Stepwise Multisensor Estimation of Shelter Hazard and Lifeline Outages for Disaster Response and Resilience: A Case Study of the 2024 Noto Peninsula Earthquake

Abstract

Addressing earthquake risk remains a significant global challenge, requiring rapid assessment of evacuation shelters for effective disaster response. Existing frameworks, such as FEMA’s Hazus, Copernicus EMS, and UNOSAT, offer valuable insights but are typically regional, static, and event-focused, lacking mechanisms for continuous shelter-level updates. This study introduces the Shelter Hazard Impact and Lifeline Outage Estimation (SHILOE) framework. SHILOE is a stepwise estimation approach integrating multisensor datasets for time-scaled assessments of shelter functionality and operability. These datasets include seismic intensity, liquefaction probability, tsunami inundation, IoT-derived power outage data, communication network disruptions, and social media. Application to the 2024 Noto Peninsula earthquake showed that ≥93.6% of designated and activated shelters were impacted by at least one hazard, with all experiencing at least one lifeline outage. The framework delivers estimates through three phases: immediate (within tens of minutes, e.g., simulation-based hazard models and lifeline data), intermediate (days, e.g., observation-based datasets), and refinement (ongoing, e.g., Social Networking Service and detailed field surveys). By progressively incorporating new data across these phases, SHILOE generates dynamic, facility-level insights that capture evolving hazard exposure and lifeline status. These outputs provide actionable information for emergency managers to prioritize resources, reinforce shelters, and sustain critical services, thereby advancing disaster resilience.

1. Introduction

Earthquake risk constitutes a global challenge that threatens lives, infrastructure, and economies worldwide [1]. Effective disaster response necessitates not only rapid hazard assessment but also the maintenance of essential services that support affected populations. The status of lifeline services at evacuation shelters is particularly critical during large-scale disasters. Services like power, communication, and water supply are essential for determining a shelter’s functionality and its operability to sustain community resilience [2]. Lifeline networks are inherently complex and interdependent [2], comprising both structural and nonstructural components [3], which makes their monitoring and management particularly challenging. A rapid and comprehensive assessment of hazard impacts and lifeline services before, during, and after a disaster is essential for effective shelter management and sustaining community resilience.

Considerable progress has been made in employing diverse data-collection technologies and tools for post-disaster damage assessment. For instance, satellite imagery is widely utilized in the immediate aftermath of disasters to generate preliminary estimates of infrastructure damage [4]. Remote sensing offers broad-scale, efficient assessments across the spectral, spatial, and temporal domains [5]. Synthetic aperture radar is particularly advantageous because it functions regardless of weather or daylight. It can capture vertical surface changes [6,7,8,9,10], enabling detailed evaluations of the damage [10,11]. Nevertheless, available satellite imagery may not always provide sufficient resolution to capture building- or facility-level impacts. Furthermore, access to very-high-resolution data is often constrained by licensing restrictions and acquisition costs, limiting its operational use in near-real-time disaster responses [12,13].

Multisensor integration methods enhance the accuracy and robustness of damage estimates [14,15]. These methods combine data from optical, radar, aerial, and ground sources. However, these methods have several drawbacks. They require complex fusion workflows and consistent preprocessing. Additionally, they face challenges such as mismatched spatial or spectral resolutions, temporal alignment issues, and registration errors, as noted in recent reviews [16]. Machine learning techniques have also been applied to automate damage detection [17,18], enabling faster analysis and the identification of subtle damage patterns. However, their performance depends heavily on the availability of large, high-quality training datasets, and they may not be generalizable across different disaster contexts [19].

Additionally, emergency mapping frameworks have been developed to facilitate disaster loss estimation and damage assessment. These include the Federal Emergency Management Agency’s (FEMA) [20] Hazards U.S. Multi-Hazard (Hazus) loss estimation tool [21], United Nations Institute for Training and Research’s Operational Satellite Applications Programme (UNITAR–UNOSAT) [22], and Copernicus Emergency Management Service (Copernicus EMS) [23]. These frameworks standardize workflows [21,24], integrate diverse information sources, and generate actionable map products for decision-makers [22,23,25,26]. These systems provide valuable loss estimates and situational awareness; however, they are generally designated for regional- to national-scale applications and focus on single-event outputs rather than continuous, facility-level updates. Consequently, most existing damage-detection applications are limited to static and discrete-time assessments. Only a few studies have investigated the outcome of the continuous time-scaled integration of multisensor data through a stepwise framework. A “stepwise framework” incrementally refines damage estimates as new data becomes available over time—beginning with rapid, coarse assessments and incrementally integrating higher-resolution or complementary information to enhance accuracy [27].

In addition to technological advancements, researchers working in the fields of emergency management and spatial optimization have examined the siting and accessibility of emergency facilities. For instance, location–allocation and accessibility models have been employed to optimize the placement of shelters [28] and medical facilities [29]. Recent studies have extended these approaches to pandemic contexts, such as the equitable distribution of COVID-19 testing centers [30,31]. These studies underscore the importance of spatial equity and service coverage. While informative, such approaches differ from the present focus, which emphasizes dynamic, shelter-level hazard, and lifeline status estimation to support operational functionality during disasters.

Research on lifeline infrastructure systems, which are critical for sustaining shelter operations, has progressed through the utilization of satellite imagery, remote inspection, field surveys, multisensor networks, social media analytics, strain gauges, and empirical models [2]. These approaches differ in terms of coverage, precision, and timeliness following a disaster, with a few being capable of delivering rapid, high-precision data that are publicly accessible. Notably, there is a paucity of tools that can provide detailed, publicly available, shelter-level lifeline status in near real-time [2]. Functional recovery frameworks developed by agencies such as FEMA and the National Institute of Standards and Technology emphasize system-level coordination and sectoral recovery objectives [24,32,33]; however, they seldom incorporate time-scaled, shelter-specific estimations from heterogeneous data sources.

