48 of 97 Active Disasters Are in Compound Cascades — May 2026 GDACS Audit

We built gdacs-disaster-alerts to turn the GDACS global disaster feed into operational intelligence with correlation analysis, geographic clustering, incident continuity, and per-event hazard intelligence. To test its compound-cascade detection on the live feed, we ran it on 11 May 2026 against the full UN OCHA + EU JRC GDACS stream with no decision-floor filter. The 97 active disasters it returned, and the 10 named cascades the correlation engine grouped them into — including an 11-incident Southeast Asia Tropical Cyclone Cascade that mainstream coverage reported as nine separate named storms — are the data this post documents.

The problem: Disaster reporting is built around the named event. One named cyclone, one named earthquake, one named wildfire — each gets its own story, its own death toll, its own landfall map. The framing is a habit borrowed from the meteorological agencies, who name events for tracking, not for narrative. The cost is that emergency-coordination realities — multi-hazard clusters, cross-border events, overlapping landfalls in the same week within the same geographic basin — disappear into the gaps between stories. A 2026 ApifyForge audit captured the answer for the live GDACS feed on a single date.

What is GDACS? The Global Disaster Alert and Coordination System is a cooperative framework run by the United Nations Office for the Coordination of Humanitarian Affairs (UN OCHA) and the European Commission's Joint Research Centre (EU JRC). It publishes a continuously updated machine-readable feed of active earthquakes, tropical cyclones, floods, wildfires, volcanoes, and droughts that meet international humanitarian-coordination thresholds, with population-exposure proxies and severity tiers (Green / Orange / Red) assigned per event. The official site is gdacs.org.

Why it matters: GDACS is the closest thing to a canonical "what is currently happening, at what scale" feed for the international humanitarian system. Its alert tiers route attention; its event IDs are the cross-reference key UN agencies, the European Civil Protection Mechanism, and the major INGOs use to coordinate. Treating its feed as a portfolio rather than as a wire of standalone alerts surfaces structural patterns — compound cascades, regional saturation, multi-region critical activity — that are invisible at the per-event level.

Use it when: reporting on a specific natural disaster's regional context, briefing humanitarian-response or supply-chain teams on multi-hazard exposure, sourcing a reproducible quantitative anchor for "is this a bad week globally?" coverage, or producing a snapshot of which regions the international system is currently treating as overloaded.

Key findings

• 97 active disasters globally on the GDACS feed as of 11 May 2026 — 19 Red-tier and 78 Orange-tier alerts.
• 48 of 97 events (49.5%) are members of a compound cascade — a geographic cluster containing 2+ linked incidents within ~500km.
• 10 named compound cascades were detected by the correlation engine; the largest contains 11 linked incidents.
• 7 of 11 GDACS regions are classified as overloaded — South America, Africa, Southeast Asia, South Asia, East Asia, North America, and Europe.
• The hazard mix is 29 Floods (30%), 26 Earthquakes (27%), 20 Tropical Cyclones (21%), 13 Wildfires (13%), 5 Volcanoes (5%), and 4 Droughts (4%) — 60 unique country / country-group combinations affected.
• The portfolioPressureIndex is 100/100 and the system has fired three systemic-risk signals: multi-region-critical-activity, multi-region-cyclone-activity, and seismic-cluster-detected.

In this article: What's a cascade · The 10-cascade leaderboard · Per-region breakdown · Southeast Asia cyclone cascade · Aegean cluster · Hindu Kush · Iberian wildfires · Melissa singleton · Methodology · Caveats · Press lift-out · FAQ

What is a compound disaster cascade?

Definition (short version): A compound disaster cascade is a group of two or more active disasters whose epicentres or impact centroids fall within a single geographic cluster — in this audit, within ~500km of a shared centroid — and which are therefore treated by GDACS-correlation tooling as one operational situation rather than as independent events, regardless of whether they share a hazard type.

The cascade concept matters because it changes the unit of analysis. Newsrooms report disasters per named event: one cyclone, one earthquake, one flood. Emergency-coordination bodies operate on regional pressure: how many simultaneous incidents are stretching the same response infrastructure in the same geographic basin. Those two units rarely line up. A reporter covering "Tropical Cyclone FENGSHEN hitting Vietnam" is writing about one storm. The UN OCHA regional office in Bangkok is coordinating a response across nine named cyclones, two floods, and a volcanic eruption within the same six-month window in the same geographic cluster.

