Forensic Review · Wildfire Causation · Sicily 2003–2026
The official answer is “predominantly deliberate — the hand of criminals.” This review tests that answer against 133,102 independent satellite fire detections, ten comparator fire regimes, and a house-by-house reconstruction of one Sicani comune — to ask whether “human-caused” is a measurement or a convenient administrative default.
Sicily's own institutions, the national government, the press and the public converge on a single claim: Sicilian wildfires are almost entirely human-ignited and predominantly dolosi — deliberate — the work of arsonists and organised criminal interests. This review finds that the claim is correct in the part that is nearly meaningless and substantially incorrect in the part that carries the policy.
The technical anchor is the Piano Regionale AIB of the Corpo Forestale della Regione Siciliana: “the clear preponderance of deliberate fires leaves no doubt,” and “almost the entirety of fires (about 95%) occurs by the hand of man.” The single published number is a decade-old chart giving a mean deliberate share of 77.5%. The political layer runs higher and without method — 98% (Assessore Savarino, May 2025); the July 2025 attribution of a 380-fire day to “the criminal hand of unscrupulous arsonists” (President Schifani).
Near-total human ignition is real — natural fires average 0.36% of the record, and the satellite data confirms it independently: the weekday signal in 133,102 detections is impossible under natural ignition (p ≈ 10⁻¹⁰⁸). But this is nearly uninformative — it does not separate an arsonist from a farmer whose stubble fire escaped.
The specific claim that most Sicilian fires are deliberate in the criminal sense is not a measurement. It is an administrative coding produced by the agency that fights the fire, without forensic investigation, whose two headline categories behave as a single pool (deliberate and “unclassifiable” correlate at r = −0.989), whose split shifts 15–17 points exactly at a change of recording software, and which the responsible regional minister repudiated on the record as “a proportion absolutely unconnected to reality.” The judicial system corroborates it at roughly one denunciation per 110 fires and near-zero convictions. And it describes a landscape that does not burn: 84% of what burns is non-forest.
The accurate description, on the measured evidence, is not a forest-arson wave. It is an agro-pastoral burning regime on grass and abandoned land, on a depopulating interior, weather-gated into catastrophe. The identity of the igniter is a second-order determinant of the disaster; the official framing indexes the entire problem on the one variable that barely moves it.
The question was worked by three parallel adversarial research councils and one independent quantitative arm. Each council fanned out across sources, then subjected its own load-bearing claims to hostile verification — agents instructed to refute each proposition and concede only what the evidence forced. The three councils covered, respectively: (I) the Sicilian doctrine and its evidentiary test across all nine provinces; (II) twelve comparator fire regimes and the methodology of cause attribution itself; and (III) the decadal history of Sicilian fire since 1950 and a micro-study of Alessandria della Rocca. The quantitative arm pulled the full NASA FIRMS active-fire archive for Sicily and analysed it against the agricultural calendar.
Because most of this evidence is administrative, every claim is graded. The single most important distinction in the entire review is between what has been measured and what has been officially coded — a national statistic saying “62% deliberate” is a coding decision made by an institution under incentives, which is not the same as a measurement of intent.
Limits held throughout. A satellite detection is not an ignition. MODIS/VIIRS have a minimum detectable fire size that systematically under-samples exactly the small agricultural burns most relevant to the Sicilian case — this bounds, and is stated as bounding, several conclusions below. Four 5-day windows in March–April 2010 (off-season) failed to download and are absent. No conclusion rests on a single source; where sources conflict, the conflict is surfaced rather than resolved.
The decisive question is not whether arson happens in Sicily — it does, and it is prosecuted — but whether the share the state publishes is a measurement. It is not, on five independent grounds, each verified against the primary documents.