To address these gaps, this study introduces the Shelter Hazard Impact and Lifeline Outage Estimation (SHILOE) framework. This is a stepwise approach that systematically and continuously updates shelter-specific, time-scaled damage information by integrating multisensor datasets. The specific objectives of our study were to (1) summarize information about both designated and activated evacuation shelters; (2) incorporate hazard impacts, including seismic intensity, liquefaction, tsunami inundation, and lifeline service outages (power outages, communication disruptions, and water supply outages); (3) provide a spatiotemporal summary of the evacuation shelter status alongside relevant Social Networking Service (SNS) data; and (4) develop a stepwise framework for SHILOE that incorporates both disaster-specific and non-disaster-specific datasets. As a case study, we applied the proposed framework to the 2024 Noto Peninsula earthquake.

The novelty of this research lies in the SHILOE framework’s ability to dynamically generate shelter-specific, time-scaled damage estimations. This is achieved by incorporating diverse multisource sensing data. Unlike existing approaches, SHILOE emphasizes continuous updates at the level of individual shelters, thereby strengthening the theoretical foundation for dynamic damage assessment. In doing so, it enhances shared situational awareness of disaster impacts on evacuation shelter status among stakeholders, thereby supporting a more effective evaluation of shelter functionality, prioritization of operations, and allocation of resources to bolster community resilience.

This paper proceeds as follows. Section 2 details the methodological framework, study area, and datasets. Section 3 presents the main findings, and Section 4 offers discussion of results and limitations. Finally, Section 5 concludes and highlights avenues for future research.

2. Materials and Methods

2.1. Overall Methodological Workflow

The methodological workflow, shown in Figure 1, follows four core stages, forming a stepwise framework to generate shelter-specific, time-scaled information products on hazard impacts and lifeline damage. First, hazard impacts—such as seismic intensity, liquefaction, and tsunami inundation—are correlated with shelter data. Second, the operational status of shelter lifeline services, including power outages, communication disruptions, and water supply disruptions, extracted from multisensor-derived datasets, is associated with shelter data. In the third step, time-series summaries of shelter-level damage estimations and SNS-related inputs are presented. Finally, a stepwise estimation framework is constructed by integrating these datasets. The subsequent sections provide a detailed workflow analysis.

2.2. Overview of the 2024 Noto Peninsula Earthquake and Study Region

A magnitude 7.6 earthquake occurred off the northeastern coast of Japan’s Noto Peninsula at 16:10 JST (07:10 UTC) on 1 January 2024 with the epicenter located at 37.50° N, 137.27° E. This seismic event followed a prolonged earthquake swarm in the region, during which more than 23,700 earthquakes of magnitude 1.0 or greater were recorded between 1 December 2020 and the day of the event [34]. Severe shaking and subsequent aftershocks devastated nearby municipalities such as Suzu, Noto, Wajima, and Anamizu [35,36]. These triggered multiple cascading hazards such as tsunami waves, landslides, surface deformation, and widespread fires [37], leading to significant destruction of infrastructure and residential buildings [38].

According to official reports, the disaster resulted in over 500 confirmed deaths, nearly 300 related fatalities, 390 severe injuries, and damage to more than 115,000 residential structures in Ishikawa prefecture [39,40]. For this study, six heavily affected municipalities in the northern Noto Peninsula—Anamizu, Nanao, Noto, Shika, Suzu, and Wajima—were selected (Figure 2).

2.3. Evacuation Shelter-Related Data

2.3.1. Designated Evacuation Shelter Data

We employed evacuation shelter data curated by the National Land Information Division, National Spatial Planning and Regional Policy Bureau, Ministry of Land, Infrastructure, Transport, and Tourism, Japan [42]. Owing to the absence of detailed data for Noto in the national dataset, this study augmented the information with records obtained from the Noto local government [43]. Shelters classified as regional or welfare evacuation shelters were excluded from the analysis.

2.3.2. Activated Evacuation Shelter Data

In mid-January, the Ishikawa prefectural government disseminated unified datasets of activated evacuation shelters via its disaster prevention portal and official disaster information system [44,45]. This dataset amalgamates shelter information reported by various entities, including municipal governments, Disaster Medical Assistance Teams (DMAT), and the Japan Self-Defense Forces. It encompasses both originally designated shelters and those newly activated in response to seismic events. In this study, we utilized a version of the dataset published on 22 January 2024. To ensure consistency and eliminate redundancies, duplicate entries were removed by cross-referencing the shelter names, addresses, and geographic coordinates.

2.4. Hazard-Related Datasets

2.4.1. Presumed Seismic Intensity Distribution

In Japan, the National Research Institute for Earth Science and Disaster Resilience (NIED) produces a presumed seismic intensity distribution at a 250 m mesh resolution utilizing a real-time damage estimate system (RDES). This system initially integrates ground motion data from approximately 5300 observation stations, encompassing nationwide strong-motion networks (Kyoshin Network [46] and Kiban Kyoshin Network [47]), local governments, and the Japan Meteorological Agency [48]. Site amplification factors are applied to estimate ground motion parameters such as peak acceleration, peak velocity, seismic intensity, spectrum intensity, and velocity response spectrum. These factors were obtained from the Japan Seismic Hazard Information Station and wide-area subsurface models, including those covering the Kanto and Tokai regions [48]. The estimated seismic intensity distribution is usually available within 10–20 min after a mainshock [48,49,50]. For the 2024 Noto Peninsula Earthquake, the initial distribution was issued at 16:29 JST (07:29 UTC) on 1 January 2024 [50], followed by an update at 18:40 JST (09:40 UTC) on 3 January 2024 [51]. The delay in releasing the update was attributed to the extensive hypocentral area, frequent aftershock activities, and power outages, which disrupted data transmission. Figure 3a illustrates the spatial distribution of the presumed seismic intensity of the event.