This audit captures the operational unit, not the named-event unit. The actor's correlation engine uses single-link clustering at a ~500km radius across all hazard types: any two events within that radius get grouped regardless of whether they share a hazard category. A drought, a flood, and a volcano in central Java would cluster together. An earthquake on the Türkiye side of the Aegean and a wildfire on the Greek side, 200km apart, cluster together. The clustering is geographically valid but not always operationally meaningful — a wildfire and an earthquake within 500km share emergency-coordination geography but not response playbooks. The methodology section discusses this tradeoff in detail.

Also known as: compound natural hazards, multi-hazard cascades, regional disaster clusters, simultaneous emergency situations, multi-incident operational situations, GDACS-correlated event clusters.

The 10-cascade leaderboard

Captured 11 May 2026 via the gdacs-disaster-alerts Apify actor.

A note on labels. The actor's raw output uses generic region tags for three of these clusters — "Other Wildfire Cascade Event" for the Aegean, "Other Earthquake Cascade Event" for the Hindu Kush, "Southeast Asia Earthquake Cascade Event" for the Central Philippines mix — because the actor's region-pack does not cover Mediterranean / Hindu Kush / Philippines as named regions. The descriptive labels in the table above are derived from each cluster's lat/lon centroid using standard place names. The companion CSV at /data/compound-disaster-cascades-may-2026/dataset.csv preserves both forms for reproducibility.

Aggregate stats:

• 10 of 15 detected geographic clusters qualify as named operational situations (cascades with 2+ linked incidents).
• 48 of 97 events (49.5%) are members of one of the 10 cascades.
• The largest cascade (Southeast Asia) contains 11 linked incidents across at least 8 distinct countries.
• 9 of 10 cascades have a peak severity of critical; only the Lake Kivu Flood Cluster sits at high.
• Three cascades mix hazard categories (the Aegean mixes EQ + WF; the Hindu Kush mixes EQ + FL; the Central Philippines mixes VO + EQ + FL).

Per-region breakdown

Seven of eleven regions are classified as overloaded on this date. The actor's regional-saturation classifier sets the overloaded threshold at active-event density relative to that region's baseline incidence — the actual threshold value is encoded in the region-pack and varies by region, because the same raw count means different things in Southeast Asia (where multi-cyclone seasons are seasonally expected) versus North America (where 3 simultaneous Red/Orange events is well above normal).

The headline cross-region finding is multi-region critical activity: a systemic-risk signal the actor fires when multiple regions simultaneously cross into the overloaded tier. It fired on this date alongside multi-region-cyclone-activity (the Pacific basin overloading on tropical-cyclone count) and seismic-cluster-detected (the Hindu Kush + Bangladesh / NE India earthquake clusters firing in the same window).

Story A — the 11-incident Southeast Asia cyclone cascade

The South China Sea / Vietnam coast cluster contains 11 linked incidents — nine named Pacific cyclones plus one flood and one cyclone-flood overlap, all clustered within ~500km of centroid 20.0N 109.9E across the Aug-Nov 2025 window. The nine named cyclones, with peak wind speeds from the US Joint Typhoon Warning Center (JTWC):

• RAGASA-25 (18-24 Sep 2025) — Philippines, China — peak 269 km/h (Cat 4+).
• FUNG-WONG-25 (4-13 Nov 2025) — Philippines, China, Japan — peak 213 km/h.
• KALMAEGI-25 (1-6 Nov 2025) — Vietnam, Cambodia, Philippines, Laos — peak 213 km/h.
• KAJIKI-25 (22-25 Aug 2025) — China, Vietnam, Laos — peak 176 km/h.
• MATMO-25 (1-5 Oct 2025) — China, Vietnam, Philippines — peak 167 km/h.
• PODUL-25 (7-13 Aug 2025) — Taiwan, China — peak 167 km/h.
• BUALOI-25 (23-28 Sep 2025) — Vietnam, Philippines, Laos — peak 148 km/h.
• WIPHA-25 (18-22 Jul 2025) — China, Vietnam, Philippines — peak 120 km/h.
• FENGSHEN-25 (18-23 Oct 2025) — Vietnam, China, Cambodia, Philippines, Laos, Thailand (6 countries) — peak 93 km/h.

