From Remote To Social Sensing:Multi-Sensing Based Approaches To The Assessment Of Seismic Disaster Risk

A thorough understanding of risk is a prerequisite for efficient and effective risk management.Natural disaster risk is the product of natural hazards,vulnerability(exposure),and capacity to cope with risk(resilience).In terms of earthquakes,the changes in vulnerability and resilience are the most obvious part of earthquake disaster risk(EDR)in certain regions,especially in China,a country undergoing rapid social and economic development.The continual,rapid,and accelerating changes in vulnerability and risk have made it difficult to ensure that earthquake disaster risk reduction strategies for such an area remain fact-based,practical and effective over time.Certain strategies may become irrelevant after a short period.Methods to recognize changes in social vulnerability and its resilience scientifically and rapidly are needed to thoroughly understand the latest state of EDR and formulate more valid and time-sensitive policies for resilience enhancement.However,considering the rapid development of society and economics,using EDR assessment methods based on conventional on-site investigations,which are always time and labor consuming(e.g.,on-site investigations for collecting and/or updating detailed building data),is a significant challenge.Fortunately,the rapid development of Earth Observation(EO)technology and modern Information Network and Communication Technology(INCT),especially geographic big data collection and analysis techniques,including remote sensing and social sensing,has provided unprecedented hope for addressing the above challenges.On the one hand,from a multiperceptual/sensing perspective,remote sensing and social sensing allow multidimensional observations of the natural environment and social environment at multitemporal and spatial scales and on multisemantic dimensions.These sensing techniques generate a large number of new data resources and provide geographic big data with a large volume and substantial content for large-scale and rapid regional EDR assessment.On the other hand,with the improvement of computing processing capacity,artificial intelligence technologies such as machine learning and deep learning provide technical support for the efficient integration and comprehensive utilization of multisource geographic big data.A built environment refers to the basic surroundings that provide a setting for human activities,and it is extremely vulnerable to the effects of natural hazards.Rapidly collecting and regularly updating housing exposure and its vulnerability attributes are the most urgent issues and primary bottleneck questions in EDR assessment and management.These problems have always been a hot spot in both the EDR assessment practice realm and research.Therefore,based on a built environment in the Tangshan area,a special study entitled"From Remote to Social Sensing:Multi-Sensing Based Approaches to the Assessment of Seismic Disaster Risk"was carried out.This study is expected to contribute to the abovementioned EDR assessment challenges.The main contributions and findings are as follows.1.This dissertation summarized several main methods of EDR assessment,including those based on field investigation,comprehensive indices,and remote sensing.On the basis of the above summary,by learning and drawing lessons from the understanding of social sensing in the fields of computer science and human geography,the basic concept of social sensing was extended in this dissertation.In addition,the application progress and prospects of typical social sensing data and social sensing approaches in earthquake disaster research and EDR reduction practice were reviewed.Then,an overall framework for EDR analysis,which is characterized as being an in-depth,step-by-step framework,was established.This framework advances with time(i.e.,it makes use of new social data resources in a timely manner),as described in the"From Remote to Social Sensing"study.From the overall framework,the geometric shape parameter of houses based on remote sensing,the structure/function/age of urban houses based on passive social sensing and of rural houses based on active social sensing can be extracted,and the earthquake disaster loss can be estimated based on artificial intelligence combined with multiattribute vulnerability parameters.2.A method for extracting house exposure information and earthquake vulnerability attributes was developed based on the combination of remote sensing and building-relevant local knowledge(Br-LK).This method consists of two interdependent core steps:(1)using a single high-resolution optical remote sensing image to accurately and quickly extract the height and footprint area of a house on a large scale and(2)establishing the mapping relationship between the house height and the number of floors and the mapping relationship between the house height and the structure type based on Br-LK.Then,the number of floors was estimated,and the structure type was identified through the above mapping relationship.Next,the footprint area parameter was added to calculate the building area of each structure type.Using this method,the number of floors,structure types and total building area of residential and public buildings in the Tangshan area in 2009 were extracted.The results of the field survey showed that the overall extraction accuracies of the number of floors and structure type were 91.61%and 95.18%,respectively.The overall mean absolute percentage error(MAPE)of the extraction of the total building area was 2.99%.This method was proven to be time-saving.In the Tangshan research area with dense buildings(up to 469 km~2),after the high-resolution images and Br-LK were made available,it took 2 people approximately 10 days to extract the parameters required for the EDR assessment of all residential and office buildings.3.An“internet+”method for extracting house exposure information and earthquake vulnerability attributes was developed.To address the limitations of the high economic cost of commercial high-resolution remote sensing images and the insufficient accuracy of the abovementioned methods in identifying the structure types of“2-6”-story houses,the“internet+”method,which is based on the abovementioned remote sensing and local knowledge,was proposed,and abundant internet house-related resources were increasingly