2.4.2. Liquefaction

The liquefaction probability was generated by NIED utilizing a liquefaction damage estimation system with a 250 m mesh resolution [52]. This estimation was derived from Equations (1) and (2), which are prediction equations based on the maximum velocity and measured seismic intensity, respectively. Equation (3) incorporates the liquefaction area ratio to calculate the final liquefaction hazard probability.

$$P_v\left(PGV\right)=\Phi \left[{\displaystyle \frac{\mathrm{I}\mathrm{n}\left(PGV\right)-{\lambda }_v}{{\zeta }_v}}\right]$$

(1)

$$P_i\left(I\right)=\Phi \left[{\displaystyle \frac{I-{\lambda }_i}{{\zeta }_i}}\right]$$

(2)

$$P_R\left(PGV\right)=P_v\left(PGV\right)\times P_A$$

(3)

where PGV represents the peak ground velocity at the surface [cm/s], and I denotes the instrumental seismic intensity. The liquefaction occurrence probability, denoted as P_v(PGV), is calculated using PGV and represents the proportion of grid cells expected to experience liquefaction). P_i(I) signifies the liquefaction occurrence probability derived from the seismic intensity I, and Φ denotes the standard normal cumulative distribution function. λ_v and ζ_v represent the mean and standard deviation of ln(PGV), respectively, while λ_i and ζ_i represent the mean and standard deviation of the seismic intensity I. P_A indicates the liquefaction area ratio, and P_R(PGV) denotes the final liquefaction hazard probability. For the 2024 Noto Peninsula earthquake, the initial liquefaction probability was generated at 16:41 JST (07:41 UTC) on 1 January 2024 [52]. Areas with liquefaction probabilities exceeding 2% were extracted through comparison with field survey data [53]. Regions surpassing this threshold were classified as having a high liquefaction risk.

2.4.3. Tsunami Inundation

In response to the earthquake, the Geospatial Information Authority of Japan (GSI) promptly commenced data collection and aerial surveys in alignment with its national mandate for geospatial analysis and hazard mapping [54]. The findings were systematically compiled and disseminated in the immediate aftermath of the disaster. We utilized GSI’s tsunami hazard outputs, which were derived from the aerial imagery captured on 2, 5, 11, 14, and 17 January 2024 [37].

2.5. Lifeline-Related Data

2.5.1. Power Outage Status

To assess the status of power outages following the primary seismic event, we employed television viewing data, specifically an IoT-derived home appliance dataset provided by the SHARP Corporation [55] (hereafter referred to as IoT-derived data), aggregated at the zip-code level. This dataset recorded the viewing activity from 00:00 to 17:00 JST on 1 January 2024 (15:00 UTC, 31 December 2023, to 08:00 UTC, 1 January 2024) at one-minute intervals. For analytical purposes, the data were segmented into two temporal phases: a pre-mainshock period (00:00–16:15 JST) and a period immediately following the mainshock (16:15–17:00 JST). Within each phase, the maximum number of active viewing signals was extracted for each zip-code area. Subsequently, the percentage change in active signals between the two phases was calculated. The power outage status was then classified into three categories based on the signal drop: (1) power outage: ≥50% decrease in viewing signals; (2) no outage: increase, no change, or <50% decrease; and (3) no information: insufficient or missing data. The ≥50% threshold was established by comparing it with official outage data from the regional power company [56]. This value was selected to align with reported household outages while minimizing the risk of underestimating data from the affected areas, particularly for disaster response. Concurrently, the official power outage reports from the Ishikawa prefectural government (1–31 January 2024) [38] were summarized and visualized to examine temporal trends and compare them with the IoT-derived results.

2.5.2. Communication Network Data

NTT-Docomo, a leading mobile network provider in Japan with the most extensive nationwide coverage [57,58,59,60], provides restoration area maps [57] that detail service disruptions and recovery progress. For this study, we utilized the restoration map released at 13:14 JST (04:14 UTC) on 2 January 2024, which likely represented the peak extent of the communication network disruption. This map was digitized and transformed into a geospatial dataset for analytical purposes.

2.5.3. Water Outage

Water supply service areas [61] situated within 250 m mesh grids that exhibited a presumed seismic intensity exceeding “6 Lower” were identified. For each service area, the maximum seismic intensity recorded within its boundary was designated as the representative value. Areas that met this intensity criterion were deemed to be at high risk of experiencing water supply disruptions.

2.6. Social Networking Service Information

FASTALERT, an AI-driven real-time risk information service developed by the JX Press Corporation, Japan [62], was employed in this study. This platform aggregates real-time disaster-related data from social media and other sources to detect incidents nationwide, such as natural disasters and accidents [63]. It disseminates pertinent information, including timestamps, geographic coordinates, disaster categories, associated images, and user-generated text.

To ensure reliability, FASTALERT employs a two-stage processing workflow. Initially, AI models score both images and texts: image analysis evaluates similarity to previously confirmed disaster-related content (e.g., fires, accidents, and landslides), whereas text analysis applies morphological and contextual parsing to distinguish actual ongoing events from jokes or hearsay. Approximately 99.7% of the posts are discarded at this stage, with only the higher-scoring posts forwarded for human verification. Human operators then refine the event classification and geographic attribution [64].

Second, additional operational measures are implemented to further mitigate misinformation. These measures include blocking recurrent false-information accounts, dynamically adjusting data-collection rules according to disaster contexts (e.g., prioritizing place names and filtering entertainment-related terms), and referencing a knowledge base of frequent misclassification cases (e.g., industrial flares misidentified as fires). As additional data become available, disseminated information is continuously updated, and geographic inaccuracies are corrected based on new inputs or user feedback [64].

In this study, only hazard- and damage-related information with confirmed geographic attribution was utilized.

