Sources and coverage

ISPRA's dataset description states: "Alla data dell'elaborazione (2020) non era disponibile lo scenario a pericolosità elevata per il territorio della ex Autorità di Bacino Regionale delle Marche" — at the time of processing, in 2020, the high-hazard scenario was not available for that basin authority's territory. There the "pericolosità elevata" class is not mapped, and its absence does not mean the scenario does not occur.

ISPRA's terms of use state that it does not warrant the completeness or the currency of the information. The inventory grows when a Region sends new surveys: a recent landslide may not be in it yet.

The density of recorded landslides varies greatly between regions because of how thoroughly each was surveyed, not only because of how prone to landslides it is: Puglia has 839 landslides in the inventory. The absence of a recorded landslide means different things from one region to another, and in none of them does it mean the slope is stable.

The open data give each landslide its type of movement and its location, not whether it is active nor when it happened: a recorded landslide may be active, dormant or stabilised, and the report cannot tell. For 54,543 landslides the type too is «n.d.».

58,932 landslides are recorded as a line and 43,502 as a point only (at the crown): they have no perimeter. They are drawn as the 25 m of the declared positional accuracy, which marks where the landslide is recorded, not how much ground it involved.

29,726 of 99,229 stretches are shorter than 5 m, 10,223 under 10 cm: the fragments the overlay of the 2006 and 2020 lines leaves between one stretch and the next, about 40 km in all. Below the 5 m accuracy ISPRA declares their class cannot be resolved, and TerraOverview does not read them: read within 250 m, they brought «erosion» into about one coastal report in twenty where no stretch of a metre or more was eroding.

31,157 of 99,229 stretches, about 10,600 km, carry the dynamics «Nessuna informazione»: defence works, harbours, river banks, high coasts and stretches without a reliable comparison with 2006. ISPRA draws their line but does not assess how it moved.

The class compares the 2006 line (IT2006 orthophotos, 2004–2006) with the 2020 one, drawn on imagery from 2007 to 2020: there is no shift in metres and no yearly rate, and the time between the two dates changes from one stretch to the next. A shoreline also moves with seasons and storms.

The source does not say why a stretch moved landward or seaward — storms, works, nourishment, river supply — nor whether it will go on doing so. It is not a hazard or a risk map.

ISPRA's dataset record states that the database covers «le sole aree di pianura» — the plains only. Hills and mountains are not surveyed: there this source says nothing.

ISPRA's project summary describes northern Italy as territory not yet visited and where the census is still under way. Piedmont and Valle d'Aosta have no records at all.

Friuli Venezia Giulia alone holds almost half the records, the result of an agreement with the Region; Marche and Trentino-Alto Adige have two each. The absence of a record means very different things from one region to another, and in none of them does it mean that no sinkhole exists.

ISPRA notes that not every case investigated turned out, on site, to be a sinkhole in the strict sense: some records are anthropogenic, volcanic or karstic collapses, and for 35% of records the type is empty or «Undefined». The type is quoted as the source writes it, question mark included.

MPS04 has no grid node in Sardinia: the nearest one to Cagliari is 248 km away. The ZS9 seismogenic zonation the model is built on defines no sources there. For the whole region this source reports "no data", which is not a low value.

The nearest node lies between 2.6 km (Ustica) and 5.5 km (Lampedusa) away, beyond the 1.6 km within which a cell speaks for its own node, and at Lampedusa and Linosa that node carries a value of zero anyway. These are holes in the grid, not in our reconstruction. INGV published separate assessments for the islands left outside MPS04 in deliverable D1 of project S1 (2007), which is not part of this grid. Lipari, Vulcano, Salina, Ponza, Ventotene, Ischia, Capri and the Egadi islands do have nodes and are covered.

At sea south and west of Sicily the model computes no estimate and the file records zero. Those cells are kept and read as the absence of an estimate, never as a low acceleration.

The value holds for rigid ground on level topography. Amplification by the soil and the landform of the site is not included and can be substantial; it is the subject of seismic microzonation studies, which this source does not contain.

Of the 114,436 Italian observations, 51,711 are below intensity 4 in the source's own numbering — among them «F» (felt) and «NF» (not felt) — and are not listed. 2,146 localities have no observation at 4 or above and are not in the archive: there the reading is «no data», not «no earthquake».

A locality's history depends on the studies that cover it: the median is 2 observations, the ninetieth percentile 20, the maximum 315. A short history says little was documented, not that the place was little shaken.

A macroseismic intensity is defined over a set of buildings, not over an address: the EMS-98 scale says it should not be assigned to a single building nor to a whole administrative area. Within one village the effects vary with the ground and with how vulnerable the houses are.

The source marks some observations as referring to a territory, an isolated building, a small settlement or a deserted locality. The source's code travels with the record and is not reinterpreted.

Abruzzo, Basilicata, Calabria, Campania, Emilia-Romagna, Lazio, Marche, Molise, Sicily and Umbria had not completed the first identification of priority areas by the date of the list. In their territories this source states nothing: not that there are no priority areas, but that the authority has not yet identified them.

A "priority area" is an estimate about a municipality's building stock — more than 15% of its buildings above 300 Bq/m³ as an annual average — not a measurement of a building. The Istituto Superiore di Sanità is explicit: a radon map does not allow the radon of a particular dwelling to be estimated reliably, because concentrations can vary greatly even between neighbouring buildings. For a building, a direct measurement is needed, preferably over a year.