integrated.The“internet+”method consists of 3 interdependent steps:(1)extracting the floor area of houses based on free high-resolution Google Earth(GE)images;(2)comprehensively utilizing GE images,commercial websites on houses,volunteered geographic information(VGI),street views,online crowdsourcing data,and Br-LK to estimate the number of floors in a house;and(3)judging the type of house structure by comprehensively using commercial websites on houses,network crowdsourcing data and Br-LK.This method was used to extract the number of floors,structure type and total building area of residential and public buildings in Tangshan in 2015.The results of the field investigation showed that the overall extraction accuracies of the number of floors and structure type were 88.98%and 97.54%,respectively.The MAPE of the extraction of the total building area was 4.64%.The timeliness of this method is equivalent to the method based on the combination of remote sensing and Br-LK.4.A multisensory-based method of extracting house exposure information and earthquake vulnerability attributes was developed.In the traditional calculation of earthquake damage of a group of houses,only the structure type and total building area parameters of different types of houses are used.To enrich the parameters for the calculation of earthquake damage to buildings,on the basis of the“internet+”method combined with machine learning and deep learning,the attributes of house vulnerability to earthquakes were refined by deeply integrating multisource data such as remote sensing and active and passive social sensing data.This method was used to extract the detailed parameters reflecting the vulnerability of groups of houses to earthquake disasters.This method employs different strategies for urban blocks and rural blocks.(1)For urban blocks,this method mainly uses remote sensing,passive social sensing(e.g.,POI/AOI),and machine learning/deep learning.The extracted parameters include the footprint area,number of floors,function,structure type,construction age,and total construction area of the house.(2)For rural blocks,the main strategy is to use active social sensing.In other words,by developing an app for rural housing investigation,the structure type ratio,construction age ratio,function,and full attribute information of typical houses(e.g.,single building attributes such as wall materials and structural column settings)were obtained.This method was used to extract the abovementioned attributes of vulnerability of all houses in the Lunan district of Tangshan,in 2020(except for the attributes of typical single buildings in rural areas).The results showed that the overall extraction accuracy of the urban housing function on the sample test set was 87.67%.In contrast to the results of a field survey in one randomly selected rural block in the study area,the relative error of the total area of rural houses was approximately 4.13%,and the difference in the proportion of structure type and construction age was less than 7%.This method was proven to be efficient and effective.Because a large number of training samples were generated,which is required for deep learning and machine learning through passive social sensing(such as POI/AOI),this method saves a large amount of time and human resources compared with field investigation approaches to obtain samples.For dense buildings and various types of houses in the Lunan district(62 km~2),after high-resolution image and positive and passive social sensing data were made available,it only took approximately 15 days for two people to extract the multiattribute parameters of all houses.5.A method for EDR assessment based on machine learning,which incorporates the multivulnerability attributes of houses,was developed.On the basis of various attributes of house vulnerability(including the number of floors,structure type,function,construction age,and total construction area of the house)obtained above,a large amount of earthquake damage prediction data of houses was used as the training dataset.Then,a random forest-based earthquake damage risk analysis method for houses was proposed.On the sample dataset,the overall test accuracy of the results from the random forest model was 98.57%.Through questionnaires and interviews,the parameters for calculating the direct economic losses and casualties caused by the destruction of houses,including the unit cost of houses with different structural types,the population density of houses with different functions,and the occupancy rate,were determined.In the context of a hypothetical and preset magnitude earthquake in the Lunan district of Tanshan,this method was adopted to calculate the damage to houses,direct economic losses,and casualties and numbers of people who need to be relocated.The results of EDR assessment showed that the houses in Lunan District meet the earthquake resilience-establishment requirements of"Not damaged in small earthquakes,Repairable in moderate earthquakes,and Not collapsed in large earthquakes".To summarize,an overall framework for EDR analysis,which is characterized as being an in-depth and step-by-step framework that advances with time,as described in the"From Remote to Social sensing”pathway,was proposed in this dissertation.Taking the EDR of buildings as an example,a series of extraction technologies for building exposure and its attribute parameters of vulnerabilities were developed based on the fusion of multisource data such as remote sensing and social sensing data.Then,a method for assessing the EDR of buildings based on machine learning was developed.Finally,a series of technical methods were developed for the rapid analysis of the EDR of buildings.On the basis of rapid social and economic development,the proposed methods are expected to provide methodological support for the analysis of large-scale EDR and its changes as well as insights into regional earthquake disaster risk reduction plans,earthquake emergency plans,and earthquake insurance policies.With the development of various relevant new social resources,technologies and methods,the technical method system guided by the“remote sensing-social sensing”pathway proposed in this dissertation may have broad development prospects.In addition,technical concepts based on the integration of"remote sensing and social sensing"can also be widely used as postearthquake rapid loss assessment and emergency response references.