2.7. Data Incorporation

The data from evacuation shelters, both designated and activated, were superimposed on the outputs generated in Section 2.4, Section 2.5 and Section 2.6. Data regarding hazard damage identified within a 50 m buffer surrounding each shelter were incorporated. The corresponding hazard impact values and indicators of lifeline service outages were appended to each shelter’s attribute table. For this study, FASTALERT entries located within a 1 km buffer around each evacuation shelter and posted between 1–14 January 2024, were included. These entries were subsequently summarized and visualized alongside the shelter impacts to facilitate the validation of the estimations.

Table 1. Main specifications of the data used in this study.

Data/Estimation Outputs/ Information	Issued/Acquisition Date	Spatial/Temporal Res.	Agency
Designated evacuation shelter	2012 2022	–	Noto [43] MLIT [65]
Activated evacuation shelter	22 January 2024	–	Ishikawa prefectural government [66]
Presumed seismicity distribution	1 January 2024	250 m	NIED [67]
Liquefaction probability	1 January 2024	250 m	NIED [67]
Tsunami inundation	5 January 2024~	–	GSI [68]
Television viewing data	1 January 2024	1 min	SHARP [55]
Communication network service area	2 January 2024		NTT-Docomo [69]
Water supply service area	2012		NLID [61]
Official damage report	1–31 January 2024	–	Ishikawa prefectural government [38]
FASTALERT real-time risk information in Japan	1–14 January 2024	–	JX Press Corporation [62]

provides an overview of the key specifications of the datasets, estimation outputs, and SNS inputs utilized in this study.

3. Results

3.1. Spatial Distribution of Evacuation Shelters

Figure 4 illustrates the spatial distribution of both designated and activated evacuation shelters as of 22 January 2024. The highest number of designated shelters (134) were present in Nanao, followed by Suzu (77), Anamizu (51), Noto (42), Shika (36), and Wajima (34) (

Table 2. Number of designated and activated evacuation shelters by seismic distribution level and municipality.

	Seismic Intensity
	5 Lower	5 Upper	6 Lower	6 Upper	7	Total
Nanao	0 (0)	7 (4)	48 (19)	70 (17)	9 (0)	134 (40)
Suzu	0 (0)	0	10 (12)	53 (61)	14 (16)	77 (89)
Anamizu	0 (0)	0	6 (9)	43 (57)	2 (7)	51 (73)
Noto	0 (0)	7 (7)	24 (52)	10 (18)	1 (0)	42 (77)
Shika	0 (0)	10 (8)	25 (26)	1 (11)	5	36 (50)
Wajima	0 (1)	2 (11)	9 (80)	20 (81)	3 (9)	34 (182)
Total	0 (1)	26 (30)	122 (198)	197 (245)	29 (37)	374 (511)

Note: Values in parentheses represent the number of activated evacuation shelters within each seismic intensity category.

). These shelters were predominantly situated along the coastal regions (Figure 3a). Conversely, the numbers of activated shelters across the northern Noto Peninsula on 22 January 2024 were significantly greater: Wajima (182), Suzu (89), Noto (77), Anamizu (73), Shika (50), and Nanao (40) (Table 2). A substantial proportion of both designated and activated shelters were located in areas that experienced a seismic intensity of “6 Lower” or higher, with dense clusters observed in severely affected municipalities, such as Wajima and Suzu.

3.2. Lifeline Damage Status

3.2.1. Power Outage Status

Figure 5 presents the total number of households impacted by power outages from 1 to 31 January 2024, as reported in an official damage report [38]. The peak in outages occurred on 2 January across all examined municipalities, totaling 33,200, with Wajima experiencing the most significant impact (10,600 on 4 January), followed by Suzu (8100 on 1 January), Noto (6200 on 2 January), Anamizu (5400 on 3 January), Nanao (550 on 7 January), and Shika (20 on 18 January). Data were not reported from 1 to 5 January in Nanao, nor from 1 to 17 January in Shika. The majority of power outages were progressively resolved in the subsequent weeks.

Figure 6 illustrates the spatial distribution of outages across three critical lifeline services in the study area: electricity (Figure 6a,b), communication networks (Figure 6c), and water supply (Figure 6d). Figure 6a depicts the municipality-level power outage status as of 1 January 2024, whereas Figure 6b and

Table 3. IoT-derived power outage status by municipality.

	Power Outage Status
	Outage	No Outage	Unknown	Total
Wajima	59 (27.1)	37 (17.0)	122 (56.0)	218 (100.0)
Nanao	51 (26.0)	78 (39.8)	67 (34.2)	196 (100.0)
Suzu	24 (33.8)	13 (18.3)	34 (47.9)	71 (100.0)
Shika	20 (18.0)	47 (42.3)	44 (39.6)	111 (100.0)
Noto	17 (20.7)	26 (31.7)	39 (47.6)	82 (100.0)
Anamizu	16 (26.7)	15 (25.0)	29 (48.3)	60 (100.0)
Total	187 (25.3)	216 (29.3)	335 (45.4)	738 (100.0)

Note: Values represent the number of zip-code areas categorized by power outage status with percentages in parentheses. “Unknown” indicates areas without available outage information.

provide the IoT-derived power outage status at 17:00 JST (08:00 UTC) on the same date, visualized at the zip-code level. Within the study area, zip-code areas were categorized as experiencing a power outage (25.3%), not experiencing an outage (29.3%), or unknown (45.4%) (n = 738). A high prevalence of outages was observed in Suzu (33.8%), Wajima (27.1%), Anamizu (26.7%), and Nanao (26.0%). Despite some data gaps, the IoT-derived dataset provided more detailed spatial information—particularly in the most severely affected regions—than the aggregated prefectural data.

3.2.2. Communication Network Disruption

Figure 6c illustrates the disruption of the communication network at 13:14 JST (04:14 UTC) on 2 January 2024, as depicted in the restoration area map released by NTT-Docomo. Significant network outages occurred across the northern Noto Peninsula, particularly in Wajima, Noto, and Amamizu. These disruptions likely resulted from the combined effects of physical infrastructure damage, extensive power outages, and limited access to restoration crews in the immediate aftermath of the earthquake.

