Abstract
Remote-sensing images are visually interpreted in this study to obtain information on buildings in the urban and rural areas of Ningxia, China. Overall, area estimates yielded by the proposed equations followed a normal distribution. Correlation and error analyses indicated that the coefficients are reasonable and reliable and that the building area estimates have an accuracy of 90% and are also reliable. These results were used in conjunction with drone aerial images, Baidu street view images, and paper maps to determine the seismic performance (SP) of the buildings in the study area. On this basis, the buildings were classified into three groups, namely, those with the required SP, suspected substandard SP, and substandard SP. Examination based on the field survey data collected from at least one sample site in each village and township in all 22 county-level divisions (CLDs) of Ningxia showed an average SP accuracy of 76% for all 22 CLDs and an SP accuracy exceeding 70% for 20 (91%) of the 22 CLDs. Based on this approach and the results obtained, the ArcGIS spatial analysis method was employed to determine the percentages and distribution patterns of the buildings in the three SP groups in the 22 CLDs. The results revealed the following features. Buildings with the required SP were clustered in the urban areas of each CLD, with a few in the village and township government seats. Buildings with suspected substandard SP were distributed predominantly in the rural-urban fringe (RUF) areas and the village and township government seats. Buildings with substandard SP were found primarily in urban villages, RUF areas, and urban areas. The soundness of the spatial analysis results was corroborated by the field survey data, lending credence to the feasibility of the proposed calculation method. This method can satisfy the real-world need for rapidly assessing the SP and distribution of buildings in a region before an earthquake occurs and provide a reliable reference for disaster prevention, mitigation, and relief efforts.
1. Introduction
Earthquakes are unpredictable and highly destructive natural disasters that gravely threaten social and economic development, the safety of lifeline facilities, and people’s lives and property [1]. Rapid economic progress and accelerated urbanisation have broadened and deepened the impact of destructive earthquakes on urban development in China.
Aboveground structures often suffer damage and even collapse during destructive earthquakes, constituting the principal cause of casualties and property losses. Massive volumes of building data considerably affect the timeliness and accuracy of rapid damage assessments after strong earthquakes. Generally, postdisaster building damage can be assessed through either a detailed field investigation of the damage sustained by individual buildings or the determination of the damage sustained by buildings over a large area using an efficient analysis method [2]. Findings from available studies suggest that the number of buildings within an assessment zone increases exponentially as the earthquake magnitude increases [3]. Field investigations are time- and labour-intensive and thus unsuitable for assessing damage over a large area after a strong earthquake [4]. Hence, the focus of research examining earthquake damage has gradually shifted from postearthquake investigations to preearthquake disaster mitigation preparations. Well-prepared emergency management strategies can alleviate the losses from earthquakes [5].
Because of its ability to provide timely, multipurpose, multiangle image-based services, remote-sensing (RS) technology has become a convenient tool for obtaining information before and after an earthquake and facilitating postearthquake emergency response and recovery efforts. The earliest use of RS technology to acquire earthquake damage information dates back to 1906 when G. R. Lawrence photographed San Francisco in the United States after a magnitude-8.3 earthquake using kites. Technological advances have enabled substantial progress in RS technology in recent decades. With higher resolutions and update rates, satellite images can now better facilitate assessment of the damage caused by historical earthquakes and the acquisition of earthquake damage information. A growing number of researchers have retrieved information on the damage sustained by buildings over large areas during earthquakes from RS imagery [4, 6–17]. Satellite RS technology has become a convenient and efficient tool for obtaining and disseminating information before and after earthquakes, facilitating a more effective emergency response to earthquakes and minimising their impact. Swiftly understanding the seismic performance (SP) of buildings in an earthquake-prone area before the occurrence of an earthquake plays a crucial role in urban disaster mitigation and emergency rescue efforts.