Officially coded The only cause distribution the Regione Siciliana has ever published is Grafico 21 of the 2020 Piano AIB update — per-year shares of the number of fires, 2010–2020: deliberate 91.5% (2010) falling to 67.1% (2020), “unclassifiable” rising from 5.2% to 29.5%, natural 0.1–1.3%. The current 2023–2025 plan reproduces the identical decade-old data with the percentage labels stripped. Neither the November 2024 nor the December 2025 update touched the causes chapter. In 2026, Sicily's standing evidence base on why its land burns is a fifteen-year-old graph.
Measured Across the eleven years, the deliberate and “unclassifiable” series correlate at r = −0.9890, and their sum is a near-constant 96.08% ± 1.41. The deliberate line moves 24 points; the two lines together move 5. The split jumps 15–17 points precisely at the 2014–2015 boundary where the AteSO system was built and the ASTUTO system entered service. The famous “arson is declining” story is, on the arithmetic, a reclassification at a software changeover, not a change in anyone's behaviour.
Documentary Assessore Toto Cordaro told the Sicilian Regional Assembly's Antimafia Commission that summer 2021's ratio of 494 deliberate to 21 negligent fires was “a proportion absolutely unconnected to reality,” because the officer writing the report, uncertain between deliberate and negligent, “prefers to ascribe it to the deliberate offence, because he believes this will secure a more penetrating investigation.” That testimony predates the plan that relies on the coding by seventeen months.
Officially coded In the plan's own definitions, burning stubble appears under both “negligent” (“the burning of stubble, the cleaning of cultivated fields”) and “deliberate” (“burning of agricultural residues such as stubble and scrub, to clean the ground before sowing”). The Agrigento provincial plan calls escaped stubble burning “the principal cause of forest fire” in the province while conceding in the same document that it is “difficult to classify as negligent or deliberate.” Reading “77% deliberate” as “77% criminal arson” is a definitional error the public narrative makes systematically.
Officially coded Sicily 2025: 654 fires, 376 offences, 6 denunciations, zero arrests — one denunciation per 110 fires. Across 2015–2026 the council found three first-instance arson convictions in Sicily, all resting on CCTV, intercepts or self-recorded video — never on fire forensics. Prosecutors reject the organised-crime frame outright: Palermo's chief prosecutor called the idea of Cosa Nostra directing the fires “a suggestion,” and framed the damage as “the downstream result of a chain of failures and omissions by those who should tend the territory.”
A skeptic on the council correctly noted that “unclassifiable” is a real, separate, large residual box, not a crude default. True — and it does not save the statistic. The residual exists and the line dividing it from “deliberate” is unstable, moves with the recording system, and is drawn by the agency that fights the fire without independent adjudication. That is an administrative artefact by any reasonable definition. Verdict: the official cause breakdown is not a measurement. Confidence: high.
Official statistics cannot resolve the question because the agency that produces them also fights the fires and assigns the cause. So this review measured Sicilian fire independently, from space, using the one method that owes nothing to any Italian institution: the joint distribution of when, where, and how hot fires burn, across 23 years of NASA FIRMS active-fire detections.
Measured 14.7% of raw FIRMS detections over Sicily are not vegetation fires — 6,625 volcanic (Etna), 7,605 other static land sources, 5,366 offshore. Anyone computing “Sicilian fire trends” from raw FIRMS without filtering the type flag inflates the record by about a sixth. Every figure below is Etna-excluded, vegetation-fire only: 94,465 detections.
Measured Sicilian fire activity is episodic, not trending. A Mann–Kendall test on the methodologically consistent MODIS series (2003–2025) returns z = −0.37, p = 0.71; the VIIRS series (2013–2025) z = +0.55, p = 0.58. Extreme years (2007, 2012, 2021, 2023) tower over quiet ones (2015, 2018, 2024) with no monotonic climb. This is the fingerprint of a weather-gated system, not a rising tide of ignition. (Caveat: detection counts are not burned area and saturate on very large fires.)
