2.1 Vulnerability matrix

The damage to buildings is impacted by many factors, such as hazard intensity and building vulnerability (Papathoma-Köhle et al., 2015). Generally, intensity parameters such as impact force, velocity, and deposition depth are derived from numerical simulations or field investigations. Building vulnerability is obtained by the matrices, curves, and indicators approach (Papathoma-Köhle et al., 2017). All the methods require a detailed investigation of specific watersheds and settlements, which is not suitable for large areas due to a large amount of work and high cost.

The relative distance, including the horizontal distance (HD) and vertical distance (VD), of buildings to debris flow channels, reflects the debris flow intensity, such as the velocity and impact force, at a given position (Akbas et al., 2009; Hürlimann et al., 2008; Jakob et al., 2012; Lo et al., 2012; Papathoma-Köhle et al., 2012). The building wall distance to a torrent channel greatly influences the impact force of fluviatile sediment transport. The flow velocity and depth near the channel are generally greater than those farther away, which may result in a high impact force on a building (Sturm et al., 2018). Without considering the inherent building characteristics, a building with a small VD to a channel is more likely to be destroyed. Therefore, in the absence of debris flow parameters and detailed characteristics of buildings, the distance of a building from a debris flow channel can be adopted as a relatively ideal indicator to identify those buildings susceptible to debris flows across large-scale regions containing building patches. However, scientific standards are lacking in regard to the safe distance of buildings from debris flows. Du et al. (2015) conducted an experimental study of the interaction between building clusters and flash floods. Their experiment indicated that the safe VD of a building from a river channel is 13 m for buildings in inner gully areas, while the safe HD is 80 m under extreme flood conditions.

To examine the distance characteristics of damaged buildings in a specific settlement, numerical simulation of damage-causing mechanisms on the buildings exposed to debris flow is necessary. However, the analysis result is difficult to apply in other catchments due to the different characteristics of debris flow. Frequency–magnitude distributions quantify the relative probability of hazard events of various sizes, and are widely utilized in debris flow hazard evaluation (Hungr et al., 2008; Riley et al., 2013; Stoffel, 2010). In the study, a series of historical debris flow events (Figure 1) were analyzed to obtain the characteristics of damaged buildings and debris flow. These debris flow catchments are tectonically active and landform undulates greatly. The climate is an eastern monsoon climate, long-duration heavy rains, and short storms occur frequently during the wet season. The complex active geological structure, steep terrain, and abundant precipitation make these catchments the most susceptible to debris flow disasters in Southwest China. Data on 362 residential buildings destroyed by 23 debris flow events in Southwest China from 2003 to 2020 (Table 1) were compiled based on field investigation and aerial photograph interpretation. Eighteen catchments are located in Sichuan Province and the remaining five are located in Yunnan Province (shown in Figure 1 as red dots). The 23 catchments are mainly distributed in the Minjiang River basin, Ddu River basin, Jinsha River basin, and Nujiang River basin, which are most susceptible to debris flows in Southwest China (Wei et al., 2021). The majority of damaged buildings were in concentrated rural settlements, 59% of the total damaged buildings were constructed on debris fans, and the remaining 41% were located near the upper channel above the debris fan. Therein, 131 buildings were severely damaged by a catastrophic debris flow that occurred in the Qipan Valley (Wenchuan, Southwest China) on July 11, 2013 (Hu & Huang, 2017; Zhu et al., 2015). The peak discharge values of 13 debris flow events were retrieved from technical articles and government research reports. Linear regression was performed to determine the relationship between the peak discharge and maximum building HD. The F and T tests were adopted, and the linear regression results are shown in Figure 2. The results reveal that the p value of the F and T tests is much lower than 0.001 and 0.05, respectively, and the R² value is 0.823, indicating that a high peak flow discharge results in a large destruction area. Generally, a high peak flow discharge is associated with a high velocity and narrow flow section, thus resulting in extensive damage. The positions of buildings relative to the debris flow channel (the smallest and largest HD between a building and the channel centerline) were statistically analyzed. The largest distance of a building destroyed by one of the studied debris flows was 260 m in the Wenjia Gully (Yu et al., 2013). The frequency and cumulative frequency distribution of the HD are shown in Figures 3 and 4, respectively. The distance was divided into several intervals in Figure 3. The proportions of buildings in different intervals vary and buildings at distances between 15 m and 30 m take the largest proportion. Buildings at distances of less than 60 m accounted for 73.5% of the total buildings. According to Figure 4, the distances of 50% of all buildings were less than 30 m, and the distances of 80% of all buildings were less than 80 m. In contrast, buildings at distances greater than 100 m only accounted for only 16.7% of the total buildings, and all these buildings were destroyed by a debris flow that occurred in the Wenchuan earthquake region (Tian et al., 2011). In most cases, the damage degree of a building exhibits a positive correlation with its distance to a channel due to the relatively high impact force. For example, among the 362 buildings, most buildings with a distance of less than 30 m were completely destroyed by the debris flow, and these buildings always sustained structural damage or collapsed and could not be repaired. Most casualties and massive property losses occurred in these cases. In contrast, most buildings with distances greater than 100 m were partly inundated by debris sediments with a shallow depth, resulting in a high survival rate for the people in these buildings. Hence, the ratio of the building distance could reflect the degree of loss resulting from debris flows. The cumulative frequency distribution of the HD could be adopted as a classification indicator for buildings vulnerable to debris flows. Therefore, we selected the HD as an indicator of the building vulnerability. For the regional forecasting model in risk management, the probability is usually divided into a series of reference intervals, such as (0, 20), [20, 50), [50, 60), [60, 80), and [80, 100] (S. Zhang et al., 2018), different grades of warning are defined according to the preselected intervals. The value at the probability of 50% is regarded as the key for forecasting (S. Liu et al., 2020). In this study, the probability was divided into five intervals, namely (0, 50], (50, 80], (80, 90], (90, 99], and (99, 100). The distance values corresponding to cumulative frequencies of 50%, 80%, 90%, and 99% (at 30, 80, 130, and 240 m, respectively) were employed as threshold values in HD classification. Compared to the intervals divided by S. Zhang et al. (2018), the intervals divided in the study were not uniform and decreased gradually. The division method has a particular emphasis on the first and second intervals, and more buildings with high vulnerability may be involved in this method than in the uniform division methods.