In this study, using the Seismic Ground Motion Parameter Zonation Map of China (GB18306-2015) [18] and the Code for Seismic Design of Buildings (GB50011-2016) [19] as guidelines, urban and rural buildings in Ningxia, China, were preliminarily classified based on their SP through visual interpretation of high-resolution RS imagery into three groups (termed SP groups), namely, those with the required SP, suspected substandard SP, and substandard SP. The reliability of the results obtained based on visual interpretation is comprehensively examined using drone aerial images, Baidu street view images, and paper maps. However, visual interpretation of RS imagery can only yield the number of buildings and cannot give the area of the buildings in each SP group. Therefore, a method must be established to determine the total area of the buildings in each of the three SP groups and its percentage in a region, with the goal of effectively mitigating earthquake damage. Rural and urban buildings are separately classified in this study. Separate equations are established to calculate the areas of rural and urban buildings. Error analysis revealed a high level of consistency between the building areas determined based on field survey data and rapidly yielded by the models. Within a permissible error range, the calculation models can be used to quickly estimate building areas in a region. On this basis, together with the SP estimates, the proportions of buildings with different SP levels and the distribution pattern of SP in a region can be determined. This method can provide powerful support for obtaining building information for an earthquake-prone region and estimating potential economic losses before an earthquake.
2. Study Area and Image Interpretation
2.1. Study Area
In terms of the geological structure, Ningxia is situated on the northeastern margin of the Tibetan Plateau, where active faults are densely distributed. According to historical records, this region has experienced multiple strong earthquakes, including two earthquakes with a magnitude of 8 and above: the magnitude-8 earthquake that occurred in Pingluo, Yinchuan, in 1739 and the magnitude-8.5 earthquake that occurred in Haiyuan in 1920. Modern instrumental records also show that Ningxia is prone to earthquakes. Figure 1 shows the study area.
2.2. Image Interpretation
Gaofen-2 imagery is used for analysis in this study. Its high spatial resolution (>1 m) meets the requirement for visual interpretation and allows aboveground structures to be quickly classified based on their SP. Considering its offsets, the geographical information system method is employed to correct the position coordinates in the Gaofen-2 imagery in the ArcGIS platform prior to visual interpretation [20].
The Seismic Ground Motion Parameter Zonation Map of China (GB18306-2015) and the Standard for Classification of Seismic Protection of Building Constructions (GB50223-2008) [21] are used in this study as a basis for determining building SPs. To relatively accurately distinguish building SPs, urban and rural buildings are processed separately during visual interpretation. The outlines of urban buildings are extracted individually based on their image texture. In other words, one polygon corresponds to one building. In contrast, the outlines of rural buildings are extracted based on the outlines of the blocks where they are located. In other words, one polygon can contain one or multiple buildings. Figure 2 illustrates the overall process used to determine building SPs and validate the results based on field survey data. Figure 3 shows separate standard maps of building outline information for urban and rural areas.
(a)
(b)
The Code for Seismic Design of Buildings (GB50011-2016) stipulates three fortification levels, that is, “no damage under minor earthquake,” “repairable damage under moderate earthquake,” and “no collapse under major earthquake.” The degree of damage to buildings is qualitatively classified in the Classification of Earthquake Damage to Buildings and Special Structures (GB/T 24335-2009) [22]. According to the Seismic Ground Motion Parameter Zonation Map of China (GB18306-2015), Ningxia is mainly located in seismic fortification area VIII. In this study, the buildings in Ningxia are divided into three categories, namely, buildings with the required SP, suspected substandard SP, and substandard SP, according to the seismic standard for seismic fortification area VIII and with reference to the above two specifications.
Visually interpreting images of urban buildings is relatively easy. A building is considered to meet the SP standard if its polygon has a well-defined texture boundary. Buildings whose polygons have an indistinct or dark grey texture boundary are mostly structures consisting of 3–5 storeys and are suspected to be below the SP standard. Buildings lacking the above features and whose polygons are appreciably smaller than those of the buildings in the above two groups are mostly structures in urban villages and are considered to be below the SP standard. Using Baidu street view images can help determine the SP of individual buildings that appear to be ambiguous based on the information extracted from the RS images.