3.2.3. Water Outage

Figure 6d depicts the spatial distribution of projected water outages as determined by identifying water supply service zones that experienced a seismic intensity of “6 Lower” or greater. The majority of the study area was impacted, with the exception of several zones in Nanao, which were comparatively less affected.

3.3. Damage Status at Shelters

Figure 7 illustrates the status of the hazard damage—seismic intensity, liquefaction, and tsunami inundation—at evacuation shelters, as determined from the multitemporal and multisensor datasets. Figure 8 depicts the disruptions in three critical lifeline services—electricity, communication networks, and water supply—based on additional multitemporal and multisensor datasets. The findings revealed significant spatial disparities in both hazard impacts and lifeline service disruptions, with numerous shelters experiencing concurrent disruptions from multiple hazard types and lifeline systems.

Figure 9 displays the proportion of evacuation shelters in each municipality affected by various hazard combinations—seismic distribution, liquefaction, and tsunami inundation. Across the study area, approximately 93.6% of the designated shelters and 95.3% of activated shelters were affected by at least one hazard, with dual-hazard exposure (seismic intensity and liquefaction) being the most prevalent pattern (A/A/ND, 54.0%; A/A/N, 44.2%). Triple-hazard exposure was relatively rare, occurring in only 2.0% of activated shelters. Comprehensive breakdowns by municipality and hazard type are provided in

Table A1. Proportion of evacuation shelters affected by various hazard types (seismic intensity, liquefaction, tsunami inundation) in each municipality. (a) Designated evacuation shelters. (b) Activated evacuation shelters.

	(a) Designated Evacuation Shelters Affected by Various Hazard Types (Seismic Intensity/Liquefaction/Tsunami Inundation)
	N/N/ND			A/N/ND		A/A/ND			Total
Nanao	1.9			14.7		19.3			35.8
Anamizu	—			7.0		13.6			20.6
Noto	—			4.3		9.4			13.6
Shika	1.9			4.8		4.5			11.2
Suzu	2.4			2.9		4.3			9.6
Wajima	0.3			5.9		2.9			9.1
Total	6.4			39.6		54.0			100.0
		(b) Activated Evacuation Shelters Affected by Various Hazard Types (Seismic Intensity/Liquefaction/Tsunami Inundation)
		N/N/N	A/N/N		A/N/A		A/A/N	A/A/A		Total
Wajima		1.4	22.9		0		11.4	0		35.6
Anamizu		0	5.9		0.2		10.8	0.6		17.4
Shika		1.4	8.4		0		4.1	1.2		15.1
Noto		0	4.5		0		9.6	0.2		14.3
Suzu		1.2	3.1		0.2		5.3	0		9.8
Nanao		0.8	3.9		0		3.1	0		7.8
Total		4.7	48.7		0.4		44.2	2.0		100.0

Note: Codes in column headers indicate hazard combinations: the order of hazards is seismic intensity/liquefaction/tsunami inundation; N = not affected, A = affected, ND = no data. All values are presented as percentages.

(Appendix A.1).

Overall, these results indicate that most shelters in the study area were affected by multiple hazards, with dual-hazard exposure being significantly more common than triple-hazard exposure.

Figure 10 illustrates the major proportion of evacuation shelters in each municipality affected by various combinations of lifeline outage combinations, specifically, electricity, communication networks, and water supply. Lifeline service disruptions were pervasive, with 100.0% of both designated and activated shelters experiencing at least one type of outage. However, outage patterns varied substantially across municipalities. Across all municipalities, the most common pattern among designated shelters was functional electricity and communication networks and outage of water supply (N/N/A, 35.6%; 23.9%), whereas activated shelters showed higher proportions of triple outages (A/A/A, 7.0%; 11.7%). Comprehensive details by lifeline and municipality are provided in

Table A2. Proportion of evacuation shelters affected by lifeline disruptions (electricity, communication networks, and water supply) in each municipality.

	Evacuation Shelters Damaged by Lifeline (Electricity/Communication Networks/Water Supply)
	N/N/A	A/N/A	N/A/A	A/A/A	N/ND/A	ND/N/A	ND/ND/A	A/ND/A	ND/A/A	Total
Nanao	17.4 (4.9)	10.2 (2.2)	0 (0)	0.5 (0.2)	1.6 (0.0)	4.0 (0.6)	1.6 (-)	0.5 (0)	0 (0)	35.8 (7.8)
Suzu	4.0 (4.1)	7.5 (4.5)	0.8 (0.2)	0.3 (1.4)	0.5 (0.2)	2.9 (3.5)	1.9 (1.2)	1.6 (1.4)	1.1 (1.0)	20.6 (17.4)
Anamizu	2.1 (1.8)	3.2 (3.5)	2.4 (2.9)	0.5 (0.8)	0.6 (0.6)	1.3 (1.2)	1.1 (1.4)	1.1 (1.0)	1.3 (1.2)	13.6 (14.3)
Noto	4.0 (4.5)	0.3 (0.4)	1.1 (1.8)	3.7 (3.5)	0.3 (0.2)	0.8 (2.0)	0.0 (0.6)	0.3 (0.2)	0.8 (2.0)	11.1 (15.1)
Shika	7.2 (5.1)	0.0 (1.8)	0 (0)	0 (0)	1.3 (0.2)	1.1 (1.8)	0.0 (0.2)	0.0 (0.6)	0.0 (2.0)	9.6 (9.8)
Wajima	0.8 (3.5)	2.4 (7.2)	0.3 (2.3)	1.9 (5.9)	0.3 (2.0)	1.9 (3.5)	1.3 (2.9)	0.3 (1.6)	0.0 (6.7)	9.1 (35.6)
Total	35.6 (23.9)	23.5 (19.6)	4.5 (7.2)	7.0 (11.7)	4.5 (7.2)	12.0 (12.5)	5.9 (6.3)	3.7 (4.7)	3.2 (11.0)	100.0

Note: Codes in column headers indicate lifeline combinations: the order of lifeline is electricity/communication networks/water supply; N = not affected, A = affected, ND = no data. All values are presented as percentages. Values outside parentheses refer to designated shelters; values in parentheses refer to activated shelters.

