This work investigated three topics in the reviewed studies, each aiming to answer a research question (RQ) or a subquestion as stated below next to each subsection:
The following subsections are organized to elaborate on these topics and answer the research questions, respectively. However, it is worth highlighting some general findings of this review work before discussing the research questions.
There were no exclusion criteria set on the type of studies or the type of mobile crowdsensing-aided disaster management solutions to be included in this systematic review. The solutions proposed could be a mobile application, a system architecture or a framework. Any type of solution that employs mobile crowdsensing or makes use of smartphone sensors in the context of a disaster seeking crowd involvement is included in the scope of this work.
The majority of the studies (12 papers) propose a
System Architecture: Frommberger and Schmid [
29] present a system architecture for an integrated disaster alerting and reporting system composed of an Android app, web interface, and disaster management server named Mobile4D. Visuri et al. [
30] present a two-tier system for building collapse detection in earthquakes that uses residents’ smartphones as distributed sensors. Anagnostopoulos et al. [
31] propose a Four-Layer system integrating crowdsourcing, crowdsensing, and an LSTM (long short-term memory) inference model for municipality resource allocation. Bhattacharjee et al. [
32] present a post-disaster map builder system including trace management and map inference modules utilizing GPS sensors in smartphones. Asiminidis et al. [
33] perform an empirical study presenting a Bluetooth-based cheap and autonomous system for indoor localization determination that is tested for a fire emergency scenario in a motorway tunnel. Piscitello et al. [
34] propose a system architecture for the detection and management of emergencies in smart buildings which is called Danger-System, composed of a DangerCore server structure and mobile application with two interfaces for building managers and tenants. Sadhu et al. [
35] discuss the architecture for a real-time 3D map** system of the disaster scene where data collection is made cooperative through crowdsensed data from bystander agents. Sarbajna et al. [
36] introduce a decentralized map** service relying on a blockchain backend to generate a complete, current, and accurate graph of accessible paths in a disaster-affected region to ensure any responder can contribute to the navigational map and everyone works on the last version of the map. Vahdat-Nejad et al. [
37] present an information gathering system for earthquake disasters which is composed of sensing, fog, cloud, and application layers and equipped with data collection, processing, and storage capabilities. Villela et al. [
13] propose a conceptual design of the RESCUER system, a smart and inter-operable decision support system that uses crowdsourced information in emergency and crisis management. Salfinger et al. [
38] present a situation-adaptive system design capable of exploiting both conventionally sensed data and unstructured social media content. Gao et al. [
26] discuss a conceptual system architecture of groupsourcing to facilitate efficient collaborations among various organizations responding to a disaster incident.
4.1. Use of Smartphone Sensors in Addressing Disaster Management Categories
This section aims to answer the research question 1-a. With the advancement in sensor technologies, there is an ever-increasing number of sensors built into smartphones such as GPS, camera, microphone, accelerometer, gyroscope, barometer, pedometer, light and proximity sensors [
49]. These sensors collect information on visualization (i.e., camera), localization (i.e., GPS, bluetooth), directionality (i.e., gyroscope), or mobility of objects (i.e., pedometer, accelerometer), which provide valuable information at a low cost in disaster management.
During the review process, it was observed that the
crowd as reporters phenomenon, where users report on a situation using the sensing capabilities of smartphones and the report gets verified by an official, appears as a frequently used sensing solution. This idea adopts the concept of smart citizens for smart cities crowdsensing [
50]. Smart citizens, or
crowd as reporters as stated in this paper, are believed to be major drivers of smart cities and they have been increasingly active in sensed data generation through smart city applications [
50]. Ludwig et al. [
23] define the
crowd as reporters as users with a concise and conscious use of existing knowledge to achieve a specific task or goal. Kamel Boulos et al. [
22] also refer to “human sensors” or “human-in-the-loop sensing” concepts to define the increasingly active role of humans in the sensing environment. Although this study initially intended to focus on smartphone-based sensors, after reviewing several papers that have been retrieved through the specific search query, it is concluded that human intervention still appears to be vital in disaster and emergency scenarios. Moreover,
crowd as reporters stands as a representation of the participatory sensing of mobile crowdsensing. Hence, the
crowd as reporters notion is recognized as a type of sensor in this review. The
crowd as reporters sensor is used in 14 of the studies which is a considerably high number. The use frequency of the sensors in the reviewed studies is demonstrated in
Figure 7.