Rural and rural-urban fringe (RUF) areas are home to buildings of complex and varied structures and types. Therefore, a uniform standard for determining the SP of buildings needs to be established to facilitate laboratory visual interpretation. Prior to visual interpretation, field surveys were conducted to determine the typical features of the buildings in representative areas identified in the RS images. It is difficult to visually interpret RS images for buildings in rural and RUF areas and to extract their information (e.g., geometries and roof and shadow features) from RS images. Considering these factors in conjunction with field sample survey data, buildings in rural and RUF areas are classified based on their SP into three groups. The classification criteria, as shown in Figure 4, are described as follows:(1)Buildings designed and constructed in rural areas by their owners themselves lack a uniform plan and layout and display various texture patterns and colours in the RS images. These buildings are classified as self-constructed (SC) buildings in rural areas. Field surveys show that most SC buildings are not seismically fortified and do not meet the SP standard. In the RS images, these buildings feature blurry texture boundaries and are mostly dark grey. During field surveys, extremely few SC buildings were found to possess seismic structures, such as structural columns and ring beams. These buildings do not fully meet the SP standard and are suspected to be below the SP standard. Their texture boundaries appear relatively distinct in the RS images.(2)A relatively uniform plan and layout can be observed for the streets within the blocks in the RS images. Buildings showing similar texture patterns and colours in a polygon are uniformly classified as government-subsidised (GS) buildings. These buildings are constructed by the government according to the Standard for Seismic Fortification of Buildings and are therefore considered to meet the SP standard.(3)RUF areas are home to alternating seismically fortified and unfortified buildings of various forms. As seen in the RS images, of the buildings in RUF areas, those neighbouring thoroughfares are mostly multistorey structures, while those distant from thoroughfares are generally densely and irregularly distributed low-rise structures. These buildings are classified as RUF buildings. Field surveys show that the SP standard is met by multistorey RUF buildings adjacent to thoroughfares but not by densely and irregularly distributed low-rise RUF buildings.
Typical buildings were photographed during field surveys. Information obtained from the photographs was subsequently statistically summarised to establish a database on the ArcGIS platform as a standard for determining building SPs in these areas.
3. Methods for Calculating Building Areas
From the principles of statistics, a sample is representative of and reflects the population, and population attributes can be estimated and inferred by analysing the sample drawn. Under the condition that the sample sites were evenly distributed in space and among administrative divisions, this paper proposes a relatively simple formula to calculate building area. The main considered factors in the formula are as follows: the building area can be quickly calculated after visually interpreting the RS imagery, a more accurate building area can be obtained, and others can obtain relatively accurate results using this formula in visual interpretation of RS imagery research. Based on these considerations, this paper proposes two equations to calculate urban, RUF, and rural building areas.
3.1. Method for Calculating Urban Building Areas
During visual interpretation, standards of different scales are used to extract building polygons from urban and rural areas. Considering factors such as the structural type, seismic fortification standard, and lateral stiffness of buildings [23, 24], urban buildings are classified into three types: low-rise buildings (i.e., buildings with 1–3 storeys), multistorey buildings (i.e., buildings with 4–6 storeys), and high-rise buildings (i.e., buildings with 7 storeys or more). Each building is treated as a zoning unit for extraction. The building area model is given as follows:where MC is the building area of an urban building, M is the polygon area obtained from the RS image, and λ1, λ2, and λ3 are the sample survey coefficients for low-rise, multistorey, and high-rise buildings, respectively. The areas of low-rise, multistorey, and high-rise buildings in urban areas are calculated separately using this equation with the help of the ArcGIS Field Calculator.
To accurately obtain a reasonable sample survey coefficient λi that can be used to calculate urban building areas across the study area, field sample surveys were conducted to collect data for low-rise, multistorey, and high-rise buildings. The numbers of storeys determined during the surveys were subsequently fitted to yield the corresponding value of λi.
