Quantifying Cascading Impacts of Natural Hazards on Power-Communication Interdependent Networks

Quantifying Cascading Impacts of Natural Hazards on Power-Communication Interdependent Networks Balaji V. Venkatasubramanian1,2, Christos Laoudias1, Mathaios Panteli1,2 1KIOS Research and Innovation Center of Excellence, University of Cyprus, Cyprus 2Department of Electrical and Computer Engineering, University of Cyprus, Cyprus [[email protected], [email protected], [email protected]] Abstract—Extreme weather events increasingly threaten critical infrastructure (CI) systems, particularly power and telecommunication networks, leading to cascading failures. This paper proposes a novel resilience assessment framework that explicitly models spatial interdependencies between electricity and telecommunication networks using a graph-based approach. Sectorspecific metrics, including Demand Not Served (DNS), power line failure rates, and affected population estimates, quantify the impact of windstorm-induced disruptions. A spatiotemporal hazard scenario generator simulates realistic storm events, dynamically propagating failures through the coupled networks. A key innovation is the integration of a telecommunication signal propagation model—combining Extended Hata, Rapport models, and a binary search algorithm—to estimate coverage loss within the resilience framework. The framework is validated in Cyprus using real-world and synthetic data, identifying vulnerable regions and informing disaster response strategies. Index Terms—Cyber-Physical Systems, Resilience, Interdependent Networks, Cascading Failures, Natural Hazards I. INTRODUCTION Power and communication networks are critical infrastructures (CIs) essential to modern society but are increasingly vulnerable to extreme events. Disruptions in one system can trigger cascading failures in the other, hindering emergency response and worsening socioeconomic impacts. For instance, Hurricane Beryl in 2024 caused massive power outages, leading to widespread communication failures [1]. With the shift to smart grids, communication networks play a vital role in real-time grid operations [2], but their interdependence heightens risks [3]. These challenges underscore the need for a unified resilience framework to safeguard cyber-physical systems (CPS) against cascading disruptions. In the literature, research on resilient interdependent CIs primarily focuses on modeling cascading failures, quantifying their impacts, and evaluating resilience within these interconnected power and communication systems. Various analytical approaches have been employed, including probabilistic modeling, complex network theory, and graph-based failure propagation frameworks to analyze the interactions between infrastructure components under stress conditions [4], [5]. Some studies incorporate intelligent algorithms to This work was supported by the European Union’s Horizon Europe research and innovation program through the project Solid Preparedness And Resilience for Robust Operations during disaster Wilderness (SPARROW) (Grant agreement ID: 101168499). assess system vulnerability. For instance, [6] applies a Qlearning approach to identify optimal sequences of attacks on communication lines, demonstrating how strategic disruptions to data flows can severely impact power grid control, resulting in large-scale outages within a Barab´ asi–Albert scale-free network model. Building on the vulnerability perspective, [7] proposes a methodology to mitigate the effects of natural hazards originating from power infrastructure. The framework emphasizes the critical role of communication systems that are essential for restoration following transmission line failures. The impact is quantified by assessing the total load lost over time and the number of cyber-physical equipment failures. Similarly, [8] proposes a Markov process-based framework to assess the resilience of interdependent infrastructure systems to weather-induced failures. It simulates bidirectional outages using fragility curves and quantifies impacts primarily through load shedding. Contributions by [9] enhance vulnerability analysis by identifying critical nodes through a criticality score that integrates topographical and electrical indicators. The approach is specifically tailored to power-line communication for transmitting power system data and does not incorporate general-purpose connectivity. Furthermore, [10] developed an open-source modeling framework within the CLIMADA risk assessment platform to evaluate largescale cascading risks across interdependent infrastructures. The framework utilizes natural hazard footprints and represents cascading failures using Boolean expressions that define the operational states of infrastructure components. Although existing frameworks for assessing interdependent power and communication networks provide valuable insights, they often rely on static models that do not capture the spatiotemporal dynamics of hazard progression and do not incorporate real-world infrastructure