Additionally, the VD between a building and debris channel is also pivotal for the safety of the building. Therefore, the VD of a building was chosen as the second indicator of the building vulnerability. However, the pre-disaster VD of buildings cannot be accurately determined due to a lack of pre-disaster terrain data. The debris flow depth was selected as a substitute to determine threshold values of the VD of buildings. The maximum debris flow depths in the main channels during 26 events in Sichuan Province were statistically analyzed (Table 2). The 26 debris flow catchments are distributed in the Minjiang River basin, Dadu River basin, Tuojiang River basin, and Fujiang River basin (shown in Figure 1 as green dots). All the depth values were obtained by field measurements of the mud depth of the debris flow near the alluvial fan or upper channel. The average flow depth was 3.9 m, and depths equal to or less than 5 m accounted for 80% of the total events, excluding the depth in curved channels. The debris flow depth often sharply increases in a curved channel, as exemplified by the curved channel with at depth greater than 10 m observed in the Tangjia Valley (Xie et al., 2013). Based on the above analysis, 5, 10, and 15 m were chosen as the threshold values in VD classification.

Finally, a vulnerability matrix for the identification of buildings susceptible to debris flows was proposed based on the above threshold values (Figure 5). Based on the common classification methods of buildings vulnerability concerning debris flows (Fuchs et al., 2007), the building vulnerability level was mainly divided into four categories, corresponding to very high, high, moderate, and low levels. Moreover, the very high, high, and low vulnerability levels were further divided into two categories, and a total of seven levels were involved in the vulnerability matrix. The subdivided methods in this study provided a more meticulous classification of vulnerability level, these specific subdivided levels make the assessment result less vague and are more convenient for practical applications.

As such, the vulnerability of buildings with HD ≤30 m and VD ≤10 m is at a very high level. The vulnerability of buildings with HD > 30 m and ≤80 m and VD ≤10 m is at a high level. Buildings with HD ≤ 30 m and 10 m < VD ≤ 15 m or 80 m < HD ≤ 120 m and VD ≤ 5 m are classified as occurring at the moderate vulnerability level. Buildings with HD ≤240 m and VD ≤15 m are classified as exhibiting a low or very low vulnerability level.

