As a human-dominated landscape, human settlements ranging in size from hamlets and villages to towns and cities play a crucial role in environmental and ecological changes [1
]. Understanding human settlement patterns, therefore, is important for a range of issues, such as city planning, environmental conservation, resource sustainability and political decision-making [4
]. Based on the statistically significant relationship between nocturnal artificial lighting signals and several urbanization and socioeconomic variables over time and space [7
], satellite-derived nighttime light data, which were previously provided by the Defense Meteorological Satellite Program (DMSP), have been widely applied when investigating socioeconomic dynamics and mapping urbanization at various spatial and temporal scales [9
]. The most notable advantage of nighttime light images is that they can provide timely and consistent observations of socioeconomic dynamics with spatially explicit characteristics and spatially represent a variety of information for human activities when compared to traditional statistical data and visible and near-infrared remote sensing data, respectively [13
]. In practice, however, DMSP nighttime light datasets have several considerable drawbacks due to factors such as a relatively coarse spatial resolution, saturated lighting signals due to the limited capability of the six-bit quantization sensors and over-glow effects caused by light diffusion from adjacent areas, which can visibly affect the quantitative measurement of human activities, particularly at a fine scale [14
]. Thus, most previous studies based on DMSP nighttime light data have generally been limited to regional or sub-regional level surveys of urbanization dynamics and socioeconomic development, even though several efforts have been made to calibrate radiance values and reduce saturation effects in DMSP images [16
Recent global composites of nighttime light images, which have been derived from the Visible Infrared Imaging Radiometer Suite (VIIRS) instrument day/night band (DNB) since 2012 [19
], show an unprecedented and detailed look at human settlements due to a wider range of radiometric detection, a higher spatial resolution and fewer over-glow effects compared to the DMSP data [20
]. Hence, VIIRS nighttime light datasets might have great potential for the detailed characterization of human settlements with various socioeconomic activities at a fine scale [23
]. Several studies have demonstrated that VIIRS images can generate more accurate measures of regional socioeconomic variables [25
], as well as the spatial extent of urbanized land [28
Apart from regional-level investigations of the quantitative responses of nighttime lighting signals to various urbanization parameters and socioeconomic variables, many efforts have been made to develop applicable nighttime light data methods to extract urbanized land features from both DMSP and VIIRS images [29
]. Dou et al. [34
] summarized these methods into three different categories: thresholding, a composite index and supervised classification. Examples of the recent use of these methods mainly include local-optimized thresholding (LOT), the vegetation-adjusted nighttime light urban index (VANUI) and support vector machine classification (INNL-SVM). For the most part, these methods mainly focus on the automatic extraction of the spatial extent of urban built-up areas that closely match the statistical data or the mapping results of high spatial resolution images. Ma et al. [35
] proposed a pixel-level approach based on the quadratic relationship between nocturnal radiance values and the brightness gradient to partition DMSP nighttime light images into five types of lighting areas for individual cities. Each type of identified area can be associated with specified subareas of human settlement, with various forms and human activities. However, this partition method might not be applicable to VIIRS data because increased spatial fluctuations in radiance values at a relatively fine scale may not support the robust quadratic relationship.
Further applications of nightlight imagery, particularly when investigating human settlement patterns at the local scale, are still less documented and are a challenge, which is largely due to the lack of partition approaches for delineating the spatial features of artificial landscapes from nighttime light images. Although both remotely sensed DMSP and VIIRS images provide us with an efficient way to survey the overall degree of regional socioeconomic activities through anthropogenic nightlight emissions, it is hardly possible to produce spatially explicit classifications of the land cover feature composite from nocturnal brightness signals without connections to spectral and texture characteristics of lit artificial objects. Hence, a spatially explicit partition method is required for delineating human settlements using nighttime light images. Moreover, an association between nightlight signals and social media-derived information can further improve our understandings of spatial patterns in demographic and socioeconomic activities [36
The primary objective of this study is to develop a spatial partition approach to identify different types of lit areas in human settlements from VIIRS nighttime light images. Our method is primarily derived from a watershed concept, which is based on local lighting hotspots and identifies adjacent units like a topographic map, with a second order exponential decay model which is used for measuring changes in brightness with respect to the rank among different partitioned units. Satellite-derived building density data and social media-based human activity information are jointly used to further identify partitioned sub-regions into different types of lighting areas. Based on nightlight watershed unit, the utility of the watershed-based approach for delineating human settlement patterns is demonstrated using points of interest (POI) data to represent various settlement types at the local scale, and further applications of our method for delineating human settlements are carried out for 99 cities in 10 urban agglomerations.
Timely and consistent observations of artificial lighting areas at night provide a unique data source for investigating urbanization dynamics and socioeconomic activities across human settlements in a spatially explicit manner [43
]. The fundamental aspects of satellite-derived nighttime light data regarding their application are positive and monotonic responses (e.g., linear, exponential and power-law functions) of total nighttime brightness or lit areas to corresponding demographic and socioeconomic variables at the local and regional scales [44
]. A reduction in texture information regarding artificial objects in nighttime light images is still a major challenge when delineating the spatial patterns in human settlements using nocturnal luminosity data, even for VIIRS images with markedly enhanced spatial resolutions and less over-glow effects than those of traditional DMSP images [34
Based on the local hotspots of lighting signals and the spatial fluctuations in nighttime brightness, we developed a watershed-based partition approach for VIIRS images. This method is used to separate adjacent sub-regions like a topographic map. NTL thresholds, which can be used for the further classification of partitioned watershed divisions, are developed by a second order exponential decay model, which can outline the spatial transitions in nighttime radiances along both rural–urban and urban–rural gradients. Using satellite-derived building densities and social media-derived human activity data, we finally obtained a classification of watershed lighting divisions, including five types of high, high-medium, medium, medium-low and low-density lighting and human activity areas across the human settlement. Different types of partitioned and classified areas can be connected to distinct human activities through quantitative comparisons of the relative proportions of different types of POIs (points of interest). Moreover, we find that NTL thresholds for the partitioning of VIIRS images are inter-regionally various and likely depend on the size and form of the human settlement. Thus, nightlight image-based investigations of human settlement patterns should vary according to the features of the target region, even for VIIRS data with a high spatial resolution and few over-glow and saturation effects. In addition, our results suggest that the association of nighttime light data with geo-located social median data should further benefit our understanding of patterns in human settlement with enhanced human activity information.
Our watershed-based partition method can provide insight into the application of VIIRS nighttime light images when spatially characterizing the various patterns and degrees of human activity across the human settlement. It is noteworthy that our method uses a watershed lighting area as an analysis unit and, hence, several details regarding local human activity might be ignored. Therefore, a pixel-based partition method that considers the pixel-level response of nighttime radiance signals to local human activity at a fine scale is essential and should be investigated in order to promote future applications of VIIRS nighttime light data to urbanization and socioeconomic dynamics with spatially explicit details.