Measured Satellite orbits are indifferent to the day of the week, so any weekday structure in detections is a property of human behaviour, not of observation. Lightning would be flat. The Sicilian summer signal is overwhelming — p = 1.7 × 10⁻¹⁰⁸ — with a Sunday peak (index 1.15) and a Wednesday trough (0.87). Human causation is beyond dispute.
But the shape is diagnostic. A Sunday peak is not the weekday-heavy pattern that salaried agricultural or forestry work produces; it is the signature of non-professional, discretionary land use. And winter inverts the whole pattern — Tuesday 1.37, Saturday 0.77 — consistent with a second, professional ignition population (pruning-residue burning) operating on a working-week rhythm. Two distinct human fire economies, neither of which resembles a coordinated criminal campaign.
Measured The PHOENIX area of interest does not follow the regional pattern (Kolmogorov–Smirnov D = 0.162, p = 1.0 × 10⁻⁹⁷ against the rest of Sicily). Its season starts about three weeks later — 10th-percentile date 5 July vs 16 June island-wide — with the sharpest rise in the 10–20 July block, and it has almost no winter fire (0.8% vs 4.1%). That late onset tracks the interior durum-wheat harvest at ~700 m, which is cut weeks after the coastal plain and is immediately followed by stubble burning. ADR fires are also small: median 7.75 MW, 59% under 10 MW, against a Sicily maximum of 2,450 MW — the radiative-power signature of field fires, not forest crown runs.
Measured As this review was compiled, the FIRMS record showed a record-breaking fire episode active at Alessandria della Rocca. 19 July 2026 produced 93 detections within 15 km — the single highest day in the district's entire 23-year satellite history; 21 July produced 65 — the second highest; the fire was still burning through the night of 22 July (detections at 01:07 and 02:34 local, peak 542 MW). Four of the eight worst days on record fell in one five-day window. It is burning at night in fuel — a genuine wildfire, not a smouldering stubble escape — near Santo Stefano Quisquina, and it is independently corroborated by local reporting of the 20–21 July Bivona fire 6 km away.
One discriminator was tested and discarded honestly: the hour-of-day peaks (14:00, 02:00) are satellite-overpass artefacts (Aqua ~13:30, Terra ~10:30, Suomi-NPP ~13:25/01:25), not fire behaviour. Diurnal timing is unusable with these products and is not used here.
The strongest test of whether Sicily's number means anything is to place it beside other fire regimes and their cause-attribution systems. Do this, and the “deliberate” share turns out to track two things — how large the “unknown” residual is and where it is forced, and whether the classification scheme even has an agro-pastoral category — far more tightly than it tracks any measurable difference in human behaviour.
| Region | “Deliberate” share | Unknown residual | The tell |
|---|---|---|---|
| Sicily coded | ~77% (77.5% mean 2010–20); politically 95–98% | 5→30% (rising) | Deliberate + unknown fixed at 96%; minister disowned it; 3 forensic-free convictions |
| Sardinia closest analogue | 68.4% of investigated fires | — | 79.3% of “deliberate” fires have no determined motive. Pastoral burning only 4.2%; pyromania 2.0% |
| Galicia | 81% intentional | — | Only 18–31% forensically corroborated; ~2% with an identified culprit; 60% of “intentional” fires have unknown motive |
| S. mainland Italy (same country, different corps) | 53.9% deliberate (2024) | 26–30% | Ordinary-statute institutional counterfactual to Sicily — lower share, larger honest residual |
| Puglia (2001) | 54.1% deliberate | 2.0% | A near-zero unknown residual is not credible for any fire service |
| Basilicata (’21–’23) | “the remaining 49% is certainly of deliberate origin” | ~35% “doubtful” | Residual-as-certainty, stated verbatim in the plan |
| Aragón | 18% of fires · 4% of area | — | Lightning ignites more fires (21%) than arson; deliberate ignition near-irrelevant to area |
| Greece (’84–’09) | 12% arson | 57% | When the residual is kept honest, the deliberate share is modest |
| Corsica / Prométhée island | arson 32% (S) vs 66% (N) | ~70% “cause inconnue” | Independent ODARC pastoral database finds 15–36% pastoral ignition — a category Prométhée does not have |
| Portugal | arson 38% of determined / 24% of all | coded by why it failed | Lightning = 32% of 2025 burned area from 1–2% of ignitions |
Sardinia is the closest thing Sicily has to a mirror. Its official system codes most fires “deliberate” — and then records that four in five of them have no motive at all.
