the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 29 Jul 2026
Review article: How hazard-related disruptions become system-wide crises: event-level evidence on failure pathways and continuity functions, 1980–2025
Abstract. Cascading disasters are increasingly recognized as systemic escalation processes, yet comparative evidence on how initial hazard impacts are transformed into wider crises remains fragmented. Resilience dimensions clarify how system weaknesses interact with triggering hazards and determine whether disruption is absorbed, contained, or transmitted. Drawing on a PRISMA-guided search and targeted supplementary searches, this review synthesizes 43 event-level cascading-disaster cases from 1980 to 2025 to identify failure pathways linking hazards, failure mechanisms, and resilience dimensions. Each event was coded for initiating hazards and transformed hazards, propagation stages, escalation points, failure mechanisms, impacts, and resilience dimensions. Results show that technological and infrastructure failures were the most recurrent transformed hazards, while lifeline failure and secondary hazards formed the main escalation points. Cascade severity depended less on the length of the propagation chain than on which critical functions failed and how disruption propagated through dependent systems. Failure mechanisms clustered around monitoring/control, transport/access, coordination, lifeline services, exposure management, and health-system protection. The resilience crosswalk revealed an interdependence–redundancy gap in which connected systems lacked fallback capacity. These pathways produced persistent recovery burdens. The review proposes a continuity-function framework for cascade-risk reduction, emphasizing monitoring and control, lifeline services, access and logistics, emergency coordination, and equity-sensitive social protection.
- Preprint
(1565 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 15 Sep 2026)
-
CC1: 'Comment on egusphere-2026-3753', Ricardo Tavares da Costa, 04 Aug 2026
reply
-
AC1: 'Reply on CC1', Homa Bahmani, 05 Aug 2026
reply
Thank you for this important comment. We agree that the previous Methods section did not explain sufficiently how Alexander’s cascading-disaster magnitude scale was applied in practice. We will therefore clarify the operational definitions of Levels M0–M5 and the event-level classification procedure in the revised manuscript.
The scale was applied as a qualitative ordinal classification of cascade complexity rather than as an additive numerical score. Following Alexander (2018) and its event-level application by Suppasri et al. (2021), classification was based on three structural components: causes, chains of effects, and escalation points. In our application, initiating and transformed hazards represented the primary and subsequent causes; coded propagation pathways represented chains of effects; and escalation points represented critical junctures at which a hazard, infrastructure failure, or institutional failure generated consequences substantially greater than the initiating impact alone.
The levels were operationalised as follows: M0 represented one simple cause and effect; M1 one cause and one chain of effects; M2 one cause and two or more chains of effects; M3 two causes, multiple chains of effects, and one significant escalation point; M4 two causes, multiple chains of effects, and at least two significant escalation points; and M5 multiple interacting causes, multiple chains of effects, and multiple escalation points producing exceptionally extensive systemic consequences.
Each event was reconstructed from the available evidence and matched to the level that best represented its complete cause–effect–escalation configuration. The classification was initially provisional and was reconsidered after targeted supplementary searches when the cascade sequence, escalation points, or downstream consequences remained uncertain. Information not reported in the evidence was not inferred, and evidence completeness was assessed separately from cascade magnitude.
We also agree that the scale does not provide independent numerical measures of propagation breadth and length. Sequential development is represented through chains of effects, while branching, cross-system spread, and the activation of additional failure pathways are represented through multiple chains and escalation points. These characteristics are therefore integrated within the final qualitative M-level rather than quantified as separate dimensions.
To examine whether the classification was functioning mainly as a measure of chain length, we separately counted the number of causally supported propagation steps. M4–M5 cases had a higher mean number of steps than M3 cases, but the difference was not statistically significant (7.57 versus 6.47; H = 2.53, p = 0.112). This supports the interpretation that higher M levels were not assigned solely because an event contained a longer sequence of effects.
