The Silence Index: How We Score a Place

Travel media ranks places by how many people liked them, which guarantees that the top-ranked place is the one you will share with the most strangers. We wanted the opposite instrument. Ninety-nine scores later, here is what it actually produced — including the finding that for most of this archive's life it never once used the bottom third of its own scale.

This is a methodology page and it describes a desk instrument. Nobody from this publication has stood in most of the places it scores, and nobody has ever put a sound level meter on the ground at any of them, or a sky photometer, or a counter at a trailhead. Every score below was assembled from published sources, operator figures and government statistics, at a desk in one country, applied to ninety-nine places in others.

The Silence Index is the spine of the site. It is a single number between 0 and 100 and it answers one question: how much of this place is left? It is not a quality score. Derinkuyu scores 54 and is among the most astonishing things human beings have cut into rock. Naoshima scores 38 and we would go back tomorrow. A low number is not a warning; it describes who else will be standing next to you.

The distribution, published in full

We have never shown this before. Showing it is why this page was rewritten.

Ninety-nine entries carry a score; the five Methodology pages carry 0 as a null value and are excluded from everything below. The scored set runs from 22 to 94. Three Instruments, One Town holds the floor, and The 800-Year Ledger holds the ceiling. The median is 76, the mean 71.0, and half the archive sits between 58 and 86. By ten-point band: one in the 20s, seven in the 30s, eleven in the 40s, nine in the 50s, ten in the 60s, seventeen in the 70s, thirty-two in the 80s and twelve in the 90s.

As proportions: forty-four of 99 entries — about 44 per cent — score 80 or above. Sixty-one score 70 or above. Eight score below 40. Nothing has ever scored below 22 or above 94, which means that of the 101 values this scale declares itself able to produce, 53 have never been used.

That is not a finding about the world. It is a finding about us. The archive selects for remoteness before it scores remoteness: we choose a place because it is unresolved, difficult or quiet, then award it a number for being unresolved, difficult and quiet. Barbara Geddes set out the general form of this error in Political Analysis in 1990 — selecting cases on the outcome of interest and then explaining the outcome — and it applies to a travel index as cleanly as to comparative politics.

The consequence is worth one plain sentence. The Silence Index is a usable comparator within this archive and a poor absolute measure of anywhere. A 66 does not mean Atacama is two-thirds of the way to silence. It means Atacama sits below Lake Natron at 81 and above Salar de Uyuni at 64, in a set of places that are all, by construction, unusually empty.

The four inputs, and what they borrow from

1. Footfall pressure — 30%

Not raw visitor numbers: visitors measured against the ground that has to absorb them. Three thousand a year across an island the size of Socotra is a different pressure from three thousand a day through a stepwell. We look at annual visitation, area and fragility, and concentration — whether arrivals spread across a year or compress into six weeks.

The lending literature is openly divided about this. A. M. O'Reilly set out the classical carrying-capacity framing in Tourism Management in 1986. Stephen McCool and David Lime answered it in the Journal of Sustainable Tourism in 2001 under the title Tempting Fantasy or Useful Reality?, concluding that the conditions required to establish a numerical capacity are rarely achieved in the real world, that most published capacities reflect capital investment rather than any ecological limit, and that the better question is what conditions are acceptable. We follow the second: a pressure, not a ceiling, and no carrying capacity claimed.

The counts are softer than they look. Iceland's tourist board is one of the better statistical operations in this field, and its own note explains that the headline figure — just under 2.3 million overnight foreign visitors in 2025 — comes from manual counting of departing passengers before airport security, and should be viewed with caution. That is the good case; elsewhere the number is an operator estimate, and the operator has an interest in it. Against that, IATA reported 2024 air traffic 10.4% up on 2023 and 3.8% above 2019 at a record load factor of 83.5%.

2. Access friction — 30%

Permits, sponsors, weather windows, boats, guides, waiting lists, physical difficulty. Friction is the most effective preservative there is. Son Doong's cap of roughly a thousand people a year through one licensed operator has held the largest cave on earth in near-original condition despite global fame; Skellig Michael's fifteen boat permits at twelve passengers put a ceiling near 180 landings a day. Those numbers and ten other regimes are in our field guide to permits. What does not count is price: an expensive place with a short flight and no cap is not protected, it is sorted by wealth.