(Appendix A.2).

Overall, the findings indicate substantial intermunicipal variation in lifeline outage patterns, with municipalities such as Anamizu, Nanao, Suzu, and Wajima experiencing higher concentrations of multisystem failures.

3.4. Spatiotemporal Summary of SNS Data on Earthquake-Induced Hazard and Damage

Between 1 and 14 January 2024, reports related to hazards (e.g., landslides, liquefaction, large-scale fires), damage (e.g., building damage, water supply outages, road damage), and other incidents (e.g., traffic accidents, emergency vehicle dispatches, rescue requests) constituted 30.4%, 46.0%, and 23.6% of all FASTALERT entries, respectively (n = 250). At the municipal level, the majority of posts originated from Wajima (38.7%), followed by Nanao (25.1%), Anamizu (13.6%), Suzu (7.9%), and Noto and Shika (7.3% each) (

Table 4. Percentage distribution of hazard- and damage-related SNS reports by municipality and date.

	1 January	2 January	3 January	4 January	5 January	6 January	7 January	8 January	Grand Total
Wajima	27.2	9.4	1.6	-	-	0.5	-	-	38.7
Nanao	9.9	13.6	1.0	-	-	-	-	0.5	25.1
Anamizu	6.8	5.2	0.5	-	-	-	-	1.0	13.6
Suzu	4.2	2.6	0.5	0.5	-	-	-	-	7.9
Noto	1.6	4.7	0.5	-	-	-	0.5	-	7.3
Shika	3.1	3.1	1.0	-	-	-	-	-	7.3
Grand Total (n = 191)	52.9	38.7	5.2	0.5	-	0.5	0.5	1.6	100.0

).

In terms of temporal distribution, 52.9% of hazard- and damage-related posts were recorded on 1 January, 38.7% on 2 January, 5.2% on 3 January, 0.5% on 4, 6, and 7 January, and 1.6% on 8 January 2024 (n = 191) (Table 4). These spatial patterns were superimposed on the time-scaled evacuation shelter status for comparative visualization (Figure 11 and Figure 12).

3.5. Time-Scaled Visualization of Hazard and Lifeline Status

On 1 January 2024, data pertaining to hazard-related estimations were disseminated, encompassing seismic intensity distribution (16:29 JST, 07:29 UTC) and liquefaction probability (16:41 JST, 07:41 UTC). For analytical purposes, a standardized hazard information release time of 16:40 JST (07:40 UTC) was employed, while tsunami inundation information was set at 9:00 JST (00:00 UTC) 5 January 2024. The activated evacuation shelter data were officially released on 22 January 2024. However, due to coordination delays in Ishikawa prefecture, these were generalized to five days post-event (5 January 2024). IoT-derived power outage data required approximately 30 min of processing; thus, 16:40 JST (07:40 UTC) was used as the release time. Although peak communication outages occurred on 2 January, data became available on 1 January 2024, so the initial availability date was applied. For illustrative purposes, the same 1 January hazard and lifeline datasets were also applied to the time-scaled analysis for 5 January 2024.

By 16:40 JST (07:40 UTC) on 1 January 2024, no hazard or only a single damage-related FASTALERT post had been distributed. In contrast, 146 posts (76.0% of the total; n = 191) were recorded between 16:40 JST (07:40 UTC) on 1 January 2024 and 18:00 JST (09:00 UTC) on 5 January 2024. These posts were spatially aligned with the incorporated hazard and lifeline results (Figure 11a,b and Figure 12a,b). Key FASTALERT entries (Figure 11 ①–⑤ and Figure 12 ①–⑥) illustrate spatiotemporal correspondence, confirming the consistency and reliability of the integrated hazard and lifeline damage estimates.

3.6. Proposed Stepwise Estimation Framework for Shelter Hazard Impact and Lifeline Outage Estimation Using Time-Scaled Multisensor Integration

Given the observed hazard impacts and lifeline disruptions at evacuation shelters, we propose a stepwise framework for assessing shelter-level hazard exposure and lifeline status (Figure 13). This methodology integrates the most reliable multisource sensing datasets—both disaster-specific and non-disaster-specific—structured across three sequential phases.

• Immediate Estimation Phase (within tens of minutes after the seismic event): Initial damage estimates were derived from datasets on designated evacuation shelters. This data were combined with simulation-based hazard information (e.g., seismic intensity, liquefaction probability, and tsunami simulation) and lifeline service outage indicators (e.g., power outages, communication disruptions, and water supply outages). Although these early estimates may lack precision, they provide essential time-critical insights to support rapid emergency responses. They integrated potential inputs, such as tsunami inundation simulations and landslide probability data, to further improve immediate estimates.
• Intermediate Phase (days following the event): The preliminary results were updated by substituting the initially designated shelter dataset with comprehensive lists of operational evacuation shelters, as obtained from official disaster prevention portals. Hazard information, including tsunami simulations and landslide probabilities, was refined using the most recent observation-based estimates, such as remote-sensing-derived assessments of tsunamis and landslides. This phase enhances the accuracy of identifying actively functioning shelters, their associated hazard impacts, and the status of lifeline services.
• Refinement Phase (ongoing): The data pertaining to activated shelters, hazard impacts, and the status of essential services are continuously updated through near real-time feeds. We incorporated SNS risk information from social media, ground-truth data from field surveys, and official post-disaster inspections to validate and enhance the estimates. Additional inputs, such as live monitoring systems (e.g., human and traffic flow), IoT-enabled home appliance data (e.g., air conditioners and refrigerators), and high-resolution satellite, aerial, or drone imagery, are integrated to further improve reliability and resolution.