Mainstream coverage reported these as nine separate named-cyclone stories — Reuters, AP, BBC, Al Jazeera each ran independent landfall maps, death tolls, and evacuation-order summaries per event, sometimes within the same week. GDACS classifies the whole sequence as one operational situation because every incident clusters within ~500km of the same centroid. The downstream emergency-coordination consequence — repeated landfalls in overlapping populations, downstream flooding accumulating across consecutive storms, response infrastructure stretched across multiple simultaneous evacuations — is invisible in the per-event framing.

Story B — the Aegean wildfire + Türkiye earthquake cascade

The Aegean cluster centres on the central Aegean Sea between the Greek islands and the Türkiye coast. Its 8 linked incidents mix wildfires across the Greek mainland and islands with earthquakes on the Türkiye side. The press framing splits these — "Greek wildfire season" and "Türkiye seismicity" are reported as independent stories — but the geography is one cluster. Two Türkiye earthquakes appear in this cluster alongside the Greek wildfire sequence.

The Aegean is one of the three cascades in this dataset where the actor's region label reads "Other Wildfire Cascade Event" in the raw output. The Mediterranean isn't covered as a named region in the current region-pack, so the cluster surfaces under the catch-all region tag with descriptive geography in the centroid. That's a labelling artefact of the actor build, not a data error — the underlying cluster is correctly detected and correctly contains the eight incidents listed.

Story C — the Hindu Kush earthquake cascade

The Hindu Kush cluster contains 4 earthquakes and 2 floods across Aug-Nov 2025:

• 31 Aug 2025 — M6.0 (lat 34.71, lon 70.79, depth 8km) — the headline event in this cluster; UN OCHA situation reports record 2,200+ fatalities (cross-referenced from reliefweb.int, not from the actor's output).
• 4 Sep 2025 — M5.6 (lat 34.72, lon 70.79, depth 10km) — first aftershock.
• 5 Sep 2025 — M5.2 (lat 34.55, lon 70.67, depth 10km) — second aftershock.
• 2 Nov 2025 — M6.3 (lat 36.59, lon 67.48, depth 28km) — second main shock north of Mazar-i-Sharif.
• Plus 2 flood incidents in the same clustering window.

Earthquake records come from the US National Earthquake Information Center (NEIC), cited per-event by the actor. The cluster is the second-deadliest cascade in the dataset by reported fatalities, behind only the Southeast Asia cyclone cascade's cumulative cyclone toll. Per-event detail for any single quake is available at the USGS earthquake catalog using the date and magnitude as the lookup key.

The Hindu Kush cluster surfaces under the "Other Earthquake Cascade Event" label in the raw actor output, again because the region-pack does not name central Afghanistan as a discrete region. The descriptive label in the leaderboard above is derived from the centroid; the CSV preserves both forms.

Story D — the Iberian wildfire cluster

The Iberian cluster is the only single-hazard cascade in the top 6 — five linked wildfire incidents on the Portugal-Spain border in northwest Iberia. It's the European-region cascade that ties this audit to US and European reader audiences: Iberian wildfires are now an annual feature of European disaster portfolios, documented in European Commission JRC wildfire reporting year after year, and the cluster pattern is the kind of cross-border, fire-season-scale event European Civil Protection coordinates on rather than national fire services individually.

The single-hazard framing makes Iberia operationally easier than the multi-hazard cascades — five wildfires share a response playbook in a way that a mixed EQ + WF cluster like the Aegean does not. But the cluster size matters for resource allocation: a single Iberian wildfire is a national event; five concurrent ones is a regional EU coordination event with cross-border resource sharing.

Story E — the Melissa singleton and the Atlantic cascade gap

Hurricane MELISSA-25 (21-31 Oct 2025) is the single highest-pressure event in the entire dataset by the actor's pressureScore metric (70/100), and is tied for the highest country-reach of any individual event at 6 countries. Yet MELISSA is not a member of any cascade. The Atlantic basin has zero compound cascades in this dataset — every Atlantic event in the sample is a standalone.

By contrast, the South China Sea / Vietnam coast is hosting 11 cascaded incidents. The Pacific is a cascade region in this snapshot; the Atlantic is a singletons region. The structural asymmetry is one of the most useful per-event readings in the dataset: Atlantic hurricane seasons produce headline-grabbing standalone events; Pacific typhoon seasons produce structural cascades. Per-event press attention is highly correlated with the Atlantic pattern; cascade attention is highly correlated with the Pacific pattern. The framing follows the storm distribution.