3. Acoustic floor — 20%

How close is the nearest engine? This input has the strongest external grounding of the four — and the widest gap between what exists and what we do.

The US National Park Service runs a Natural Sounds and Night Skies Division that collects baseline acoustic data and identifies source-specific impacts. Its national model predicts L50 in A-weighted decibels on a 270-metre grid across the contiguous United States and publishes two versions of that map, existing and modelled natural conditions; the difference is the impact. The quantities are small: the NPS records the Haleakalā crater at around 10 dBA and remote Grand Canyon trails at about 20 dBA, against 50 to 60 dBA for a typical suburb.

The definitive result from that programme is Buxton and colleagues in Science in 2017: modelling across the continent, anthropogenic noise doubled background sound levels in 63% of US protected areas, produced a tenfold or greater increase in 21%, and affected 14% of critical habitat for endangered species — concentrated around transport, development and extraction, and lower under stricter regulation.

Quiet Parks International certifies against this problem directly, listing five Wilderness Quiet Parks: the Zabalo River in Ecuador, Glacier National Park, the Boundary Waters Canoe Area Wilderness, NamibRand in Namibia and the American Prairie Reserve. One is in this archive. NamibRand — the only place we score that carries independent acoustic certification — scores 84, not 94. We cannot state the threshold behind that certification: the assessment-criteria page publishes no decibel figure, no noise-free interval and no listening-hour requirement, only a document we could not retrieve.

So a measurement discipline exists, with instruments and grids and thresholds, and we use none of it. Our acoustic floor is inferred from distance to roads, airports, shipping lanes and settlement: an educated guess dressed as a fifth of a score.

4. Night darkness — 20%

The fourth input changed on 2 September 2026, and we record the change here rather than pretend the scale was always this shape. From the archive's first entry until that date the fourth input was unresolved questions — whether a place still held a real open problem: who cut the first level of Derinkuyu, whether Socotra's dragon's blood trees can regenerate at all. It was the least measurable thing on this page. Footfall has counts, however soft; friction has permit numbers; the acoustic floor has a federal division behind it. Unresolved questions had one editor's reading of a literature, and carried the same weight as the acoustic floor, which they should not have. Fourteen of the newer dossiers had already begun reasoning about the sky rather than the bibliography — for Darvaza, which its own dossier argues produced light on an industrial scale for decades, the sky says more about how found the place is than the state of its literature does — and a scale whose fourth input depends on which dossier you open is not a scale.

Night darkness replaces it. It asks how much sky is left: whether the Milky Way is visible from the ground, how far away the nearest skyglow is, and whether the place makes light of its own — a camp, a generator, a flare, a town. It has a measurement tradition behind it, described in the next section, that unresolved questions never had: Bortle's nine classes, Falchi and colleagues' world atlas in physical units, DarkSky International's pass-or-fail certification. We still use none of the instruments. Our darkness input is inferred from the published atlas, from certification where it exists, and from the presence of settlement and industry — one step better than a reading of a bibliography, and several short of a photometer.

What this means for the archive. Fourteen dossiers were scored with night darkness as the fourth input and their scorecards say so. Eighty-five were scored before the change, with unresolved questions in the fourth position, and their scorecards still say that, because we will not relabel a score that was reasoned differently. They will be re-scored one at a time as each dossier comes up for its source re-check, and each re-scored dossier will carry a dated note. Until then the two groups are comparable on three inputs and not on the fourth — a smaller flaw than the one it replaces, and one we would rather print than hide.

What a single number can and cannot do

Two instruments already do to the night sky what we attempt for a whole place — and since darkness became a fifth of our score, they are also the yardstick that input borrows from.

The Bortle Dark-Sky Scale, published by John Bortle in Sky & Telescope in February 2001, compresses sky quality into nine classes: class 1 is a site where the naked-eye limiting magnitude reaches 7.6 to 8.0 with effort and the Milky Way casts shadows, class 9 an inner-city sky where it falls to 4.0 or less. It is observer-dependent and subjective at the boundaries, and after twenty-five years it is still the most used shorthand in amateur astronomy, because a nine-point scale everyone understands beats a precise number nobody quotes. That is the argument for our index existing at all.