Throughout the course of this review, disaster management problems that the reviewed studies have commonly addressed are noted and classified under eight categories such as evacuation/map**, hazard/risk detection, organization of rescue teams, data fusion, information exchange, situational awareness, efficient data transfer, and resource sharing/allocation. The context and use of sensors in addressing each disaster management category are discussed for each paper in the remainder of this section.
Figure 8 illustrates the number of reviewed papers addressing each disaster management category. The map** of the sensor types against the disaster management categories that are addressed by each sensor type is provided in
Figure 9. The results are also reported in a table format where citations are included per each sensor type and disaster management category addressed
Table 3.
Hazard/Risk detection: Detecting the hazard or risk in the disaster-affected area is the tip** point of rescue and relief operations [
51]. Six studies address this problem. Visuri et al. [
30] propose a building collapse detection system that uses end-user smartphones as distributed sensors accompanied by a rule-based fall detection algorithm. It is an empirical study with an evaluation of a fall detection algorithm through lab simulations and a limited field test. The main sensor used in this solution is the accelerometer and efforts are made to detect false-positive cases. Accelerometer values from devices falling on a soft or stiff surface are measured with a higher true positive rate for drops on stiff surface (98.7% vs. 95.6% for soft surfaces). An interesting finding is that jum** with phones in hand or in pocket trigger less than 1% false positive earthquake event, while running with the phone does not trigger any false positive event. A key limitation of this study is the assumption that multiple mobile phone falls indicate a building collapse. More information on the environment should be collected through smartphone sensors to ensure that there is a building collapse. Choi et al. [
47] discuss identifying major damage locations and types of the incident at the damaged site through crowdsensed image data using clustering algorithms. The study presents a cloud-based data collection, processing, and analysis process that employs a mobile crowdsensing application. This empirical study uses Icheon-si and Anseong-si rainfall 2020 data to test effective detection of incident type when image data are collected from a smartphone during an emergency. Camera and GPS sensors are used but more sensors could help to detect dynamic situations. Motivating users to share information is a challenge of this solution, however, receiving analyzed content or participating in response activities are discussed as motivating factors for users. Kielienyu et al. [
45] generate MCS-driven community risk map** to predict and prepare for future COVID-19 cases. This empirical study collects GPS signals from citizens’ smartphones and obtain their mobility patterns to estimate future movements of the detected communities and calculate a risk factor for communities ahead of time. Projected heatmaps for COVID-19 risks of the communities are good contributions to support Public Health departments’ resource allocation. However, privacy concerns and lack of incentive mechanism are shortfalls of this work. A similar study was conducted by Simsek et al. [
52], providing an MCS-enabled framework for risk map** of communities based on GPS data of smartphones and empowering autonomous vehicles to respond to the public safety needs.
Zabota and Kobal [
41] discuss a new methodology to enable quick and simple on-field data collection for past rockfall events through mobile crowdsourcing. A mobile application (Collector for ArcGIS, which is part of the Esri ArcGIS platform) is designed for data collection and WEBGIS platform is used for visual Web maps. A GPS sensor is used to obtain location of past rockfalls and deposits, a camera sensor is used to obtain additional attributes such as size, dimensions, etc. The data are reported into the app by the crowd, so crowd as reporter can be noted as another sensor used in this work. Li et al. [
40] discusses structural health monitoring, i.e., regional building damage assessment, through mobile crowdsensing. An earthquake strike in a city is simulated in ’Unity’ simulation environment where ‘Ground Eye’ acts as the city brain to coordinate emergency response and assign tasks to citizens to collect data on damage assessment. Acceleration is measured by the accelerometer sensor, strain, and inter-story drift are captured by camera sensor. Smartphones have to be installed on the damaged structure by the citizens to capture the time history data for all three parameters. Information exchange between citizens is overlooked in this study. Burkard et al. [
42] present image-based measurement methods to feed into a conceptual flood prediction system that can be used in rivers. The methods are based on three variants of inclination, reference points, and correspondence points, which are measured using the camera and orientation sensors (accelerometer and gyroscope) of a smartphone. The methods were evaluated in a demo application for Android phones, however, the conditions were assumed to be ideal. Variant targeting correspondence points provides the most accurate prediction given that the camera image taken captures four of the previously defined reference points. The implementation of these measurement methods in a real-life setting with real users remains a big challenge since the evaluation has only been performed under ideal conditions.