Read down that residual column and the mechanism is plain. Where the unknown fraction is kept honest — Greece 57%, Corsica 70% — the deliberate share is modest (12–32%). Where the residual is administratively suppressed — Puglia's 2%, Basilicata's “49% certainly deliberate” — the deliberate share balloons. Sicily sits at the balloon end. And Corsica is the decisive case: a Mediterranean island with pastoralism, whose national database (Prométhée) has no pastoral category and therefore records 70% “unknown,” while an independent agricultural database covering the same fires finds pastoral ignition running at 15–36%. The category you do not have is the cause you cannot see.
“Official Mediterranean wildfire cause statistics are administrative codings rather than forensic determinations, and their deliberate-ignition shares are therefore not reliable measurements of intent.” — survived adversarial refutation. The honest counter (forensic determination has its own reliability problems; a high true-deliberate share is logically compatible with near-zero convictions) is accepted and does not rescue the Sicilian number, because the Sicilian number is not even a forensic determination to begin with.
One methodological guard-rail, adopted from the same council: the conviction gap is corroborative, not decisive. An arson classification is about fire events and physical evidence; a conviction is about a person proven guilty beyond reasonable doubt. Arson is a structurally low-clearance offence everywhere, so a genuinely high deliberate share can coexist with near-zero convictions. The load-bearing evidence against Sicily's figure is therefore the construction evidence of §02 — the r = −0.989, the fixed 96% sum, the recording-boundary jump, the minister's repudiation — not the arrest count.
If the Sicily-wide framing is a valid generalisation, it should describe the interior comune where PHOENIX operates. It does not — and the mismatch is measurable at the parcel level.
Measured Clipped to the comune boundary against the ISPRA/Regione Siciliana Carta Natura (100% coverage), Alessandria della Rocca is 78.6% agricultural land, 14.7% dry grassland/steppe, and only 2.6% woodland-plus-plantation of any kind. Native oak woodland is 47 ha (0.76%); conifer plantation 32 ha (0.52%). There is no Natura 2000 site in the comune; the Monte Cammarata reserve lies in neighbouring communes; the comune was never in the (now-abolished) Sicani park. The mountains-oak-woods-reforestation-reserves story that the regional plan tells about “the Sicani” simply is not the land here. And the province agrees: Agrigento is 69% agricultural, 19% pasture/uncultivated, 8% forest — the least wooded province in the regional inventory. ISPRA's satellite land-cover breakdown of Sicily's 2024 burned area independently confirms the pattern: 36% agricultural, forest a minority class.
Officially coded Alessandria della Rocca peaked at 6,408 inhabitants in 1951 and stood at 2,301 on 1 January 2026 — a loss of 64.1%. The seven-comune Sicani cluster fell 57% over the same span. Crucially for causation, the modern phase is no longer emigration: over 2002–2024, natural decrease and internal out-migration split the loss roughly evenly, and cumulative births (441) ran less than half of deaths (1,103). This is not the transatlantic exodus of 1900–1960; it is an interior emptying out, leaving grass and abandoned field to accumulate as continuous, unmanaged fuel. Land abandonment is the leading structural alternative to the arson framing, and here it is not asserted — it is counted.