In the context of diverse data sources, the temporally structured SHILOE framework effectively produces time-scaled, consistent, and actionable outputs to address a broad spectrum of disaster-response requirements. This framework has several practical applications:

• Early stage prioritization of disaster response: This utilizes hazard and lifeline status estimates to swiftly identify high-impact areas and critical lifeline disruptions, thereby guiding the prioritization of resources and distribution of materials.
• Rapid needs assessment: Integrating hazard and lifeline status estimates with aggregated hazard- and damage-related SNS posts in near real-time facilitates the estimation of essential needs (e.g., food, mobile network access, and water) prior to the completion of official surveys.
• Shelter demand forecasting: Temporal spikes in posts concerning shelter crowding, access difficulties, and service outages are early indicators of increasing demand. These signals enable the anticipation of capacity needs, allowing for proactive adjustments in staffing, supply, and facility readiness.
• Improved coordination among stakeholders: The integration of hazard and lifeline status mapping with real-time SNS information provides a common operational picture for agencies, NGOs, and community groups. This enhances decision-making, and efficiency, reduces duplication of efforts, and ensures the deployment of critical resources to areas with the greatest needs.
• Optimization of shelter operations: By analyzing aggregated reports of shortages, overcrowding, or operational issues, and cross-referencing them with hazard severity, the framework guides real-time resource allocation, supply redistribution, and the establishment of additional temporary shelters.
• Enhanced community resilience: The combination of real-time SNS data with hazard and lifeline assessments strengthens recovery capacity, improves situational awareness, and supports long-term preparedness.

By aligning damage estimation with the evolving availability of data, this approach fosters shared situational awareness and supports more agile and informed decision-making throughout the disaster-response timeline.

4. Discussion and Limitations

4.1. Stepwise Estimation of Shelter Damage Using Time-Scaled MultiSensor Incorporation

A timely and comprehensive understanding of evacuation shelter damage is crucial for evaluating shelters functionality, setting operational priorities, optimizing resource allocation, and enhancing operability to bolster community resilience. However, few existing methodologies have systematically estimated shelter-level damage using time-scaled, multisource sensing data. The SHILOE framework addresses this gap by continuously integrating the most reliable datasets available at each time scale to update the hazard and lifeline status for individual shelters.

Compared with established disaster assessment systems, the SHILOE framework offers distinct operational advantages. FEMA’s Hazus tool [24] provides robust, scenario-based loss estimates, but remains static and regional in scale. The Copernicus EMS [26] and UNITAR–UNOSAT [25] deliver high-resolution satellite-based rapid maps that enhance broad situational awareness for civil protection and humanitarian coordination; however, their products are typically single-event snapshots and do not reflect evolving shelter conditions. In contrast, SHILOE provides dynamic, stepwise updates of hazard exposure and lifeline outages at the facility level. By combining multisource data—including hazard simulation, real-time monitoring, SNS risk reports, IoT-based outage records (available within ten of minutes of the mainshock), remote sensing imagery, and ground-truth data (available within days)—SHILOE enables time-scaled, near-real-time, localized monitoring. IoT datasets, in particular, provide minute-level temporal resolution and are automatically generated through IoT-enabled devices. These high-frequency, localized observations complement prefectural data, typically reported at the municipal level, and are particularly valuable during large-scale disasters, when timely, location-specific information is critical for effective emergency response. This capability supports real-time operational decision-making for evacuation shelter management and complements the primary planning- or snapshots-based approach of existing systems.

This continuous, high-frequency capability allows emergency managers to track shelter disruptions as they unfold, prioritize resources dynamically, and reinforce shelters in multihazard zones. Beyond mapping, SHILOE functions as an operational decision-support tool, providing situational awareness of critical lifeline services (electricity, water, and communication) to support evacuees and sustain overall disaster-response effectiveness. Thus, the framework directly addresses the limitations of conventional systems and advances state-of-the-art post-disaster shelter management.

4.2. Utilization of Stepwise SHILOE Approach

The proposed framework enhances the situational understanding of disaster impacts among various stakeholders in a unified manner. For instance, the status of power outages at designated shelters—an IoT-derived dataset aggregated at the zip-code level, along with data on communication networks and water supply outages–was accessible as early as 1 January 2024. In contrast, the official damage reports for the same date contained no records. Power outage information was initially disseminated at the provincial level at 07:00 on 2 January 2024, and at the municipal level at 08:00 on 3 January 2024 [70,71]. Communication networks and water supply outages were disseminated at the municipal level at 07:00 on 2 January 2024 [70]. Although some activated evacuation shelters were listed on the Ishikawa Prefecture Disaster Prevention Portal [44], the list was incomplete, and no shelter-specific lifeline damage status was provided. Even when derived from the most reliable available datasets, early stage SHILOEs may have limited accuracy. However, they are crucial for guiding operational planning and prioritizing emergency responses, particularly in the absence of comprehensive official damage reports, as occurred on 2 January 2024.

This stepwise framework not only improves the timeliness of hazard impacts and lifeline outage estimations but also fosters shared situational awareness among responders, policymakers, and the public. Developed in the context of the 2024 Noto Peninsula earthquake, its modular design is scalable and adaptable to future disaster-response scenarios. Furthermore, the framework supports the integration of newly available datasets, enabling the continuous refinement of assessments.