configurations. These models typically assume a randomly initiated event that triggers cascading failures within the interdependent system. Furthermore, while some frameworks represent the coupling of power and communication networks using network graphs, they often employ an overly simplistic one-to-one mapping, wherein each power network node connects to a single communication network node. This assumption lacks realism and does not accurately quantify the actual impact. In contrast, a limited number of studies have explored more flexible coupling—such as many-to-one mappings—they still fall short by excluding dynamic hazard modeling and of- ten overlook the spatial co-location of infrastructure assets, such as electrical poles and communication towers, which is crucial for modeling physically-driven failure propagation [11]. Additionally, population-level service disruptions, such as communication coverage loss, are rarely quantified, and communication networks are frequently abstracted or limited to control data exchange, omitting their broader role in societal connectivity. To address these limitations, this paper proposes a novel graph-based framework for quantifying the cascading impacts of natural hazards on interdependent power and communication networks. The contributions of this paper are as follows: •Graph-based modeling approach: Develops an integrated framework capturing spatial interdependencies between power and telecommunication networks by coupling substations and towers along electrical poles. It also integrates a signal propagation model to assess tower coverage under disruption. •Spatiotemporal Hazard Simulation: Models windstorm dynamics to capture cascading failures across the interconnected power-telecommunication network. •Resilience Quantification: Assesses system performance using demand not served (DNS), asset failure rates in both power and telecommunication networks, and the population impacted by telecommunication service disruptions. The remainder of this paper is organized as follows: Section II introduces the Interdependent Infrastructure Resilience Assessment Framework (IIRAF), which is specifically designed to evaluate cascading failures and resilience in spatially coupled power and communication networks. Section III presents the results of the case study and provides corresponding discussion. Finally, Section IV concludes the paper with key findings and directions for future research. II. INTERDEPENDENT INFRASTRUCTURE RESILIENCE ASSESSMENT FRAMEWORK The proposed framework for assessing the resilience of interdependent power and communication networks is illustrated in Fig. 1. The framework comprises several key components: it begins with data collection and pre-processing, followed by the spatial coupling of electricity substations and telecommunication towers. A spatiotemporal hazard scenario generator simulates windstorm events, dynamically propagating failures through the network. The failure propagation module identifies affected infrastructure, while a signal propagation model evaluates telecommunication tower coverage over time, including disruption periods. Finally, the impact is quantified using sector-specific metrics, including DNS, asset failure rates in both power and telecommunication network, and the estimated population affected by communication disruptions. Each component is explained in detail in the subsections. A. Data Collection and Pre-processing The implementation of the proposed framework requires several types of input data spanning infrastructure systems, Fig. 1. Framework for Assessing the Resilience of Interdependent Power and Communication Networks. spatial boundaries, demographic information, and hazard records. These include: •Power Network Data: This includes the topology of the electricity network, such as the location of substations and electrical lines. These data are essential for constructing the graph representation of the power system and modeling failure propagation. Furthermore, electrical pole data is collected and utilized to connect telecommunication towers with electrical substations. •Telecommunication Network Data: This consist of the geographic coordinates of telecommunication towers, their structural attributes (e.g., tower height), and the types of services they provide, including GSM, UMTS, LTE, and 5G. This information supports the spatial coupling of power network and the signal coverage modeling. •MultiPolygon Spatial Boundaries: This data refers to geospatial representation of administrative regions or zones. These are used to define the spatial units for impact analysis and to associate infrastructure elements with specfic regions. •Population Statistics: This data includes population counts associated with the defined multipolygon regions, enabling the estimation of the population affected by telecommunication service disruptions. •Historical Hazard Events: This data consists of historical records of windstorms events, which serve as the basis for generating spatiotemporal