Evacuation/Map**: Evacuating the endangered area and guiding the victims to a safer place is one of the most important problems that need to be addressed during a disaster. In the aftermath of a disaster where the road networks are damaged and navigation through existing route networks is not possible, unconventional map** techniques are required to mobilize the resources in the affected areas. The evacuation/map** problem is one of the two most studied problems in the reviewed studies. Five studies discuss the evacuation/map** problem. Three out of five studies use GPS sensor data and two studies use Bluetooth signals. Bhattacharjee et al. [
32] propose a post-disaster map builder system on Android smartphones which includes trace management and map inference modules, where traces of users are collected through GPS sensors and fed into map inference module. Pedesterian maps are built by the map inference model through the collected trajectory traces. It is a conceptual study for which evaluation is performed through simulation and Testbed on ONE simulator and the Map** Toolbox of MATLAB 2017a. The key assumption of the study is that the map builder system has to be pre-installed on all node devices. This paper pre-processes raw trajectory traces for noise reduction, performs significant point identification, and uses clustering technique and Topology inference. This study achieves successful sending of 95% of trajectory traces to their destinations in two hours of generation, which is a quite promising contribution to the map** problem.
Asiminidis et al. [
33] propose a low-cost autonomous Bluetooth Low Energy (BLE) sniffing technique to guide and show the emergency exit upon receiving the RSSI (received signal strength indication) values from users’ smartphones and thus guiding the users instantly during an emergency in a tunnel. It is an empirical study presenting a Bluetooth-based cheap and autonomous system for indoor localization determination that is tested for a fire emergency scenario in a motorway tunnel. The solution uses the Bluetooth sensor to show the emergency exit to the users using their smartphones’ RSSI values and hence poses concerns regarding user privacy. Kitazato et al. [
39] discuss a system to detect real-time pedestrian flows including pedestrian congestion, direction, and velocity through mobile sensing with Bluetooth signals for evacuation route guidance during disasters. The study analyzes separately the congestion degree by detecting the surrounding Bluetooth devices and the direction and velocity of the pedestrians through the RSSI of a Bluetooth LE beacon carried by the pedestrian. Limited Bluetooth scope and battery limitation of smartphones can be a downside of this solution. Sarbajna et al. [
36] present a decentralized map** service relying on a blockchain backend, placed in a smartphone and available to everyone to generate a complete, current, and accurate graph of accessible paths in a disaster-affected region to ensure any responder can contribute to the navigational map and everyone works on the last version of the map. GPS sensor is used in this solution. The blockchain component of the solution helps with the accessibility of the system without a central authority need. It is a conceptual study with no evaluation, hence deployment in a real-world setting is needed to ensure the validity of this work. Fajardo and Oppus [
14] introduce an Android-based disaster management system (MyDisasterDroid) to facilitate the logistics for the rescue and relief operations determining the optimum route between volunteers and victims to serve the greatest number of people in the maximum coverage of the area. The solution uses a genetic algorithm to determine the optimum route and receives data from GPS sensors. It is based on Google’s Android system and assumes Google Maps navigation is not distorted, but it is very likely so in a flood or earthquake disaster.