Where the authority has acted and the municipality is not on the list, the reading is that the municipality does not reach the 15% threshold. It is not a statement about any one building, and a municipal boundary is administrative: radon does not stop at it.

The consolidated list of municipalities is published by ARPA Lombardia with the Gazzetta Ufficiale reference of each act; the act remains the authority. Reconciling it with the Istat codes required two corrections, both typing errors in the consolidated list, each resolving to a single municipality and confirmed against the list published by the Province of Bolzano: "BRENERO" for Brennero and "CALSARO SULLA STRADA DEL VINO" for Caldaro sulla strada del vino. The Piedmont list was compared entry by entry with table 3.1 of d.g.r. 61-6054 of 25/11/2022, from which it derives: the 37 municipalities agree.

ISPRA's dataset description states: "L'ultimo aggiornamento è al 31-12-2021, è relativo esclusivamente ai procedimenti di competenza delle Regioni o di Enti da esse delegate (sono pertanto esclusi i SIN)" — the register covers only procedures run by the Regions or bodies they delegate, and Sites of National Interest are therefore excluded. All 16,067 records carry cod_sin = NO_SIN. The SIN perimeters are published by the MASE under CC BY 4.0 and are a separate source, not yet built: this one answers nothing about them, in either direction.

ISPRA's dataset description states that the register "non ha una copertura omogenea a livello nazionale a causa della mancata trasmissione da parte della Regione Sardegna e della trasmissione parziale da parte della Regione Umbria" — its national coverage is uneven because Sardinia transmitted nothing and Umbria transmitted only part. In the data Sardinia is absent altogether and Umbria holds 60 procedures against Lombardy's 5,370.

2,598 of the 16,067 procedures carry the attribute coord_conv = SI and sit at only 667 distinct positions, one coordinate accounting for 192 of them: these are administrative locators rather than site positions, and a 250 m disc drawn around one of them would be an invention. TerraOverview does not place them. Veneto is entirely in this condition — 1,886 procedures recorded and none of them placeable — so for Veneto this source never answers.

ISPRA reports: "Una ricognizione effettuata nel 2015 dal MASE ha evidenziato che lo stato di attuazione ed aggiornamento delle anagrafi è estremamente disomogeneo sul territorio nazionale" — a 2015 survey by the ministry found the registers to be kept and updated very unevenly across the country. The data shows it: Lombardy holds 5,370 procedures, Sicily 122. The absence of a record means very different things in the two regions, and in neither of them does it mean that no procedure exists.

The data stops at 31/12/2021. A procedure open on that date may since have closed, and a site contaminated then may since have been remediated by an act of the authority: for the present state, the competent authority named in each record governs.

Only the activities listed in the annexes of the directive and of the regulation report, and only above the thresholds they set. A workshop, a small plant or a depot below those thresholds is not in the register: its absence says nothing about it.

For each site the latest report is kept: for 7,701 sites that is 2024, for 2,698 it is an earlier year, back to 2007. A site that no longer reports may have closed, fallen below the thresholds or changed operator: the year is shown beside the site and is never hidden.

The source publishes one point per site, at a declared resolution of 10 m, not the outline of the plant. A large works is a single point: its boundary may be much further away, and the 500 m circle is TerraOverview's choice, not a measurement of the source.

The quantities are those the operator reports for the whole site over a year. They are not a measurement of the air, the water or the soil at the address, and TerraOverview does not use them to estimate one.

At 1,011 positions of 9,186 the source records more than one declaration: the `IT.EEA/…` identifiers of the older E-PRTR reporting beside the `IT.CAED/…` ones of the recent reporting, and different companies reporting from one address. Often it is the same plant named twice — the old record names the facility, the new one the company — but not always, and TerraOverview does not claim they are the same. There is one circle per position: the fields and the class are the most recent declaration's, and the others are listed under it with their year and identifier. At 160 positions the declarations carry different classes, and the most recent one gives the class.

The site map carries pollutant names for 885 of the 10,399 Italian sites, all of them among those that reported releases. For the others the field is empty in the source, and the report does not fill it: the quantities and the detail are in the EEA's tabular dataset, which the report links to.

An exceedance is established for the zone, which can be tens of kilometres across, and the values given are the worst measured by the stations inside it. They are not the air quality at an address: the concentration changes between a street front and a courtyard, and no zoning — nor a model grid at ten kilometres — can bridge that distance.

The archive carries 2025, the last year for which all the Italian assessments are final: on 2026-10-10 all 1,524 Italian rows for 2025 carry Preliminary = 0, as 2024's do, while all 802 rows for 2026 are preliminary. Building on 2026 would mean publishing numbers that change under the reader.

What is reported are the limit values and target values set to protect human health. Critical levels and target values for vegetation protect ecosystems and answer a different question; information and alert thresholds concern single hours rather than living somewhere.

The zone outlines are simplified to 100 metres. Within that distance of a zone boundary a point may fall in the neighbouring zone, whose assessment may differ.

«Non sono esposte le mappe sottoposte a vincoli di riservatezza o quelle sulle quali sono in corso interventi di manutenzione» (Agenzia delle Entrate). Where a parcel is missing, the report does not say the area is unregistered.