hazard scenarios in the simulation. During the data pre-processing stage, spatial datasets—including telecommunication tower locations, power infrastructure elements, population points, and multipolygon boundaries—are cleaned by removing duplicates, correcting invalid geometries, handling missing or inconsistent attribute data, and ensuring proper topological relationships (e.g., no overlaps or gaps where inappropriate). All datasets are then aligned within a consistent coordinate reference system to ensure spatial compatibility. Commonly used projections include EPSG:4326 (WGS 84) for global geographic coordinates and region-specific UTM projections (e.g., EPSG:32636 for Cyprus) for accurate distance-based computations. The selection of the coordinate system depends on the study area’s geographic context. Furthermore, invalid geometries, if present, are corrected using operations such as buffering. Where necessary, spatial joining techniques are applied to map demographic data to administrative regions. These pre-processing steps are adaptive and can be adjusted based on the structure and format of the available datasets. B. Spatial Coupling of Power and Communication Network This study presents a spatially explicit, graph-based methodology for integrating power and telecommunication networks based on proximity relationships between critical assets. The model focuses on electrical substations and telecommunication towers, given their key interdependence for service continuity. Efficient spatial mapping is achieved via the cKDTree algorithm from SciPy, which enables fast nearest-neighbor searches on large geospatial datasets [12]. A pivotal step in the integration process is the spatial coupling of telecommunication towers to electrical substations, which determines infrastructure dependencies. This is accomplished through a hybrid approach (illustrated with an example outcome in Fig. 2) that prioritizes connectivity through electrical poles. Initially, each substation is associated with its nearest electrical pole to establish a graph structure. Subsequently, for each tower, the shortest path to the substation is computed using the graph. If a valid path exists and its total length falls within a predefined threshold (e.g., 5 units), the tower is assigned based on the electrical pole route. In the absence of a valid path, the algorithm resorts to checking the Euclidean distance between the electrical substation and the telecommunication tower. A direct link is established only if this distance also complies with the threshold condition. This hybrid approach (or the two-level assignment logic) balances spatial realism with robustness, particularly in cases where direct coupling is unavoidable due to fragmented or sparse network structures. The proposed hybrid approach is depicted in Fig. 2, which presents three illustrative example cases. In the first case, a telecommunication tower (Tower A) is connected via a valid graph path composed of intermediary electrical poles, with a total distance of 4.52 units falling below the defined threshold of 5 units. In the second case, a tower (Tower B) lacks a valid path via electrical poles but is close enough to the electrical substation to be assigned through direct coupling with a distance of 3.54 units. In the third case, a tower (Tower C) which has a path via electrical poles but its total length (6.94 units) exceeds the allowable threshold. Consequently, Tower C is not connected to the substation. This highlights the significance of the distance constraint not only in enabling fallback links, but also in validating graph-based paths. C. Spatiotemporal Hazard Scenario Generation The hazard scenario generator simulates storm-induced failures in the power network using wind fragility curves for Fig. 2. Conceptual Illustration of Hybrid Spatial Coupling between an Electrical Substation and Telecommunication Towers. electrical lines, following approaches similar to [13]. The simulation initializes by loading network geodata, defining a buffered boundary, and setting up a geodetic calculator using the WGS 84 coordinate system—ensuring accurate distance calculations while improving computational efficiency. Input parameters include wind speed limits (from historical data), restoration times, fragility curve statistics, and the number of scenarios. Based on these, the hazard scenario generator produces random storm trajectories with variable headings and radii, simulating storm propagation across the coupled power and communication network. The storm progresses over a fixed time horizon (e.g., 24 hours), advancing in discrete time steps with displacement calculated based on its initial and final positions. At each timestep, the storm’s impact on the electrical lines is evaluated by checking whether each lines lies within the storm radius and estimating wind speeds at the line midpoints. The failure probability of each line is then assessed using a log-normal fragility curve, as illustrated in Fig. 3 and defined mathematically in (1). This