The other negative-space finding: TC FENGSHEN-25 hit 6 countries (Vietnam, China, Cambodia, Philippines, Laos, Thailand) — the same country-reach as Melissa — but FENGSHEN is one of 11 incidents in the Southeast Asia cascade, not a cascade lead in its own right. The press covered FENGSHEN as a tropical-depression footnote story; the data treats it as a major node in the largest active cascade. That asymmetry — same country-reach, very different press footprint, very different cascade role — is the kind of cross-event detail that only surfaces when the whole feed is read as a portfolio.

Methodology

• Phase 1 — Live-feed capture: The gdacs-disaster-alerts Apify actor (build 2.5, schemaVersion 2.5.0) was run on 11 May 2026 against the GDACS json-api endpoint at gdacs.org with hazardType: "all", alertLevel: "all", minSeverity: "low", dateFrom: null, dateTo: null. The endpoint returns the current active-events portfolio that UN OCHA and EU JRC are flagging for international humanitarian coordination. The run produced 97 alert records plus 1 batch-summary record. Run ID rRHpXtHjNKW1rQLXU, dataset ID TbefZLGr54UQPF12W.
• Phase 2 — Compound-cascade classification: The actor's internal correlation engine runs single-link geographic clustering at ~500km radius across all hazard types and surfaces clusters of 2+ linked incidents as named entries in the operationalSituations output block. The clustering is hazard-agnostic by design — a wildfire and an earthquake within 500km cluster together regardless of category.
• Phase 3 — Per-cluster enrichment: Each cluster's centroid lat/lon is extracted from operationalSituations[].centroid. The named-region label comes from the actor's region-pack lookup against the centroid; where the region-pack doesn't cover a centroid (Mediterranean, Hindu Kush, Central Philippines), the actor falls back to a generic "Other [Hazard] Cascade Event" label, which the post re-describes using standard place names.
• Coverage: primarySource: "json-api", primarySourceRichness: "full" — no GeoRSS fallback occurred on this run, so the dataset is at the actor's highest available fidelity. Source provenance is recorded per-event in each record's source and dataConfidence fields.
• Sample scope: This audit is the GDACS live feed only. Smaller-country domestic disasters not flagged by UN OCHA / EU JRC do not appear. The 97 events are the portfolio the international humanitarian system is tracking on this date — not every disaster active globally.
• Cross-reference paths: Each disaster carries a per-event gdacsUrl of the form https://www.gdacs.org/report.aspx?eventtype=EQ&eventid=NNNNNNN for direct GDACS lookup. Earthquakes cross-reference to the USGS earthquake catalog; cyclones cross-reference to JTWC; humanitarian impact figures cross-reference to UN OCHA at reliefweb.int.
• Reproduction: Anyone with an Apify account can re-run the actor against the same endpoint configuration any day and produce a current-as-of-that-date version of the same analysis. The output schema and operationalSituations block are stable across builds 2.4+.

The companion CSV (/data/compound-disaster-cascades-may-2026/dataset.csv) and the full 97-alert JSON (/data/compound-disaster-cascades-may-2026/raw.json) are downloadable from the source-actor banner at the top of this page for verification and reuse.

What the data does NOT support

Several inferences a journalist might want to draw from this dataset are not supported by it. Honest framing requires acknowledging them upfront.

1. "This is every active disaster on Earth right now." It isn't. GDACS prioritises events with international humanitarian implications — events tracked by UN OCHA or EU JRC at a coordination-relevant scale. Smaller-country domestic disasters that don't meet that threshold don't appear. A magnitude-4.8 earthquake in a sparsely-populated area, a small-scale wildfire that a national fire service handles internally, a localised flood with no international coordination — none of those show up. The dataset is the international humanitarian system's portfolio, not a global census.

2. "Cascades cause higher fatalities than singletons." The dataset does not carry consolidated death tolls. GDACS includes severity tiers and population-exposure proxies but not unified fatality figures. The 2,200+ death toll cited for the 31 Aug 2025 Afghanistan earthquake comes from UN OCHA situation reports cross-referenced separately. Cause-of-fatality and cascade-vs-singleton comparisons cannot be made from this data alone.