In future, the integration of Japan’s Evacuation Shelter Common ID system could significantly enhance the interoperability and operational utility of the SHILOE framework. Established by the Cabinet Office in 2022, this system assigns a unique 14-digit identifier to approximately 74,000 designated shelters and 112,000 designated evacuation sites nationwide [72]. It facilitates unambiguous identification, eliminates duplication, and facilitates information exchange, status tracking, and coordinated management among national and local agencies, including the Self-Defense Forces and DMAT. Wider deployment would enable the seamless linking of SHILOE outputs with official shelter registries, thereby improving data integration, multi-agency coordination, and real-time decision-making during large-scale disasters. Such integration would strengthen municipal and national capacities for evacuation planning, shelter management, and interagency communication, thereby extending SHILOE’s impact from analysis to actionable policy and operation.

Japan faces a significant risk of three major earthquakes within the next 30 years: the Tokyo Metropolitan, Nankai Trough, and Japan Trench/Chishima Trench events. The projected impacts of these events are considerable, with estimates of 25.4 million affected individuals and ¥95 trillion in losses for the Tokyo Metropolitan Earthquake; 40.7 million affected individuals, 231,000 fatalities, and ¥220 trillion in damages for the Nankai Trough Earthquake; and nearly 300 municipalities impacted by the Japan Trench/Chishima Trench event, with approximately 300,000 fatalities and ¥48 trillion in losses [73] (

Table 5. Estimated occurrence of national-level earthquakes in Japan over the next 30 years (Adapted from Reference [73]).

Earthquake NAME	Scale (M)	Probability of Occurrence (30 Years>)	Seismic Intensity (SI)	Estimate of Affected Population	Deaths (Thousand)	Debris (Million Tons)	Damage (Trillion JPY)
Tokyo Metropolitan	7.3	70%	7	25.4 M (SI 6L>)	23	98	95
Nankai Trough	9.0	70–80%	7	40.7 M (SI 6L>)	231	310	220
Japan Trench/ Chishima Trench	9.1/9.3	60%	7	Earthquake: 272 Municipalities Tsunami: 108 Municipalities	199/100	110	48

). These figures highlight the urgency of advancing disaster impact models—particularly those assessing shelter functionality and lifeline operability—to improve preparedness, responses, and recovery in future large-scale disasters.

4.3. Practical Implications for Emergency Management

Although the current study does not provide any guidance on the spatial allocation or relocation of shelters, it offers significant operational insights for emergency management professionals. For instance, in Wajima City, 182 shelters were activated despite only 34 being pre-designed, underscoring the necessity for contingency planning for non-designated facilities and the importance of adaptable resource prepositioning. Similarly, a multi-hazard exposure analysis (Figure 11) identifies shelters that are concurrently affected by seismic intensity, liquefaction, or tsunami risk. Rather than recommending relocation, these findings suggest priorities for reinforcing activating shelters, securing backup lifeline systems (e.g., emergency power, water storage, and satellite communications), and integrating hazard awareness into routine preparedness planning. Thus, the SHILOE framework facilitates real-world decision-making by providing near-real-time, shelter-specific information that can guide resource allocation and coordination during both the preparedness and response phases.

4.4. Limitations

Although the SHILOE framework exhibits significant potential for near-real-time hazard and lifeline assessment at the shelter level, several limitations remain. First, certain hazard layers—such as seismic intensity and liquefaction probability—rely on simulation-based models with a 250 m mesh resolution in the immediate phase, which may not accurately reflect localized conditions until observation-based data become available. Second, estimations depend on the availability and resolution of input datasets, which may be limited to regions with low IoT penetration or SNS activity. Third, temporal resolution varies across datasets. For example, IoT-derived power outage data update frequently, whereas water supply data are refreshed less often, limiting the precision of synchronized analyses. Fourth, restricted access to utility records and official shelter updates can delay integration. Fifth, equity-oriented siting and environmental justice were not addressed, as many shelters were activated spontaneously, limiting pre-event optimization. Lastly, the continuity of IoT-derived datasets depends on device energy efficiency and power stability, which can be particularly challenging in remote or disaster-affected areas. Recent work on ultralow-power and energy-harvesting techniques provides potential pathways to enhance the resilience of such systems [74].

4.5. Methodological Considerations

Several methodological considerations warrant attention during implementation of the SHILOE framework. First, to improve reliability when datasets are sparse immediately after an event, the framework uses the most dependable sensing datasets available at each temporal scale and incrementally refines estimates as new sources emerge. Second, although simulation-based hazard models provide rapid initial estimates, SHILOE dynamically updates them as new data arrive, reducing reliance on model-only outputs. Third, although data gaps remain, the framework is designed to integrate expanded sensor networks and advanced IoT technologies as they become available. Finally, to manage the computational load of 15 min nationwide updates, SHILOE uses in-memory storage of static datasets and region-specific processing to maintain timeliness amid high data volumes.

5. Conclusions

We present a stepwise framework designed to estimate earthquake-related hazards and lifeline disruptions in evacuation shelters. By combining disaster-specific and routine sources, this framework enables facility-level updates that evolve over time. The findings demonstrate that the progressive incorporation of multisource information can significantly enhance the timeliness assessments of shelter conditions. Incorporating additional resources—such as social media content, secondary hazard models (e.g., tsunami and landslide), IoT-derived household device data (e.g., air conditioners and refrigerators), mobility monitoring (e.g., human and traffic flows), and very-high-resolution remote sensing—offers promising avenues for further refinement. Beyond improving timeliness, the proposed stepwise framework enhances shared situational awareness among diverse stakeholders, enabling timely shelter evaluations, operational prioritization, resource optimization, and improved operability to strengthen community resilience. Future research should aim to automate data ingestion and processing for real-time use, adapt the framework to various hazard (e.g., floods and typhoons), and expand partnerships with government agencies, utilities, and research institutes to improve access to high-resolution lifelines and hazard datasets. Furthermore, the framework should be developed into a near real-time decision-support tool integrated with emergency GIS platforms. Additionally, incorporating advanced analytics, such as machine learning for image classification or anomaly detection, can further enhance the accuracy, scalability, and adaptability of the SHILOE methodology across different geographic settings.