Documentary Exhaustive searching of provincial, hyperlocal and national outlets, the comune's own portal, and the full Google News index recovered exactly two vegetation fires inside the comune across 2009–July 2026: a roadside fire (22 Sep 2014, suspected discarded cigarette or arson) and a woodland fire (14–15 Apr 2024, “almost certainly deliberate”). Neither has a published burned area. Everything else in the comune's fire file is structural, vehicular, or waste. There is no published fire cadastre for the comune, as Law 353/2000 requires. The dedicated hyperlocal outlet for exactly this cluster returns zero woodland-fire articles for the comune across its whole archive.
This absence is itself the central finding for the geographic question the review was asked: fire cause is not published anywhere below the regional level in the Sicilian system. The provincial plans manufacture a false appearance of local knowledge — Agrigento, Enna, Caltanissetta and Palermo each carry a word-for-word identical sentence attributing ignition “in via prioritaria alla natura dolosa,” then present tables with no cause column at all. There is no official evidentiary basis for any statement about why fire happens in Alessandria della Rocca, or in any specific Sicilian comune.
Officially coded Sicilian burning law fixes the high-risk period (and its absolute ban on agricultural burning) from mid-June — 15 May in 2025. But the interior grain at ADR's altitude is cut last, mid-to-late July, deep inside the ban. Stubble burning here is therefore forced into either illegal timing or a narrow pre-ban window — and the FIRMS record shows exactly the resulting compression: a sharp 10–20 July onset, and the record-breaking July 2026 episode landing precisely on the late interior harvest. The mechanism the regional plan itself names (“the burning of stubble adjacent to woodland… the principal cause”) is the one the satellite timing independently fingerprints. What the official system codes as the province's “deliberate” fires and what the calendar-plus-satellite evidence identifies as escaped harvest burning are, to a first approximation, the same fires under two names.
Sicily's own two independent data series settle the ignition-versus-spread question almost by themselves. In 2021 the region had 978 fires and 59,872 ha burned (61 ha/fire); in 2022, more fires — 1,083 — but 24,561 ha (23 ha/fire), less than half the area. The ground series repeats it: 2015 had 831 fires and 6,777 ha; 2016 had 1,014 fires and 27,928 ha — count up 22%, area up 312%. Burned-area variance is governed by the conditional probability that a start becomes large, not by the supply of starts. Confidence: high — this is arithmetic on the Region's own numbers.
What converts a start into 500 km², ranked by defensible contribution:
The framing indexes on the variable that barely varies. Yet Sicily's legal instruments say the opposite of its press releases: the 2021 crisis decree cites “exceptional meteo-climatic conditions,” and the Council of Ministers characterised July 2023 as “severe fires and an exceptional heatwave” with no cause stated. Civil-protection chief Cocina attributed the 2022→2024 decline to weaker winds and drought-reduced fuel — not to prevention or policing.
Interventions graded by strength of measured effect, not by prominence. The pattern is consistent: the instruments with the strongest evidence are the least funded, and the most funded are the least evidenced.