curve models the probability of failure as a function of storm gust speed w, characterized by a mean µ and standard deviation σ, as in [14]. Phw L(w) = Zw 0 1 xσ√2πe−(ln x−µ)2 2σ2dx (1) The wind-dependent failure probability of electrical lines is further defined in (2), where failure probability is 0 below a critical wind speed wcritical, gradually increases with a defined range, and reached certainty at the collapse wind speed wcollapse. When an electrical line fails, it remains inoperative for a randomly assigned restoration time before restored. To enhance clarity, Fig. 3 also marks the critical wind speed wcritical, below which no failure occurs, and the collapse wind speed wcollapse, above which failure is certain. Fig. 3. Wind Fragility Curve [14]. PL(w) =      0, w ≤wcritical Phw L, wcritical ≤w≤wcollapse 1, w ≥wcollapse (2) The outcome of this scenario generator encompasses the line status, indicating whether lines are operational (1) or failed (0). Additionally, it provides storm details such as the storm gust speed, storm radius, and its spatiotemporal path. D. Failure Propagation and Impact Quantification The failure propagation and impact quantification approach outlined in Algorithm 1 employs a three-phase process. In the initial phase, it processes the power line status over time to ascertain the operational state of electrical substations. Disconnected substations and their dependent telecommunication towers at timestep tare stored in Pt Nand Tt N, respectively, where Ndenotes the set of nodes and tthe discrete time index. Each tower possesses a backup power duration h, and towers disconnected beyond this duration are designated as non-operational. This phase initiates the cascade by identifying failed nodes within the interconnected power and communication network. In the second phase, the method assesses the resulting loss of communication coverage for the failed towers in Tt N. For each tower and each communication service it supports (e.g., GSM, UMTS, LTE, 5G), the transmit power is computed using physical parameters: the number of transceivers ntrx, transceiver power Pmax, power amplifier efficiency ηPA, and feeder loss σfeed. The per-transceiver power is given by (3). PAW=Pmax ηP A.(1 −σfeed)(3) The total transmit power is calculated as TP AW= ntrx.P AWand converted to dBm for compatibility with path loss models. The appropriate propagation model is selected based on the service frequency: the Extended Hata model for sub-3 GHz services and the Rapport model for higher Algorithm 1 Structured Algorithm for Failure Cascade and Impact Analysis Require: Power line status over time, network topology, coupling data Ensure: DNS, failed power/telecom assets, affected population 1: Phase 1: Failure Propagation from Power to Communication Network 2: for each timestep tdo 3: Identify disconnected substations →Pt N 4: Cross-reference Pt Nwith telecom tower couplings → Tt N 5: for each tower in Tt Ndo 6: if outage duration >backup threshold hthen 7: Mark tower as failed 8: end if 9: end for 10: Phase 2: Service Loss and Coverage Estimation 11: for each tower in Tt Ndo 12: for each service ∈ {GSM, UMTS, LTE, 5G}do 13: Compute PAW=Pmax ηPA·(1−σfeed) 14: Compute TP AW=ntrx ·PAW 15: Convert TP AWto dBm 16: Select propagation model (Hata/Rapport) 17: Binary search for coverage radius 18: Create circular buffer 19: end for 20: end for 21: Phase 3: Impact Quantification 22: Run DC-OPF to compute DNS at time t 23: Count failed lines and towers 24: Call Algorithm 2 to estimate affected population 25: end for frequency bands. These models account for distance, antenna height, frequency, and environmental characteristics. A binary search algorithm determines the maximum coverage radius where the received signal matches the receiver sensitivity threshold. This radius is represented as circular geospatial buffers centered at tower coordinates. In the third phase, the model evaluates the integrated network-wide and societal impacts. DNS is calculated using a DC Optimal Power Flow (DC-OPF) analysis that estimates the total unmet load following power line failures. The number of failed power assets is directly derived from the scenario outputs, while failed telecommunication towers exceed their backup power threshold. To estimate the affected population, the model employs a geospatial overlay technique. It merges the coverage buffers of all non-operational towers into a single polygon representing lost coverage at a specific time t. This polygon is overlayed with population datasets containing region boundaries and population counts. For each intersecting region, the area of overlap is computed and utilized to determine a coverage ratio, which is subsequently scaled to account for the region’s population. Summing these values across Algorithm 2 Geospatial Estimation of Affected Population Require: Lost coverage areas at time t, population regions with counts Ensure: Estimated affected population at time t 1: Merge coverage buffers of failed towers into a single polygon Ct 2: Overlay Ctwith population regions 3: for each intersected region rdo 4: Compute overlap area Aoverlap between Ctand region r 5: Compute coverage ratio ρr=Aoverlap Aregion 6: Compute affected population paffected r=ρr·pr 7: end for 8: Sum all paffected rto get total affected population all intersecting regions yields the estimated total population impacted by communication outage at each time step. III. CASE STUDY AND SIMULATION RESULTS A synthetic power network is constructed by merging two standard IEEE 33-bus test systems and projecting onto Cyprus. This spatial coupling of power and communication infrastructure enables cascading failure analysis while maintaining a generalized and reproducible model. Due to confidentiality, real grid data is not shared; however, publicly available geospatial data from [15] and the Cyprus Digital Twin (CyDT1) platform have been utilized to inform the spatial layout. Infrastructure data were accessed via CyDT APIs and integrated into the simulation. The CyDT, developed in-house, is a digital platform aimed at enabling spatial data integration and simulation capabilities. The long-term objective is to expand the presented scenario within the CyDT framework, engaging relevant stakeholders and supporting broader infrastructure resilience planning. The resulting configuration supports interdependent simulations between critical power and communication nodes. A visual representation of the projected infrastructure is provided in Fig. 4. The proposed framework is tested using the power and communication network shown in Fig. 4. As outlined in Section II-B, the power and communication networks are coupled by considering a threshold distance of 2000 meters between the substation and telecommunication towers. The resulting coupled network is subjected to hazard scenarios generated by the scenario generator in Section II-C over a specified time horizon. Windstorm scenarios with varying intensities and trajectories are created, considering windstorm speeds between [40, 60] m/s. Restoration times for failed power lines are randomly assigned within [4, 8] hours. A log-normal fragility curve with a mean of 3.5 and standard deviation of 0.1 models wind-induced failures. Each windstorm scenario spans a 24hour simulation window. Following the generation of hazard scenarios, failure propagation analysis is conducted as outlined in Section II-D. This analysis evaluates the impact of the coupled network 1Cyprus Digital Twin (CyDT) overview: https://youtu.be/GJfvOiLpGak over the specified time horizon, T. For each windstorm scenario, resilience metrics—including DNS, percentage of failed electrical lines, percentage of disconnected telecommunication towers, and the affected population due to communication failures—are assessed at each time interval, t. A representative scenario is generated to illustrate these metrics over a time period of T= 24 and the corresponding results obtained are shown in Fig. 5. The resilience curve, presented in Fig. 5a, shows the percentage of Demand Connected, while Fig. 5b depicts the percentage of failed electrical lines over time. These metrics reflect the impact on the power network and remain unchanged across both conditions—with and no backup power at telecommunication towers—since the backup only influences the communication network. The impact on the communication network is consolidated and presented in Fig. 5c and Fig. 5d, which compares both scenarios: with and without backup power. It illustrates the percentage of disconnected telecommunication towers and the corresponding population affected over the time horizon. As depicted in Fig. 5, the windstorm had a substantial impact on the power network, resulting in over 30% of electrical lines experiencing failures between the 9th and 12th hours. Consequently, the overall power line failures led to a decline in the Demand Connected from 100% to approximately 40% during this critical period. This substantial reduction in the system’s ability to meet electricity demand underscores the vulnerability of the power infrastructure to such events. The consequences of these failures cascaded into the communication network. In the absence of a backup system, power outages at electrical substations resulted in the disconnection of approximately 14% of telecommunication towers. Consequently, communication outages affected over 130,000 individuals. These disruptions commenced at approximately t = 6 hours and persisted for nearly twelve hours. In contrast, the scenario with backup power for communication towers, where each tower has h= 3 hours of backup duration, demonstrates a delay in the onset of communication disruption. As shown in Fig. 5c and Fig. 5d, the first significant disconnections occur at t= 9 hours instead of t= 6 hours. This three-hour shift aligns with the modeled backup duration and effectively postpones the communication impact. However, the maximum level of disruption remains similar, as the backup only provides temporary support. Once the backup duration is exceeded and