2.2 Building identification

2.2.3 Distance calculation and building identification

The process of distance calculation is shown in Figure 6 based on the GIS. First, the proximity tool in ArcGIS was adopted to calculate the HD between the building and debris channel. The nearest-neighbor distance was calculated between the polygon (building) and polyline (channel). Moreover, the corresponding nearest-neighbor points of the buildings on the polyline were extracted. Second, the building polygons were converted into points inside the boundary through the Feature to Point tool. The elevation values of the building points and corresponding nearest-neighbor points were then extracted from the DEM. Subsequently, the difference between the above two elevation values was considered the VD between the building and debris flow channel. Finally, the vulnerability levels of the buildings were determined according to the classification matrix, as shown in Figure 5. A final map of the buildings at the different vulnerability levels was generated for all debris flow catchments.

3.1 Physical setting

Puge County is located in the southwestern part of Sichuan Province, China. Its geographic scope includes an area of approximately 1918 km² with longitudes of 102°26′E and 102°46′E and latitudes of 27°13′N and 27°30′N (Figure 6). The elevation in the area ranges from 1400 to 4359 m above sea level (a. s. l.) with a relative height difference of 3319 m. Areas with slopes between 10° and 30° account for 62.25% of the total area. The area spans the two secondary tectonic units of the Kangdian platform uplift and the Dianqian syncline, largely consisting of a series of north–south folds and compression and transtensional faults. The Zemuhe and Heishuihe faults are two typical structures (Figure 7). Quaternary unconsolidated deposits mainly occur along the river, and clastic rocks, including sandstone and mudstone, primarily outcrop in the northern area. Carbonate rocks, including limestone, dolomite, and marl, occupy the central and southern areas. Magmatic rocks are mostly basalt distributed in the eastern area.

The climate of Puge County is controlled by the southwest monsoon and the dry continental air of northern India. The wet and dry seasons are distinct, and the annual range of the temperature is small, while the daily range of the temperature is large. Approximately 89.2% of the rainfall occurs between May and September, which is the period when most debris flow events occur in this area. The precipitation increases with the elevation, and the average annual precipitation reaches 1176.3 mm. The maximum rainfall in 1 h and 10 min amounts to 51.2 and 15.2 mm, respectively.

Puge County lies in the Jinsha River basin and is one of the counties in China that is the most affected by debris flows due to its active tectonic setting, abundant rainfall, and rapid urbanization. Sixty debris flow-prone catchments have been distinguished by the local government, and approximately 5630 people are at risk for debris flow hazards (Figure 6). The most recent debris flow disaster occurred on August 8, 2017, in the Tongzilin Gully, and this debris flow resulted in 25 fatalities and four injuries due to building collapse (Figure 8). Direct economic losses totaling 160 million RMB were caused by this event.

Puge County is a less developed county in China and the risk management of debris flows is still in the preliminary stage. Detailed records of historical debris flows are lacking in Puge County. Research on debris flow risk assessment in this area is scarce, and the preliminary identification of buildings prone to debris flows is mainly based on manual investigation.

3.2 Data sources and preprocessing

In this study, a DEM with a cell size of 10 × 10 m was adopted to extract the debris flow channels, and the DEM data were reprojected onto the Universal Transverse Mercator (UTM) zone 48 (WGS 1984) projection. A flow accumulation threshold value of 14,000 was applied for stream network extraction in the study area.

Gaofen-2 (GF-2) satellite images were employed for the extraction of buildings. GF-2 is a new-generation Chinese satellite equipped with sensors capable of resolving data at the submeter resolution launched in August 2014. Two barrel-mounted panchromatic cameras and two sets of multispectral cameras are assembled on this China-made satellite, which are collectively referred to as the Panchromatic and Multispectral Sensor (PMS). One 0.8-m-resolution panchromatic band and four 3.2-m-resolution multispectral bands are provided in the PMS data (Peng et al., 2019). Four GF-2 raw images acquired on February 27, 2019, two images obtained on April 7, 2019, two images captured on February 2, 2018, one image recorded on February 17, 2018, two images acquired on April 8, 2017, and one image obtained on February 2, 2017 were adopted in the study. The high-resolution panchromatic images and four-band multispectral mobile mapping system images were fused. Figure 9 shows two typical fused images obtained from the very large amount of data, and the two images exhibit high and low residential building densities.