3. "The Atlantic is safer than the Pacific." It is not. The Atlantic-vs-Pacific cascade gap reflects the structural difference between the two basins' tropical-cyclone behaviour — the Western Pacific produces more concurrent storms in a tighter geographic window because of warmer SSTs across a larger basin, not because the Atlantic is less dangerous. Hurricane MELISSA is the single most powerful event in this dataset. The basins differ in clustering pattern, not in per-event danger.

4. "Active means current." Mostly. But the South American drought (gdacsEventId 1016449) has been flagged active since Sep 2023 — over 2.5 years. Drought events have long active windows by definition; flagged-active in the GDACS feed includes long-tail slow-onset events. Two African and South American droughts skew the portfolio toward slow-onset events that journalism rarely revisits. Frame the drought rows as ongoing background context, not breaking news.

5. "The 500km clustering radius is operationally meaningful for all hazard combinations." It is not. Two events within 500km share emergency-coordination geography, but a wildfire and an earthquake don't share response playbooks. The clustering is geographically valid; the operational meaningfulness depends on the hazard combination. Reading the Aegean cluster as "one operational situation" is meaningful for regional resource allocation; reading it as "one response operation" overstates the integration.

6. "Region-pack labels reflect the data." Three of the top 10 cascades have generic region labels in the raw output ("Other Wildfire Cascade Event", "Other Earthquake Cascade Event", "Southeast Asia Earthquake Cascade Event") because the actor's region-pack doesn't cover Mediterranean / Hindu Kush / Central Philippines as named regions. The data is correctly clustered; the label is a coarse fallback. Coverage should use the descriptive geographic labels in the leaderboard, not the raw region tags.

7. "Cyclone country-reach equals cyclone impact." It does not. FENGSHEN reached 6 countries but at peak 93 km/h was a weak tropical depression by the time most landfalls happened. MELISSA also reached 6 countries but peaked at 296 km/h. Country-reach is a footprint metric; it does not measure intensity, damage, or fatalities.

8. "The actor's operationalSituations output is the only valid cascade classification." It is one classification, with one clustering radius and one set of correlation rules. Another emergency-management tool with different thresholds (200km, 1000km, hazard-pair-specific) would surface different clusters. The actor's choices are defensible and consistent across runs; they are not the only valid choice.

9. "Severity tier predicts press coverage." It does not. Some Red-tier events in this dataset have received minimal press attention; some Orange-tier events have headlined for a week. Press attention is driven by named recognition, evocative geography, and recency cycles; severity tier alone does not predict it.

Common misconceptions

• "Compound cascade means the events caused each other." No. The clustering is geographic, not causal. The Aegean's eight events share a 500km region; they don't trigger each other. Some hazard-pair combinations are causally linked (an earthquake triggering a tsunami; heavy rain triggering a flood) — the cascade classification doesn't try to claim causality, and citing it that way overstates what the data says.
• "GDACS is a complete global disaster feed." It is the international humanitarian-coordination feed. National disaster management agencies in any specific country track many more events than GDACS surfaces. GDACS shows what UN OCHA and EU JRC are flagging for international coordination, which is a much smaller filter than "all active disasters."
• "The 500km clustering radius is a hard physical boundary." It is a clustering parameter. Two events 501km apart would not cluster; two events 499km apart would. Real-world disaster geography is fuzzier than any single radius. The radius is a choice, not a physical fact, and other radii would produce different cascade counts.
• "Active for years means high current relevance." Drought events stay active in GDACS for long windows by definition — slow-onset hazards take years to resolve. A 2.5-year-old drought flagged active is not "breaking news today"; it's ongoing context. The active flag is a binary, not a recency indicator.
• "Pacific cascade dominance means the Pacific is in worse shape." The cascade-pattern difference reflects basin meteorology, not relative danger. The Atlantic produces fewer simultaneous storms in a tighter window; the Pacific produces more. Both basins are dangerous in different ways.