Information exchange: It is possible to obtain comprehensive situational information in the shortest time through mobile crowdsensing—three studies contribute to information gathering systems. Frommberger and Schmid [
29] propose an integrated disaster alerting and reporting system architecture called Mobile4D for an integrated disaster alerting and reporting system composed of an Android app, web interface, and disaster management server. It is a conceptual study but the evaluation is performed through usability testing. This study tackles two important challenges: lack of institutionalization and lack of two-way communication. A top-down, bottom-up communication channel between authorities and disaster victims with an escalation procedure is designed. The system mainly uses the
crowd as reporters sensor where users report on the situation through the system where the report gets verified by an official; it also exploits other sensory information such as GPS for location detection and a camera for sharing visual information. The solution is functional under weak network conditions and is a text-free interface to avoid misinformation, however, improvements on information visualization were noted during the evaluation phase. It was tested with a small group of people and is only limited to Android Applications. Vahdat-Nejad et al. [
37] present an information gathering system architecture for earthquake disasters which is composed of sensing, fog, cloud, and application layers and equipped with data collection, processing, and storage capabilities. The system is designed as per the requirements analysis performed through interviews with Iran’s Red Crescent Society who have lived through the 2017 Kermanshah earthquake with 7.3 magnitude. GPS and camera sensors are used in the system design, however, the
crowd as reporters sensor is also embedded through user interface to compensate the shortfall of other sensors. Each information element is tagged with GPS coordinates to produce geotagged information maps, a camera is used to gather diverse environmental information that can either be processed by machines or reviewed by humans. Processing times for the shares images and texts are left as an area for future research. Villela et al. [
13] propose a conceptual design of the RESCUER system, a smart and inter-operable decision-support system that uses crowdsourced information in emergency and crisis management under the supervision of the command control body. A mobile app with a user interaction mechanism, data analysis capabilities, and views on relevant aggregated data is a part of the system. Two types of information exchange are proposed—participatory through crowd reporting and opportunistic through smartphone sensor data sharing consented to by the citizens. GPS, camera, and
crowd as reporters sensors are used and detailed data processing steps for each type of sensed data (text, image, video) are explained by the authors.
Situational awareness: Situation awareness is about providing the real-time situation of the disaster area to the stakeholders, i.e., citizens, volunteers, and disaster management authorities [
53]. Three studies address this problem. Piscitello et al. [
34] propose a system architecture to address an urban building crisis with a proposed system called Danger-System that creates a two-way communication between residents and building administrators leveraging both building infrastructure and mobile devices to create instant situational awareness. It is a conceptual study with scenario-based evaluation and the researchers created their simulator (DangerSystem simulator) in Python 2.7. It facilitates information gathering from residents through the microphone and pedometer sensors of their smartphones to detect any dangers in a building proactively, however, privacy and limited validation of data are key concerns of this solution. Sadhu et al. [
35] propose a smartphone-enabled system named Argus that generates a real-time 3D map of the disaster area including crowdsensed inputs from bystanders. Users are allowed to share data from camera, microphone, GPS, accelerometer, and gyroscope sensors of their smartphones to help with 3D reconstruction. It is a conceptual study simulating a fire scenario and evaluated through prototy**. This paper exploits the MARL framework (Multi-Agent reinforcement learning) for data collection that is implemented with a distributed Q-learning approach to direct the agents to capture data on the areas of interest. However, this framework is not evaluated in this paper. Salfinger et al. [
38] discuss a situation-adaptive SAW (situation awareness) system capable of exploiting conventionally sensed data and unstructured social media data and continuously optimizing itself through situational feedback loops. It exploits the CrowdSA framework and uses Hawaii Hurricanes 2014 dataset and General Disasters (Twitter) 2014 dataset. The
crowd as reporters sensor is mainly used in this solution. The proposed crowd-sensing enhanced SAW system architecture keeps track of the monitored real-world situation’s evolution and reuses the detected (and projected) situational context to optimize its crowd-sensing configuration and processing. Situational feedback loops are introduced in between processing levels to improve situational awareness. The system is tested on limited datasets so large-scale studies on various types of crises would be necessary. It is noted that to gain situational awareness of the disaster area, social media data and human input are still heavily consulted.