3.3 Result

A total of 35,009 building polygons covering an area of 8.5 km² were extracted from the GF-2 satellite images. Figures 10 and 11 show examples of the extracted buildings in two settlements, and certain buildings in Figure 11 were segmented with hexagons.

The comparison between extracted buildings and actual buildings was carried out in 10 sample areas with a single area of 1 km². The results indicated that the correctness of the results exceeded 90%, and the completeness of the extracted buildings was higher than 85%. The extraction errors mainly occurred in the high mountainous areas. The accuracy in the alluvial fans was relatively high, while that in the high mountainous areas was relatively low due to the effect of vegetation and other factors. Most buildings extracted from the satellite images precisely conformed to reality. The buildings that were not successfully extracted from the imagery were mainly located in the high mountains, and these buildings are usually not easily destroyed by debris flows due to their large VD to the debris flow channel.

Careful scrutiny of the satellite images reveals that the settlements susceptible to debris flows are largely located near the debris flow channels at a distance of less than 1 km. To obtain the total area in all the settlements susceptible to debris flows, buffer zones with a radius of 1 km were plotted around the main channel using ArcGIS, and the total area of the buildings in these buffer zones was calculated. The results reveal that the area of buildings located in settlements near the main channels or debris flow fans reaches 2.1 km², which accounts for 24.8% of the total area of buildings in Puge County. These buildings are always constructed with a high density in communities, and some buildings have been impacted by debris flows or flash floods. Statistics of the areas of the buildings at the different vulnerability levels in all settlements prone to debris flows are listed in Table 3. The results reveal that approximately 10% of the buildings exhibit a very high vulnerability level, 15.53% of the buildings exhibit a high vulnerability level, and the buildings at the moderate and low vulnerability levels account for 5.71% and 13.47%, respectively, of the total buildings. The highest frequency is observed at the high (1) vulnerability level, and the next highest frequency occurs at the very high (1) level. The total area of the buildings at the very high and high vulnerability levels amounts to 539,316 m², which means that approximately 1078 buildings covering an area of 500 m² are recognized as being at a very high or high risk of being destroyed by debris flows.

Among all these settlements threatened by the 60 debris flow-prone catchments in Puge County, the communities located near 13 debris flow channels are the most vulnerable. The proportions of the areas of the buildings at the very high and high vulnerability levels in these settlements are higher than 50%. Moreover, the settlements located near six out of the 13 channels are the most vulnerable (Figures 12-15), and some have even experienced several debris flow events in the past.

Luomo village, as shown in Figure 12, is affected by the debris flow of the Niunaidu Gully and experienced four debris flow disasters in 1987, 1997, 2006, and 2016. The 1987 debris flow occurred at approximately 2:30 a.m. on July 2, which destroyed approximately 40 buildings near the forest protection station and resulted in one death. The buildings at the very high and high vulnerability levels were mainly located in the deposition area of the 1987 debris flow (Figure 12). Three debris flow events then occurred in 1997, 2016, and 2017; fortunately, these events neither cause very high property losses nor numerous casualties. Currently, debris flows still greatly threaten the lives and properties of approximately 80 households and 400 villagers. Deyu and Luobo villages, as shown in Figure 13, are two settlement centers located near Luoji Mountain, which is a well-known tourist attraction in Puge County. Two debris flow gullies, namely, the Qinshui Gully and Jialuobo Gully, which are located in the upstream and downstream directions, respectively, threaten these two villages. The majority of the buildings, including population centers such as government buildings, kindergartens, and middle schools, in Luobo village, occur at the high and very high vulnerability levels. The very high- and high-vulnerability level buildings in Deyu village are mainly supermarkets, hotels, and restaurants targeting tourists attracted by Luoji Mountain. Similar to Luobo village, the Huashan community, as shown in Figure 14, is a popular center of Luowu town, and an elementary school, a middle school, a post office, and government departments are located in this community. More than 50% of the buildings in the above community are prone to damage at the very high and high vulnerability levels. The buildings in Weilati village are relatively sparse, as shown in Figure 15, and approximately 50% of these buildings occur at the very high and high vulnerability levels. In recent years, several households in steep mountainous areas have relocated to debris flow fan areas, such as Luomo village, after the implementation of the policy involving the relocation of impoverished residents in China. Newly constructed buildings in debris flow fan areas are likely to increase in the future, which may result in risk management challenges. More attention should be given to these communities, and further investigation is imperative for the implementation of detailed measures.