| Intervention | Evidence | Applicability to Sicily / finding |
|---|---|---|
| Early detection → rapid initial attack | strong (elsewhere) | Time-to-detection is the highest-leverage lever in the international literature; the Sicilian gap is real (satellites miss the small fires; ~200 lookout towers already supply 71% of reports). Directly relevant to a ground-sensor programme. |
| Fuel & abandoned-land management, extensive grazing | moderate | The mechanism that actually fits the Sicilian regime (grass/agro-pastoral, spread-limited). Supported but under-deployed. |
| Lookout towers (retain/modernise) | demonstrated | 71% of 2017–21 fire reports came from tower personnel vs 17% from the citizen line. Cheap, working, quietly effective. |
| Prescribed burning | refuted at achievable scale | Honest negative: the proposition that prescribed burning at the scale Sicily could realistically achieve would measurably cut burned area was refuted. Sicily runs no programme; the treatment area required to move outcomes is not attainable here. |
| Aerial suppression | weak vs prominence | Enormously expensive and politically central; independent effectiveness evidence is thinner than its budget share. Availability collapses on peak simultaneous-demand days regardless. |
| Fuel breaks (viali parafuoco) | weak | Fail under high fire intensity; maintenance-dependent; strategic placement > blanket application. |
| Surveillance-for-attribution (drones, mapping) | failed on its own terms | ~€290k of drones identified 3 perpetrators in 3 years; a €36.4m EU surveillance grant was lost to an unqualified bidder (Corte dei Conti). High confidence it has failed. |
| Penalty escalation / the mafia frame | unsupported | 293 arrests on 117,509 national offences over 24 years; the organised-crime thesis is rejected by the prosecutors who would have to prove it. |
| Fire cadastre (Law 353/2000 art. 10) | design flaw + non-compliance | Penalises changing land use while the plan ranks pasture renewal first among deliberate motives — so an arsonist who wants the land to stay pasture is undeterred. >1/3 of comuni non-compliant; ~41% of 2010–22 burned area sits in non-compliant comuni. |
The seasonal-forestry-worker perverse-incentive hypothesis — the recurring claim that Sicily's seasonal fire-fighting employment creates an ignition incentive — was tested as a proposition and returned: widely asserted, never rigorously tested anywhere. It is neither confirmed nor refuted; it is unexamined, and should be labelled as such rather than deployed as fact by any side.
Pulling the arms together: how much weight can the phrase “human-caused” actually bear in the Sicilian case?
“Human-ignited” is knowledge. Near-total anthropogenic ignition is supported by the official record (natural < 0.4%), by comparative Mediterranean lightning climatology, and — decisively and independently — by the FIRMS weekday signature at p ≈ 10⁻¹⁰⁸. Confidence: high. The one honest wrinkle: the near-zero natural share is itself probably an underestimate, because holdover lightning and dry-lightning ignition are barely investigated in Italy — but not by enough to overturn the human-dominant conclusion. Lightning is a real, under-counted minority, not the hidden main cause.
“Deliberate, in the criminal sense” is convention. It is a residual-driven administrative code, not a forensic finding; it disagrees with independent statistical attribution wherever both exist; it bundles genuinely ambiguous agro-pastoral burning into a criminal box; and its cross-national variation tracks classification convention rather than behaviour. The multiple-ignition-point inference that investigators treat as proof of arson is itself unreliable under extreme fire weather, where long-range spotting produces the same signature. Confidence: high that the criminal-arson framing is not a measurement; medium on exactly what share of Sicilian fire is escaped agro-pastoral burning versus true malicious ignition, because no one has measured it.
“Human-caused” is true. “Deliberately, criminally human-caused” is where measurement stops and convenience begins.
Was the arson framing, then, a convenient answer? The comparative record documents a real and repeated pattern of states over-attributing wildfire to deliberate criminal ignition in the aftermath of catastrophe — from Australia's Black Summer misinformation to ministerial claims in Greece and Algeria that investigation did not later substantiate. Sicily fits the pattern: the attribution externalises responsibility away from land management, prevention budgets, and detection failure, and onto an anonymous criminal. It is not necessary to allege bad faith to reach this conclusion — the incentive structure and the classification convention produce it without anyone deciding to lie. But the effect is the same: a genuine landscape-and-climate emergency is administratively re-described as a policing problem, and the remedies follow the misdescription.
The official assessment is substantially incorrect as a measurement, more specific to a subregion (the Palermo–Trapani wildland–urban interface) than admitted, and stale — its data end in 2020 at 67%, twenty-four points below where it began. It is not wrong that people start Sicily's fires. It is wrong about who, wrong about where that story fits, and wrong to spend the policy on ignition when the disaster is made by weather, fuel and abandonment.