Efficient data transfer: During a disaster or a crisis, an extensive amount of information is shared either on social media or on conventional emergency platforms. However, it is highly critical to transfer data effectively and efficiently during a disaster, considering the limited bandwidth and intermittent connectivity [
54]. Honarparvar et al. [
55] address the energy consumption problem in wireless sensor networks where sensor nodes are located far from the base stations. The authors propose an integrated location based social network that can reduce energy consumption up to 42% through reduced routes to BSs. In our review, there are two studies that discuss efficient data transfer issue. Fahim et al. [
44] present a novel data-efficient framework to transfer very limited data to the central server and yet still detect global events or disasters with high accuracy. Only 1% of the crowd-sensed data are consumed for highly accurate detection of global events. Through parallelization of visual content, the average delay of content retrieval is reduced by 67%. This solution can help use the limited bandwidth and connectivity more efficiently and effectively during a disaster. Felice and Iessi [
46] introduce a software service to reduce processing times of Tweets during emergencies to speed up the analysis of post-earthquake on-site messages. Efficient data transfer in limited bandwidth and limited battery life is a key problem in mobile crowdsensing during disasters and it is a topic to be further explored to help increase the utilization of smartphones in emergencies. This empirical case study uses GPS and the
crowd as reporters sensors through the TwittEarth mobile app. Territorial data of Abruzzo region (center of Italy), an area repeatedly affected by destructive earthquakes, as provided by the Italian Institute of Statistics, and OpenStreetMap Data are used for this study. The study presents promising results in terms of fast data processing to identify exact damaged locations, however, the test sample was not very inclusive since a low number of Twitter users turn on location sensors to preserve privacy.
Resource sharing/allocation: Sharing and allocating scarce resources during a disaster is a major problem and is one of the key reasons for communicating in disasters [
56]. Two studies discuss this problem. Ae Chun et al. [
43] introduce the PEER Framework with a vision of a centralized database to facilitate resource sharing capabilities and an intelligent recommender system to match citizen resources with emergency tasks. No specific sensor type was mentioned except for the
crowd as reporters. The conceptual framework aims to support a comprehensive and unified disaster management system integrating social media channels and smartphones. It is evaluated through a Flood Warning Community System prototype, however, the prototype is not tested. Anagnostopoulos et al. [
31] try to solve the resource allocation problem of municipalities and propose a resource allocation system that can potentially be used for efficient disaster planning. The study uses a Greece Papagos–Cholargos municipality dataset (a smart city located in Athens, Greece) and utilizes Citify Crowdsourcing System Architecture as a data processing platform. The proposed system is composed of four layers. The first layer is the environmental crowdsourcing—the physical layer of the municipality where citizens are located and perform crowdsourcing activities. The second layer is the smartphone crowdsensing where citizens act as human sensors and annotate the environmental situation with the use of sensors. The third layer is the inference engine model which makes resource allocation (assigns problems to specific departments) possible and with the help of the LSTM classifier, the system can propose solutions for future cases. In the last layer, municipality headquarters’ staff work on the solution to the disaster recovery problem. The proposed system heavily relies on crowd reporters as a sensor. Spatial positional data, camera or audio data obtained through smartphone sensors are not further used by the system but instead they are used for annotation. Although this study describes a comprehensive system architecture, the evaluation is largely performed on the inference model and prediction accuracy. It could be an interesting topic to analyze autonomous resource allocation among distributed parties.
Organization of Rescue teams: Coordinating private rescue and relief activities during disasters became very common with the rise of social media [
57]. Two studies try to propose a more structured method for organizing rescue teams. Ludwig et al. [
23] introduce a web application called CrowdMonitor using mobile crowdsensing principles to align official emergency services with physical and online volunteer activities. The main objective of the study is to examine the potential of social media-generated information for situation assessment and at the same time, the potential for involving volunteers into the current work of emergency services avoiding duplications or conflicts in responding to authorities’ information requests. The crowd as reporter is the main sensor used in this study to report the requested information either on Open Street Map or in the mobile application. Gao et al. [
26] discuss an approach to facilitate efficient collaborations among various organizations responding to a disaster incident. This solution highlights a gap in other crowdsensing applications which is the lack of a central authority and a unified mechanism. Hence, the study suggests a central authority that will ensure data integrity and quality, and a subscription mechanism for other response groups which is named as “groupsourcing”.