power is still unavailable, towers begin to fail similarly to the no-backup case. The deployment of backup power systems in critical telecommunication infrastructure effectively mitigates temporal impacts by providing a valuable delay—though not eliminating peak impact—that enhances emergency response and public safety during the initial stages of cascading failures. IV. CONCLUSION The proposed framework integrates a hazard scenario generator with a coupled power and communication network model to evaluate cascading failures during extreme weather events. It incorporates a telecommunication signal propagation model Fig. 4. Map of a Synthetic Power Network Featuring Telecommunication Towers and Electrical Poles. Fig. 5. Resilience Metrics: (a) System Performance (b) Electrical Asset Failures Over Time (c) Telecommunication Towers Lost and (d) Population Affected Over Time with 3-Hour Backup Power to estimate dynamic coverage loss from tower failures and assess the population affected by communication outages via geospatial analysis. Simulation results show that severe windstorms can cause extensive failures in the power grid, which subsequently propagate into the communication network, underscoring the vulnerability of interdependent infrastructure. The incorporation of backup power at telecommunication towers demonstrates mitigation of the onset of service disruptions, providing a critical buffer period that can minimize the immediate impact on affected populations. While focused on impact quantification rather than vulnerability assessment, the framework supports identifying critical assets across networks through diverse hazard scenarios. Future work will address backup power uncertainty, redundant energy supply efficacy, and bidirectional cascading effects to enhance infrastructure resilience and inform disaster preparedness, with planned integration into CyDT. REFERENCES [1] S. Lam, “Power out risks rising! 5 ways extreme weather causes blackouts,” Sep. 2024, accessed 2025-02-28. [Online]. Available: https://bit.ly/3Dc3Um0 [2] H. Zhang, B. Liu, and H. Wu, “Smart grid cyber-physical attack and defense: A review,” IEEE Access, vol. 9, pp. 29 641–29 659, 2021. [3] X. Liu, B. Chen, C. Chen, and D. Jin, “Electric power grid resilience with interdependencies between power and communication networks–a review,” IET Smart Grid, vol. 3, no. 2, pp. 182–193, 2020. [4] A. Salehpour, I. Al-Anbagi, K.-C. Yow, and X. Cheng, “Modeling cascading failures in coupled smart grid networks,” IEEE Access, vol. 10, pp. 81 054–81 070, 2022. [5] R. Atat, M. Ismail, S. S. Refaat, E. Serpedin, and T. Overbye, “Cascading failure vulnerability analysis in interdependent power communication networks,” IEEE Systems Journal, vol. 16, no. 3, pp. 3500–3511, 2021. [6] W. Huang, T. Zhang, and X. Yao, “Optimization for sequential communication line attack in interdependent power-communication network,” Physica A: Statistical Mechanics and its Applications, vol. 592, p. 126837, 2022. [7] T. Liu, R. Sun, Y. Liu, and C. Wang, “A resilience enhancement scheme of cyber-physical power system for extreme natural disasters,” in 2020 IEEE Sustainable Power and Energy Conference (iSPEC). IEEE, 2020, pp. 1684–1689. [8] M. Lotfi, M. Panteli, I. L. Wells, K. B. Laskey, J. W. Van De Lindt, Y. Amir, A. Babay, and I. Chiu, “A resilience assessment framework for coupled power and communication infrastructure,” in 2023 IEEE Power & Energy Society General Meeting (PESGM). IEEE, 2023, pp. 1–5. [9] Q. Zhang, S. Yao, Q. Zhao, M. Zhang, R. Ma, and Z. Ren, “Critical nodes identification for power communication network under electrical interdependent effects of smart grid,” in 2022 IEEE 22nd International Conference on Communication Technology (ICCT), 2022, pp. 1230– 1234. [10] E. M¨ uhlhofer, E. E. Koks, C. M. Kropf, G. Sansavini, and D. N. Bresch, “A generalized natural hazard risk modelling framework for infrastructure failure cascades,” Reliability Engineering & System Safety, vol. 234, p. 109194, 2023. [11] X. Lu, K. Hinkelman, Y. Fu, J. Wang, W. Zuo, Q. Zhang, and W. Saad, “An open source modeling framework for interdependent energy-transportation-communication infrastructure in smart and connected communities,” IEEE Access, vol. 7, pp. 55 458–55 476, 2019. [12] S. Documentation, “scipy.spatial.ckdtree,” Online Documentation, accessed: March 28, 2025. [Online]. Available: https://bit.ly/4iJdL26 [13] M. Panteli, C. Pickering, S. Wilkinson, R. Dawson, and P. Mancarella, “Power system resilience to extreme weather: Fragility modeling, probabilistic impact assessment, and adaptation measures,” IEEE Transactions on Power Systems, vol. 32, no. 5, pp. 3747–3757, 2016. [14] S. Dunn, S. Wilkinson, D. Alderson, H. Fowler, and C. Galasso, “Fragility curves for assessing the resilience of electricity networks constructed from an extensive fault database,” Natural Hazards Review, vol. 19, no. 1, p. 04017019, 2018. [15] Government of Cyprus, “Open Data Portal of Cyprus,” 2024, accessed: 2024-02-26. [Online]. Available: http://data.gov.cy