Here, multistorey buildings are used as an example. Multistorey buildings at 88 sites were surveyed to determine their numbers of storeys, which were subsequently fitted to yield λ2 (5.32). Table 1 summarises the polygon areas, actual numbers of storeys, areas determined based on survey data (treated as the actual areas), and estimated areas at the survey sites. The polygon area referred to here is the area of a polygon obtained during visual interpretation, that is, the area of a building interpreted from remote-sensing images. The field-surveyed area was calculated by multiplying the number of storeys obtained in the field survey by the area of the corresponding polygon, and the estimated building area was calculated by multiplying the fitted multistorey building coefficient by the corresponding polygon area. Then, a linear correlation analysis was performed on the actual and estimated building areas. The reliability of the fitting coefficient was examined through correlation analysis. Figure 5 shows the correlation analysis results for the actual and estimated areas of the multistorey buildings at the survey sites. In the figure, the small blue dots represent the linear trendline, and the large blue dots represent the estimated building areas. An analysis of Figure 5 reveals a correlation coefficient of 0.92 between the areas of the multistorey buildings determined based on survey data and those estimated with λ2. This finding suggests that building areas estimated with λi are strongly correlated with those determined based on the actual numbers of storeys and that λi can be used to calculate urban building areas across the study area.
3.2. Methods for Calculating Areas of Buildings at Rural and RUF Sites
Most buildings in rural and RUF areas appear to be sheet-like structures in the RS images and are thus difficult to extract as individual polygons. In this study, buildings in rural and RUF areas are extracted based on block outlines. In other words, the outline of one block is extracted as one polygon, the area of which is denoted by M.where MS is the building area in a block in a rural or RUF area, βi is the regression coefficient (i.e., the ratio of the building area in the block determined based on survey data to the block area determined through visual interpretation; i = 1, 2, and 3 for SC, GS, and RUF buildings, resp.), and σ is the error compensation coefficient for the areas of standalone buildings (e.g., the building polygon areas indicated by B in Figure 6). The areas of SC, GS, and RUF buildings are calculated separately using this equation with the help of the ArcGIS Field Calculator. The polygon area here refers to the area obtained by visual interpretation. That is, the area is one block interpreted on the RS image. The definition of the survey point in the RUF area is consistent with the sample point in this area.
Here, the method used to calculate the regression coefficient for uniform GS buildings, β2, is described. β2 is the average ratio of the GS building areas determined based on field survey data to the corresponding polygon areas, that is, the proportion of the area of a GS building block that is occupied by all the buildings within the block (see Figure 6). To ensure accuracy, the obtained value of β2 is validated based on survey data obtained at 88 sites (see Table 2). Figure 7 shows the results of the statistical error analysis. In the figure, the blue dots indicate the percentages of building area, and the short, vertical lines are error bars, each with a standard deviation of 1. The two horizontal lines above and below the error bars correspond to 40% and 20% of the total building area, respectively. The standard deviation statistical analysis indicates that the β2 value is within a reasonable range. The same method is used to calculate β1 and β3, and thus the process is not described again.
4. Result Analysis
4.1. Estimated Area Correction
Normal statistical analysis was conducted to estimate the frequency distribution with a known mean and standard deviation. To verify whether the accuracy of the building areas calculated by equations (1) and (2) met the requirements, another 88 sample sites were selected from 22 districts and counties across the entire Ningxia area for field surveys to verify the reliability and accuracy of the calculated building areas. The results indicated that the calculated building areas had an accuracy of 90%. The results of normal statistical analysis are shown in Figure 8.
The main influencing factors limiting the accuracy are as follows. First, the accuracy of visual interpretation varies among people. Second, some of the RS images are satellite images captured in the second or third quarter, which makes building boundaries difficult to interpret because of the occlusion of trees, thus resulting in an inaccurate acquisition of polygon boundaries.