Organizing rescue teams need a more structured approach than casual efforts on social media in order to prevent duplicate efforts, speed up responses and ensure reliability of shared data. The reviewed studies have taken a conceptual and qualitative point of view, without utilization of sensor data other than the crowd as reporters.
Data Fusion: Data fusion refers to the integration of heterogeneously sensed data from actual traditional sensors with the crowdsourced data to provide a more comprehensive understanding and an improved information structure [
58]. Two studies reviewed in this work discuss elements of data fusion in responding to disasters. Tripathi and Singh [
48] introduce C-Sense as a new paradigm of heterogeneous crowd sourcing and investigate the impact of training on volunteer participants’ contribution to disaster operations. The study presents a data fusion model of human virtual sensors and actual traditional sensors for disaster response. The crowd as reporter sensor is heavily used in this solution. Random and trained participants are evaluated for their contribution in the participatory sensing. This model envisions real-time data collection from various communication sources but the complexity of sensors appears as a challenge. Nguyen et al. [
27] introduce a novel mobile application design for integrating crowdsourced data collection and validation activities in disaster risk reduction processes. Heterogeneous data from ‘crowd as reports’ sensors as disaster reports or communication through notifications and related social media posts are integrated in the design. The key contribution is adding a validation mechanism for others’ reports by up-voting or down-voting to mitigate unreliable data from social media. The reviewed studies emphasize the significance of data quality and reliability in data fusion which is also one of the key problems of crowdsourcing. The value of data is a quite critical concept in prioritizing the data and hence enhancing the quality of fused data.
4.3. Guidance for Disaster Management Authorities
This section aims to answer the second research question (RQ2), which seeks to identify any guidance in the form of a framework, approach, or business model that is proposed in the reviewed studies to guide the disaster management authorities or organizations through the use of crowd sensed data in their decision-support systems. Out of 25 reviewed studies, only 4 studies are deemed to propose a process-based guidance for disaster management authorities to handle crowdsensed data in their operations.
There are existing studies proposing business models or frameworks on decision support for mobile crowdsensing for participant recruitment [
63], incentive mechanisms [
64], task allocation [
65,
66], optimum sensing coverage [
67], or task execution under budget constraints [
68]. However, what is sought in this study is more of a holistic and process-based guidance for disaster management authorities hel** them either to select the MCS-based disaster management solution that can address their needs, or to integrate the proposed solution into their existing decision-support systems.
Four criteria have been defined to determine whether a study also proposes guidance on how to integrate the solution in decision making while explaining the technicality of the solution:
Existence of a data flow process: Are the data generation, data collection, data processing, or data storage processes explained? Are there references on the server technology, communication infrastructure, data collection or generation platform, bandwidth requirements or data quality measures to let the authorities make realistic choices?
Definition of an information exchange mechanism: How does the information dissemination take place? Is there a two-way communication mechanism built in the solution between volunteers and disaster authorities or are there subordinate units?
Incorporation of human factors: Is human behaviour modelling or user feedback taken into consideration in the design or improvement of the solution?
Evaluation of the proposed solution: Is there any type of evaluation performed for the solution such as prototy**, simulation, or field testing?
The selected studies are reviewed for those criteria and only the ones that meet all four criteria are deemed fit to propose guidance to disaster authorities. Four studies are concluded to encompass the existence of a data process flow, an information exchange mechanism, the incorporation of human factors to design or fine-tune the solution, as well as the evaluation of the solution. The assessment results of the identified studies based on these criteria are summarized in
Table 4.
The majority of the reviewed studies focus on the data collection, generation, or processing steps of a specific solution. However, a holistic view of the full process from data generation to decision making is critical to speed up the adoption of mobile crowdsensing in disaster management.
It should be noted that no separate search query has been generated to identify studies proposing guidance for disaster management authorities. Most probably, there are other studies driven to provide guidance on the integration of crowdsensed data in current disaster management decision support systems. However, for this review work, only the studies selected through the main search query have been assessed for the existence of process-based guidance on the integration of crowdsensed data.