The Methods section will be revised as follows:
After coding the primary source pool identified through PRISMA, each event was assigned a provisional cascade magnitude using Alexander’s cascading-disaster scale (Alexander, 2018), operationalized following its event-level application by Suppasri et al. (2021). The scale was applied as a qualitative ordinal classification of cascade complexity rather than as an additive numerical score. It distinguishes events according to the configuration of causes, chains of effects, and escalation points. In this study, initiating and transformed hazards were treated as primary and subsequent causes, coded propagation pathways represented chains of effects, and escalation points represented critical junctures at which a hazard, system failure, or institutional failure produced consequences substantially greater than the initiating impact alone.
The levels were interpreted as follows: M0 represented one direct cause and effect; M1 one cause and one chain of effects; M2 one cause and two or more chains of effects without a clearly established escalation point; M3 multiple causes or transformed hazards, multiple chains of effects, and one significant escalation point; M4 multiple chains of effects and at least two significant escalation points; and M5 multiple interacting causes, multiple chains, and multiple escalation points producing exceptionally extensive systemic consequences. Each event was matched to the level that best represented its complete cause–effect–escalation configuration. Borderline classifications were revisited after targeted supplementary searches, and information not reported in the evidence was not inferred.
The scale does not provide separate numerical measures of propagation breadth and length. Sequential propagation is reflected in chains of effects, while branching, cross-system spread, and the activation of additional failure pathways are reflected in multiple chains and escalation points. Consequently, the M level was treated as a holistic measure of cascade complexity rather than as a measure of chain length alone. Propagation depth was therefore assessed separately through the number of causally supported propagation steps. Cases classified as M3 were treated as moderate cascades, while M4 and M5 cases were combined as high-complexity cascades for group-level analysis; the exact event-level classifications were retained in the database. Cases classified as M0–M2 were excluded because they did not exhibit sufficient cascade complexity to meet the objectives of this review.
Citation: https://doi.org/10.5194/egusphere-2026-3753-AC1
-
AC1: 'Reply on CC1', Homa Bahmani, 05 Aug 2026
reply
-
RC1: 'Comment on egusphere-2026-3753', Anonymous Referee #1, 20 Aug 2026
reply
Specific observations:-
The second sentence of the abstract confuses resilience with vulnerability, which is [almost] its opposite.
Lines 24-31: This paragraph states things that are self-evident. It could usefully be deleted.
Lines 41-42: This was a point made by Pescaroli & Alexander (2018, Figure 2, p. 2253).
Lines 32-45: This paragraph would be better if the analysis were conducted from the point of view of vulnerability rather than hazard.
Lines 60-61: "...the methods used to describe and map cascades remain heterogeneous." - This is a very good point, but the subject matter is heterogeneous as well!
Line 64 and elsewhere: by territorial you mean geographical. There is a subtle difference.
Line 81: It is not clear what 'standardized cross-case empirical coding' involves. If practical studies of actual cascades are in short supply it would be difficult to develop norms from a body of evidence.
Lines 90-91: "...a broad comparison across many types of real-world cascading disasters is still needed to better understand and manage cascade events." - That is exactly what the cascading disasters magnitude scale was designed to facilitate.
Line 97: The trouble with Hagen's categories is that most of them do not betoken cascading effects at all, merely interruptions in continuity of events or decisions.
Line 106: I recommend you read Ken Hewitt's 1983 book, which established the primacy of vulnerability over hazard long before Kelman dealt with it. This was the so-called 'radical critique'. It still has plenty of lessons for the present day.
Lines 111-113: Resilience cannot be applied if vulnerability is not understood first.
Lines 117-118: "...resilience dimensions are therefore used as an analytical bridge between observable failure mechanisms and the deeper system weaknesses that allowed those failures to propagate" - This is a thoroughly muddled statement.
As the cascading disasters magnitude scale shows, not all cascading effects involve escalation. Rather than assuming that it occurs, it would be better to analyse why it does in the cases where it can readily be identified.
Line 131 et seq.: It beats me as to why PRISMA is conjured up so often in the social sciences when it was specifically designed to evaluate medical research. Full use of the PRISMA methodology involves looking for weaknesses in empirical studies that might bias results and lead to unwanted outcomes in patients. In fact, this paper does not use PRISMA as described by its originators in their 2020 article in the Annals of Surgery. Hence, the description of methodology is misleading.