4.2. SP Estimates
The building SPs in all the CLDs of Ningxia were preliminarily determined by visually interpreting the RS imagery. Subsequently, at least one sample site was surveyed in each village and township in all 22 CLDs of Ningxia. During field surveys, the building SPs at the survey sites were determined based mainly on three factors: (1) the SP data for the survey sites obtained from the local construction authorities, (2) the external appearance and structural features of the buildings (i.e., whether the buildings were equipped with seismic structures such as structural columns and ring beams), and (3) feedback from the occupants with respect to whether seismic measures were implemented during construction. The error correction of the original SP results was carried out based on the field survey results. The SP accuracy in Table 3 refers to the percentage of the sample sites in which the SP based on visual interpretation and the SP based on field survey data at the sample site are consistent. Here, the sample site is a zoning unit displayed by a building in the urban area or a block in the rural or RUF areas on a remote-sensing image. The results contain the following information. The SP accuracy is greater than 80% for Yuanzhou District, Xiji County, Delong County, Jingyuan County, Pengyang County, Haiyuan County, Hongsibu District, Tongxin County, Qingtongxia City, Litong District, and Lingwu city (50% of all the CLDs). The SP accuracy is greater than 70% for 20 (91%) of the 22 CLDs. The SP accuracy is considerably higher for the CLDs in southern Ningxia than for Xingqing, Jinfeng, and Xixia Districts and Yongning and Helan Counties under the jurisdiction of Yinchuan City and Pingluo County and Huinong and Dawukou Districts under the jurisdiction of Shizuishan City. Based on the field survey results, this significant difference can be mainly ascribed to the following factors. First, buildings in the CLDs of northern Ningxia are complex and varied. Here, seismically fortified buildings alternate with unfortified buildings. As a result, interpreting the corresponding RS images is difficult and prone to errors. Figure 9 shows the distribution of SP accuracy for the buildings across the study area. Second, the RS images do not have a high spatial resolution; thus, some building SPs are difficult to accurately determine. Third, for the RUF buildings, using block-based acquisition in SP determination tends to ignore buildings with a certain SP that account for a relatively small proportion within a block.
4.3. Distribution Pattern of Building SPs
Based on the area and SP estimates, ArcGIS spatial analysis was used in this study to analyse the correlations between spatial location and SP attributes of buildings throughout the Ningxia region. The basic process of ArcGIS spatial analysis is as follows.
Spatial join analysis is a method that joins attributes from one feature to another based on the spatial relationship. Target features and the joined attributes from the join features are written to the output feature class. We used “many-to-one” to associate the join features with the target features. The join features were buildings, and the target features were CLDs (consisting of districts, county-level cities, and counties). The Match Option of different buildings in one CLD was set to the CLD.
Overall, the proportion of buildings with the required SP in each CLD is higher in northern Ningxia than in southern Ningxia, while the proportion of the buildings with substandard SP in each CLD is appreciably higher in southern Ningxia than in northern Ningxia, as shown in Figure 10. An analysis of the field survey results identifies two factors primarily responsible for the high proportion of buildings with substandard SP in southern Ningxia: (1) The efforts of nearly four decades of resettlement and nearly a decade of targeted poverty alleviation have improved the living environment and rural buildings in southern Ningxia. However, earlier government poverty alleviation efforts were mainly dedicated to substantially ameliorating rural buildings in terms of safety and comfort and failed to consider their SP. Most rural buildings with the required SP have been constructed in the last decade. (2) New and old buildings generally coexist in rural areas, presenting a certain challenge to the determination of building SPs and, therefore, the classification of buildings based on their SP.
Figure 11 shows the overall distribution of the buildings with the required SP, suspected substandard SP, and substandard SP, as obtained using the ArcGIS spatial analysis method; the figure reveals the following features. Buildings with the required SP are predominantly clustered in urban areas, with only a few distributed in villages and townships. Buildings with a suspected substandard SP and substandard SP are distributed mainly on the outskirts of urban areas and in the residential areas of villages and townships. This finding conforms to the field survey results. An analysis of the building SPs across Ningxia determined based on visual interpretation reveals generally high levels of SP in urban buildings and relatively low levels of SP in rural buildings.