Glossary

• Cascade (this audit's definition) — a geographic cluster of two or more active GDACS events within ~500km of a shared centroid, surfaced as a named entry in the actor's operationalSituations output.
• Operational situation — the actor's umbrella term for a multi-incident geographic cluster being treated as one emergency-coordination unit rather than as independent events.
• Red / Orange / Green tier — GDACS severity classification, assigned per event based on hazard magnitude and population exposure. Red is the highest tier; Green denotes a monitored event below the operational-attention threshold.
• portfolioPressureIndex — the actor's composite portfolio-stress metric, scored 0-100 against simultaneous-event count, region saturation, and severity-tier distribution. 100/100 is the maximum.
• Single-link clustering — a clustering algorithm where two points are grouped if they're within a fixed distance of each other; clusters grow by chain-of-proximity. The actor uses 500km single-link clustering across hazard types.
• Systemic-risk signal — a boolean flag the actor fires when portfolio-level patterns cross thresholds (e.g., multi-region-critical-activity fires when more than three regions are simultaneously in the overloaded saturation tier).

Press lift-out for journalists

A copy-paste-ready version of the headline finding for newsroom use:

"A 2026 ApifyForge audit of the live GDACS Global Disaster Alert and Coordination System feed on 11 May 2026 found 97 active disasters globally (19 Red-tier, 78 Orange-tier), 48 of which (49.5%) are members of one of 10 compound cascades — geographic clusters of two or more linked incidents within ~500km of a shared centroid. The largest cascade contains 11 linked incidents along the Southeast Asia Tropical Cyclone Cascade in the South China Sea / Vietnam coast basin, covering nine named Pacific cyclones plus one flood and one cyclone-flood overlap reaching at least eight distinct countries. Seven of eleven GDACS regions are simultaneously classified as overloaded. Three systemic-risk signals are firing: multi-region-critical-activity, multi-region-cyclone-activity, and seismic-cluster-detected."

Source for citation: ApifyForge, "48 of 97 Active Disasters Are in Compound Cascades — May 2026 GDACS Audit," 11 May 2026. Underlying data: GDACS live feed (UN OCHA + EU JRC), captured via the gdacs-disaster-alerts actor on 11 May 2026 (run rRHpXtHjNKW1rQLXU, dataset TbefZLGr54UQPF12W). Cross-references: USGS earthquake catalog, JTWC, UN OCHA ReliefWeb.

Embeddable visuals

The three charts below are described as image-prompt blocks for a designer or chart tool to reproduce. Each is reproducible from the tables in this post.

Chart 1 — The 10 compound cascades, ranked by linked-incident count

Horizontal bar chart, sorted descending. The top bar (Southeast Asia Tropical Cyclone Cascade) extends to 11 linked incidents — roughly 1.4× the second bar (Aegean Wildfire + Türkiye Earthquake at 8) and almost 2× the third (Hindu Kush Earthquake at 6). Bars 4-6 (Central Philippines, Bangladesh / NE India, Iberian Wildfire) cluster at 5 incidents each. Bars 7-10 shrink to 2 incidents each forming a clear long tail. Y-axis: cascade name + geographic label. X-axis: linked-incident count. Colour-code bars by dominant hazard type (TC blue, EQ red, WF orange, FL teal, mixed grey). Source line: "GDACS live feed via gdacs-disaster-alerts Apify actor, captured 11 May 2026."

Chart 2 — The Southeast Asia cyclone cascade as one operational situation

A horizontal timeline showing the nine named Pacific cyclones in the Southeast Asia cascade across Jul-Nov 2025, with each cyclone plotted as a coloured bar spanning its start-to-end date. Stack them vertically by name (WIPHA at top in July, FUNG-WONG at bottom in November) so the visual collision in Sep-Oct is obvious — five storms overlap or near-overlap in that window. Annotate each bar with peak wind speed in km/h. Overlay a single horizontal bracket labelled "South China Sea / Vietnam coast cluster — 500km radius around centroid 20.0N 109.9E" spanning the full Jul-Nov range. Source line: "JTWC best-track data via GDACS, captured 11 May 2026."

Chart 3 — Pacific cascade region vs Atlantic singletons region

Two side-by-side maps. Left: Western Pacific basin map centred on the South China Sea / Vietnam coast, with the 11 linked incidents of the Southeast Asia cascade plotted as filled circles sized by peak wind speed; a translucent 500km circle around centroid 20.0N 109.9E shows the cluster boundary. Right: Atlantic basin map centred on the Caribbean, with Hurricane MELISSA-25's track plotted as a single bold line covering Jamaica, Cuba, Bahamas, Bermuda, Haiti, and Canada. No cluster boundary on the right — there's nothing to cluster with. Headline above the two maps: "Same hazard category, two basins, two different structures." Source line: "GDACS live feed via gdacs-disaster-alerts Apify actor, captured 11 May 2026."