Figure 1 needs improvements in legibility.
Figure 2a refers only to the cases extracted from the literature and is not a portrait of cascading disasters worldwide. It is difficult to see what point can be made from it.
Regarding Figure 2b, my understanding was that the author wanted to get away from a hazard-based interpretation of cascading disasters, and yet here the cascade seems to be from one type of hazard to another.
Figure 2c is a table.
Line 262 et seq.: In this paragraph there is a great confusion between causes and effects. Although in cascading disasters an effect can become a cause, to achieve a reasonable level of explanation there needs to be much better separation between the two.
I have struggled unsuccessfully to understand Figure 3. It does not have a clear message. It also seems to be critically dependent on a limited data set. It may well be that the results would be broadly similar if the data set were very much larger, but that is difficult to argue without more evidence. The figure also needs serious improvements in legibility.
Lines 293-294: "...ongoing hazard exposure, reconstruction, infrastructure restoration, and spatial recovery often combined to form a recovery burden." - Yes, but this is hardly novel. We have known it for many decades.
Lines 296-297: "Prolonged infrastructure recovery/resilience was observed in 45.8% of high-complexity cascades" - Complex cascades are probably inevitable in large disasters, which, by virtue of their size, take longer and require more resources to remedy.
Line 301, and elsewhere: The term 'spatial recovery needs' crops up at intervals in this paper. What does it mean? Cascading disasters produce panarchy, which occurs at several spatial scales simultaneously.
Line 317: "failure mechanisms capture the functional pathways of failure" - this is a self-referencing statement.
Line 396: "Closely related to lifeline failure, road and transportation network failure" - roads and other transportation corridors are lifelines!
Lines 455-457: "The central policy recommendation of this review is that cascade-risk reduction should not be organized solely around sectors, but around the system functions that prevent disruption from spreading across sectors." - That is a very valid point, but it needs to be illustrated with proper examples (which it isn't).
Line 486: "recovery planning should not assume that risk disappears once the triggering hazard has ended" - who says it does? The whole principle of the very popular slogan 'build back better' is predicated on assumptions of continuing hazard or threat.
Overall evaluation:-
Although the methodology is unusual, most of the observations and conclusions in this paper are not novel. The paper does not adequately bring out which of its findings are innovative, and there are some.
As interactions among different kinds of vulnerability are the main source of escalation in cascading disasters, this paper needs more emphasis on vulnerability, as this is the main explanatory variable. The paper tends to use resilience as a surrogate for vulnerability, which is confusing and perhaps misleading.
One major problem is that the consequences of disaster are critically dependent on the context in which it occurs. We can define 'context' as the social, economic, cultural, psychological and environmental milieu that surrounds disaster risk and to some degree interacts with it. If necessary, we can disaggregate different types of context. However, overall, context should be considered as the sum of elements that have no direct causal relationship with disaster but, paradoxically, are (or should be) essential to any attempt to explain it. For example, how serious is the failure of electricity supply to the 900 million people in the world who have no access to electricity, or the 1.8 billion whose access is limited? There are many other examples. This suggests that a scenario-based approach to cascades is still very much needed.
The sort of macroscopic analysis given in this paper does not identify cascades, but merely causes and effects. There is little actual analysis of the propagation of cascades through chains and networks, yet the most useful thing that could be done is to find unusual and unexpected consequences and explain them in terms of cascading disaster.
Hence, I think that this paper is a brave attempt to advance our understanding of cascading disasters, but it is methodologically flawed. For example, as noted above, cascades of high complexity prolong the recovery of infrastructure - but what about the size of the disaster and the context of investment, political priorities, power relations and all the rest?
Citation: https://doi.org/10.5194/egusphere-2026-3753-RC1
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 110 | 31 | 13 | 154 | 21 | 15 |
- HTML: 110
- PDF: 31
- XML: 13
- Total: 154
- BibTeX: 21
- EndNote: 15
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Please report how M-level assignment was operationalised in practice and how each was placed on the 0–5 scale. This would let readers see whether the scheme actually separates breadth and length or conflates them.