5. Conclusion and Discussion
RS technology can be used to examine the SP and seismic vulnerability of building groups. Assessing the SP of buildings before earthquakes can save time and labour and produce timely data. This study presents a set method for extracting building area attributes and determining the SP of buildings that involves identifying building features in high-resolution RS images and statistical analysis of field sample survey results. The conclusions of this study are summarised as follows:(1)Urban buildings are first classified based on the number of storeys into three types: low-rise buildings (i.e., buildings with 1–3 storeys), multistorey buildings (i.e., buildings with 4–6 storeys), and high-rise buildings (i.e., buildings with 7 storeys or more). Each building is treated as a zoning unit for polygon identification. The values of the fitting coefficients λ1, λ2, and λ3 are determined based on sample survey data. Subsequently, the areas of low-rise, multistorey, and high-rise buildings are calculated by directly multiplying their polygon areas by λ1, λ2, and λ3, respectively. Correlation analysis reveals a strong correlation between the areas estimated using this method and determined based on sample survey data, suggesting that this method can be used to calculate urban building areas.(2)For rural and RUF areas, the outline of a block is treated as a zoning unit for visual interpretation. An equation is established to estimate building areas. Building areas estimated with βi and subsequently checked and corrected based on field sample survey results follow a normal distribution.(3)The SP estimates are validated based on data collected at one or more sample sites in each village and township in all 22 CLDs of Ningxia. The results show an SP accuracy greater than 80% for 50% of the CLDs and an SP accuracy greater than 70% for 20 CLDs, suggesting that the SP estimates are satisfactory.(4)Based on the SP estimates, the ArcGIS spatial analysis method was used to determine the distribution of the buildings in the three SP groups across Ningxia. The results reveal the following features. Buildings with the required SP are clustered in the urban areas of each CLD, with a few distributed at the seats of the village and township governments. Buildings with a suspected substandard SP are distributed predominately in RUF areas and at the seats of the village and township governments. Buildings with a substandard SP are clustered in urban villages and RUF and rural areas. The field survey results lend concrete credence to the correctness and reliability of this analysis.
To summarise, the methods for determining building SP and calculating building area proposed in this study provide new ideas and approaches to solving the problem of identifying building SPs on a large scale. Ningxia is an earthquake-prone region that has experienced numerous destructive earthquakes, all of which caused many casualties and much damage. The available findings of seismic activity tectonic studies indicate that the region faces a high seismic risk. The results of this study play a guiding role in the seismic strengthening of buildings in the region by screening buildings with substandard SP.
In addition, despite the extensive research on applying RS technology in a postearthquake disaster assessment, few studies have applied RS images to evaluate preearthquake building SP on a large scale. This study has the advantages of high output efficiency, low input cost, and the ability to carry out a large-scale evaluation of building SPs. The research findings enrich the applied research on RS technology in the prevention and control of preearthquake risks and provide new ideas and new means for evaluating building SPs.
RS images are the basis for this analysis. To produce reliable results, RS images obtained during the same period that have sufficiently high resolution and are representative of the current situation should be used for the same region. The ages of buildings in a region can also be approximated based on their polygon features, positions, and spatial combinations in high-resolution RS images in conjunction with the regional characteristics to improve the accuracy of SP estimates. An overlay of the temporal and spatial distribution patterns of the population and buildings in a region can be used to quickly assess the impact of earthquakes of different magnitudes, such as potential economic losses and the number and distribution of casualties. All these areas warrant in-depth discussion and investigation.
Data Availability
The data of the remote-sensing images used in this study are from GF-2 satellite.
Conflicts of Interest
The authors declare that they have no conflicts of interest.
Acknowledgments
This study was supported by the (1) National key R&D Program of China (no. 2017YFB0504104) and (2) the First National Survey on Natural Disaster Risk (the Project on Seismic Risk Investigation and Elimination of Major Hidden Dangers of Ningxia).