Frequently asked questions

What does GDACS classify as a compound cascade?

A compound cascade in this audit is a geographic cluster of two or more active GDACS events whose epicentres or impact centroids fall within ~500km of a shared cluster centroid, detected by single-link clustering across all hazard types and surfaced as a named entry in the actor's operationalSituations output. Hazard categories can be mixed within a cluster — an earthquake and a wildfire in the same 500km radius will cluster together. The classification is geographic, not causal: a cascade is a regional cluster, not a chain of triggering events.

Why is the Southeast Asia cyclone cascade treated as one event when the press covered nine separate storms?

Because all nine named cyclones, plus one flood and one cyclone-flood overlap, cluster within ~500km of centroid 20.0N 109.9E across the Aug-Nov 2025 window. From an emergency-coordination standpoint, repeated landfalls in overlapping populations, compounding flood risk, and shared response infrastructure across consecutive storms is one operational situation rather than eleven independent events. The press names each storm for tracking; GDACS groups the regional sequence for coordination. Both framings are valid for different purposes.

Is Hurricane MELISSA in a cascade?

No. MELISSA is a standalone — the single highest-pressure event in the entire dataset by the actor's pressureScore metric (70/100) and tied for the highest country-reach at 6 countries (Jamaica, Cuba, Bahamas, Bermuda, Haiti, Canada), but no other Atlantic event clusters within 500km of its track centroid. The Atlantic basin produces fewer, more powerful, more spaced storms; the Western Pacific produces more concurrent, tighter-clustered storms. The cascade-density difference is basin meteorology, not a relative danger signal.

Does the actor have historical GDACS data?

No. The actor queries the live GDACS feed. Setting dateFrom / dateTo filters the currently-flagged-active portfolio; it does not query GDACS history. Time-series analysis requires repeated captures stored over time, not a single dated query. The companion raw.json for this audit is the captured 11 May 2026 snapshot, archived for reproducibility.

Why are some cascades labelled "Other Wildfire Cascade Event" or "Other Earthquake Cascade Event"?

Because the actor's region-pack — its lookup table of named geographic regions for centroid-to-region resolution — does not currently cover Mediterranean, Hindu Kush, or Central Philippines as named regions. When the correlation engine detects a cluster whose centroid doesn't match any named region, it falls back to a generic "Other [dominant-hazard] Cascade Event" label. The clustering is correct; the label is coarse. The leaderboard in this post uses centroid-derived descriptive labels; the CSV preserves both.

Can I reproduce this audit myself?

Yes. The methodology section documents the full pipeline. Sign up for an Apify account, open the gdacs-disaster-alerts actor page, configure the input as specified (hazardType: "all", alertLevel: "all", minSeverity: "low", dateFrom: null, dateTo: null), and run it. The full live-feed capture completes in under five minutes against the GDACS json-api endpoint. Note that because the live feed is continuously updated, an exact reproduction of the 11 May 2026 figures requires the captured raw.json companion file; re-running today produces a current-as-of-today version of the same analysis structure.

How does GDACS compare with USGS or JTWC?

Different scope and different focus. GDACS is the portfolio layer — it covers earthquakes, cyclones, floods, wildfires, volcanoes, and droughts across all hazard types in one feed, with international humanitarian coordination as the editorial threshold. USGS is the high-fidelity single-hazard feed for seismic events (epicentre, magnitude, depth, fault mechanism). JTWC is the equivalent for tropical cyclones (best track, wind history, forecast cone). The audit uses GDACS as the portfolio surface and cross-references to USGS and JTWC for per-event detail. They are complementary feeds, not competitive ones.

Ryan Clinton publishes Apify actors and MCP servers as ryanclinton and builds developer tools at ApifyForge. The cascade leaderboard above was produced via the gdacs-disaster-alerts actor against the live GDACS json-api endpoint (run rRHpXtHjNKW1rQLXU on 11 May 2026); the methodology, analysis, and framing are independent of any product positioning. Related ApifyForge data audits in the same methodology pattern: tech podcast cemetery 2026 · OSS maintainer burnout index 2026 · Stack Overflow question decline 2020-2026.

Last updated: May 2026