Every published cause figure is region-wide; the province plans carry identical boilerplate and no numbers. The uniformity is manufactured — by publication scale, by size-threshold censoring that removes exactly the small agricultural escapes, and by importing national percentages that statutorily exclude Sicily. The variation that is measured is enormous, and it is not a variation in criminal intent.
| Province | Dominant real driver | Official framing | Note |
|---|---|---|---|
| Palermo | Escaped agro-pastoral burning in abandoned, WUI-saturated land; waste-site fires; a real arson component in coastal reserves | partial fit | Worst province 2010–20; adjudicated arson exists (Capo Gallo, Casaboli) — but 77–86% of burn is non-wooded, and burned forest fell 3,174→104 ha in one year |
| Trapani | Wind-driven runs through maquis and non-wooded ground | partial fit | Largest mean fire size in Sicily; a 55 km² fire on 20 July 2025 — five days before the “380-fire” arson attribution |
| Messina | Abandoned private terraces (Peloritani WUI); pastoral land-creation (Nebrodi) | poor fit | 2017: >90% of 850 ha was private uncultivated land; inquiry against unknowns. Nebrodi mafia convictions are subsidy fraud, not arson |
| Catania | Many small peri-urban stubble/vegetation fires; Etna volcanic ignition | poor fit | Most true forest fires but lowest mean size; volcanic ignition recorded at ~0% and excluded from the risk model |
| Agrigento | Escaped cereal-stubble & pasture burning across a near-treeless, depopulating interior | poor fit | ~6–8% forest; lowest wooded burn in 2023 yet Italy's most-burned province in 2025 (96% non-forest). PHOENIX AOI. |
| Caltanissetta | Annually renewed grassland & stubble on the gypsum-sulphur interior | poor fit | 114 km² in 2025, 8% forest |
| Enna | Grassland burn cycle eating into fragmented tree cover | poor fit | ISPRA: ~42% of tree cover burned over 2021–25 with <10% re-burn overlap; no cause attributed |
| Siracusa | Insufficient data | n/a | No cause data, no case material |
| Ragusa | Insufficient data | n/a | 77% agricultural land use; absent from every cause and conviction record |
Where the framing does have adjudicated content is narrow: the Palermo–Trapani wildland–urban interface, where the documented convictions sit and the fires are high-value and reserve-adjacent (note the selection effect — none was proved by fire forensics). Where it does not is the whole central-southern interior. Agrigento, Caltanissetta and Enna together were ~67% of Sicily's 2025 burned area — and all three were outside the July 2023 state of emergency, which covered only Catania, Messina, Palermo and Trapani. The island's real burned-area anomaly is genuine (Sicily has been Italy's most-burned region four years running); its causal uniformity is an artefact.
Four gaps stand between this review and certainty, and none of them is unfillable:
What would actually settle the causal question is executable with largely existing open data. Take a random sample of Sicilian fires with known ignition points (FIRMS/VIIRS first-detection pixels, Sentinel-2 burn perimeters), and for each compute the independent ignition signature — day-of-week, coincidence with the durum-harvest and burning-ordinance calendar at the correct interior altitude, distance to fields versus woodland versus roads versus settlement, event-size distribution — then compare that data-driven attribution against the official coding for the same fires. Where independent inference and official coding disagree, the coding is what fails. The weekday and calendar arms of exactly this test are already built and already point one way. What remains is the spatial-covariate and holdover-lightning work, and a systematic re-investigation of a random fire sample to a defensible forensic standard — the one thing no Sicilian institution currently does.
Until then, the defensible statement is the one this review will stand behind: Sicily's fires are lit by people, overwhelmingly; they are not, on any measured evidence, the criminal-arson wave the state describes; and the disaster they become is made by heat, fuel and a hollowed-out countryside far more than by the intent of whoever strikes the match.
Primary and peer-reviewed sources underpinning the load-bearing claims. Comparative figures (Sardinia CFVA, Galicia/MITECO EGIF, Greece OECD/GFMC, Corsica Prométhée & ODARC, Portugal AGIF/ICNF, Aragón, southern-Italy CFS/Carabinieri series) are attributed in the text to their issuing bodies; the quantitative arm's figures are the author's own analysis of the NASA FIRMS archive.