for mapping, quantification, and documentation of the extent, growth rates, and percentage change in urban areas [26
], and can also be used to predict possible future urban growth [29
]. Especially the grid-based urbanization velocity analysis, which typically involves the use of a “grid-based moving window” on remote sensing imagery to detect the spatial gradient changes of grid-based land use and land cover [31
], nighttime luminosity changes [30
], population changes [35
], or temperature changes [36
] through time, were popularly defined to describe the urbanization process. Grid-based urbanization velocity analyses do not require extensive auxiliary data to obtain spatial-temporal urban growth [27
], it is able to avoid some of the redundancies of estimations that are caused by many landscape parameters [8
] when attempting to describe quantitatively the human settlement patterns [37
]. It aims to quantify the local urbanization velocity across a landscape [32
]. Spatial and temporal grid-based gradient information can therefore provide consistent, spatially explicit parameters that can be used to record the expansion of urban areas by estimating the speed and direction of urban growth [28
Grid-based urbanization velocity analyses are, however, sensitive to the size of the “grid” (or the spatial resolution) used to compute spatial and temporal characteristic of the urbanization process [32
]. Grid-based urbanization velocity analyses can, in theory, be carried out at any scale and descriptions of urban growth may therefore need to be tied to a specific scale. There is little knowledge about the effects of spatial resolution on urban velocity analyses [47
]. At very high spatial resolutions, transitional patches of human settlement tend to be complex, while coarser resolutions tend to smooth out the effects of urbanization. By contrast, although a coarse resolution urbanization velocity map may show more instances of the same urban expansion features than a high resolution map, it may demonstrate the agglomeration of land cover types and reveal greater urbanization evolution process that are not easily detected from the high resolution map [48
]. The challenge is therefore to establish a meaningful and useful multi-resolution urbanization velocity model to investigate the spatial resolution issue on grid-based urbanization velocity analyses.
The first objective of this research was to investigate the effect of imagery spatial resolution on grid-based urbanization velocity analysis, which required an investigation into the most appropriate size to use for the “moving window” (expressed as the grid size, which is an effective surrogate for spatial resolution). Since the widespread use of remote sensing data has generally increased interest in studying the capability issues relating to spatial resolution of images, the second objective was thus to investigate whether or not the available fine resolution remote sensing data for grid-based urbanization velocity analyses in the case study of Pearl River Delta are appropriate for the spatial resolution at which the investigated processes operate, or the spatial resolution at which decisions are required. The third objective was to take the grid-based urbanization velocity as the independent variable, and test the inherent characteristics of spatial autocorrelations and spatial structure heterogeneities. These characteristics are useful to effectively validate the results of grid-based urbanization velocity analyses at different spatial resolution.
Increasing numbers of flexible variance methods are providing effective ways of identifying spatial resolution thresholds and spatial resolution domain problems. Frequently used methods for dealing with different spatial resolutions (referred to as multi-scale methods) include local variance analysis [49
], geographical variance analysis [51
], semivariance analysis [53
], multifractal analysis [55
], wavelet transform analysis [57
], and Fourier transform analysis [59
]. All of these methods are able to quantify landscape characteristics of different spatial resolutions in their mathematical formulations and procedures, but their selection heavily depends on the nature of the data and the objectives of the investigation [53
In this study, the grid-based urbanization velocity analysis is based on spatio-temporal changes within an urban area compared to neighboring areas, using a moving window. This research systematically organizes the moving window (i.e., spatial resolution or grid size of remote sensing data) into a hierarchical, grid-based nested dataset. Through a thorough and succinct literature analysis [50
], the empirical variogram model, local variance model and the geographical variance model, were selected to investigate the urbanization velocity analysis results at different spatial resolutions. They show the advantages of analyzing the spatial autocorrelation and spatial structural heterogeneities for hierarchical grid size. Meanwhile, these three methods have rarely been used to identify spatial resolution thresholds and domains within urbanization velocity analyses. This research also can quantify and validate their capability in investigating the influence that the hierarchical grid size (i.e., spatial resolution) has on quantitative descriptions of urban growth. The investigation of grid size (i.e., spatial resolution) effects on urbanization velocity analysis was set out in the Pearl River Delta study area between the years of 2000 and 2015.
6. Conclusions and Outlook
Remote sensing data are widely believed to be constrained by the minimum spatial resolution of the sensor. This research has, however, shown that a resolution of 30 m in Landsat satellite data was more than adequate for a multi-scale urbanization velocity analysis over the whole of the PRD. Coarser resolution satellite data (>30 m) clearly presents new possibilities for multi-resolution urban growth analysis, particularly with regard to differentiating urban structures within cities. Nevertheless, fine to coarse resolution imagery data are indispensable for long-term observations of spatial distributions and fluctuations in built-up areas, and enable planners to track urbanization processes in a spatially and temporally explicit manner. Compared to other ways of grid-based urbanization velocity analysis utilizing built-up area information, which are based on using the variations of vegetation cover [33
], impervious surface [95
], night-time luminosity data [96
], and land surface temperature data [97
], are somewhat limited by the time-consuming nature of image classification.
Spatial resolution can be used to refer both to the magnitude of a study (e.g., its geographic extent) and also to the degree of detail (e.g., its spatial variance) [77
]. Concepts of spatial autocorrelation and intra-pixel heterogeneity within the structure of multi-resolution data were thus used in this research as a basis for understanding the effects of spatial variability in a grid-based hierarchy. With the advantages of being simple, well grounded in theory, and largely compatible with multi-scale grid models, the empirical variogram, local variance and geographical variance models used in this research have provided an understanding of the nature and cause of spatial variability in satellite imagery. Nevertheless, investigations into the effects of scale on irregularly grouped hierarchical data now have access to a growing arsenal of techniques. Among them this research encountered various kinds of scale-based decomposition of variation. Further research is required into developing systematic procedures for extrapolating information from one scale to another.
Spatial resolution has emerged as a critical issue in the description of the hierarchical structure of complex urban systems. Remote sensing data and geographic technologies will undoubtedly advance the development of research into multi-scale urban geography. When exploring multi-resolution issues in urbanization velocity analyses, this research found that urbanization velocity analysis across multiple resolutions may yield accurate representations, but the specifications of feature characteristics may differ for different scales. Meanwhile, relationships between socio-economic indicators and urbanization velocity that are independent of scale may provide an initial insight into likely quantitative relationships between the spatial structures within urban landscapes and the social characteristics of urbanization. This may aid in the understanding of how social and economic factors, as well as population growth, might influence urban expansion and density within the PRD. The usefulness of grid-based urbanization velocity analysis may be critically affected by the nature of the grid size in the original remotes sensors. Nevertheless, the remote sensing-based conceptual distinctions and methodological guidelines regarding spatial resolution in this research can help resolve “the spatial resolution question” in urban geography.