Alongside it sits the measured tradition. Falchi and colleagues published the world atlas of artificial night sky brightness in Science Advances in 2016 in physical units, finding the Milky Way hidden from more than a third of humanity, 60% of Europeans and nearly 80% of North Americans. DarkSky International then certifies against lighting management rather than a score — 270-plus places in 22 countries covering more than 160,798 square kilometres, in six categories. A place meets the standard or does not. There is no 84 out of 100.

The two traditions disagree about the rate of change, and we leave that standing. Kyba and colleagues, in Science in 2023, used 51,351 citizen-scientist star counts to infer sky brightening of roughly 7 to 10% a year — explicitly faster than satellite radiance predicts, which the authors attribute to spectral shifts satellites see poorly.

Our archive holds the lesson in miniature. Aoraki Mackenzie is an International Dark Sky Reserve and scores 66; NamibRand is a dark sky reserve and a certified quiet park and scores 84. Darkness and silence are correlated, not identical, and folding them into one digit throws the difference away.

Where the index fails

There is a mature literature on how weighted composite indicators fail, and it convicts us on five counts. The standard reference is the OECD and Joint Research Centre Handbook on Constructing Composite Indicators of 2008; Greco, Ishizaka, Tasiou and Torrisi reviewed the field again in Social Indicators Research in 2018 and found that weighting and aggregation are precisely where the paramount criticism lands.

• The weights are arbitrary. 30/30/20/20 is a judgement, never tested against anything. Worse, declared weights are not the weights that operate: Paruolo, Saisana and Saltelli showed in the Journal of the Royal Statistical Society in 2013 that because component variables are correlated and unequally dispersed, a component's nominal weight rarely equals its main effect — across six indicators including the Human Development Index. We have never computed our own effective weights, so we do not know what this index is really weighting.
• It is fully compensatory. Because the four inputs are added, a place can buy back a motorway with a permit regime. The UN Development Programme faced exactly this and changed the HDI in 2010 from an arithmetic to a geometric mean to limit substitution between health, education and income; Martin Ravallion's 2012 critique in the Journal of Development Economics then showed that the fix introduced tradeoffs of its own. We have not taken the first step.
• There is no uncertainty estimate. Saisana, Saltelli and Tarantola argued in 2005 that uncertainty and sensitivity analysis should accompany a composite indicator throughout, so that a ranking's robustness can be stated. Every Silence Index here is a bare integer. A 78 and an 80 are almost certainly the same number.
• One scorer, no reliability test. Every score was assigned by the same publication. Cohen's kappa and the intraclass correlation exist for exactly this situation — Hallgren's 2012 tutorial is the standard account — and we have run neither, having never had a second rater.
• It rewards specialisation. Nicky Rogge documented the equivalent problem for the Environmental Performance Index in Ecological Indicators in 2012. Here, a place with extreme access friction and a black sky can score in the 80s while being busier than a quieter place that happens to be easy to reach.

Add the bias we have always admitted and never fixed: the index leans toward difficulty, and difficulty is not a virtue. The lean is structural. Watch us for it.

What we will do about the skew, and what we will not

Two remedies are available and we are committing to one.

The first is rescaling: stretch the observed 22-to-94 range across the full scale so the numbers use their declared span. We are not doing that. It would move every published score without adding a single observation, and disguise a sampling problem as a scaling problem.

The second is a control set, and we commit to it. We will score twelve deliberately crowded, famous, easy places against the identical rubric — the kind of destination this publication does not otherwise cover — and publish all twelve with their working, whatever they come out at. That anchors the bottom of the scale with cases instead of an assumption, and it is the only way to learn whether the floor here is 38 because the world stops there or because our reading list does. Target: before 31 March 2027. If it is not published by then, this paragraph is the receipt. Sooner and cheaper: from the next review cycle every score will carry its four component values.

What popular coverage gets wrong

The genre this page belongs to is destination ranking — the ten places to visit this year, the fifty to see before you die — and its central defect is not bad taste. It is circularity. A ranking published by a travel outlet changes the thing it ranks. The list moves readers; the readers change the place; the place is re-ranked in its altered state, usually by the same outlet, often as a warning about crowds it helped create.

The cleanest documented case is not a magazine list at all. In 2007 a privately run internet poll announced the New Seven Wonders of the World, and on 9 July that year UNESCO issued a formal statement that there is no link whatsoever between its World Heritage programme and the campaign, that it had declined repeated invitations to collaborate, that the poll reflected only the opinions of those with internet access, and that such a campaign cannot in any significant or sustainable manner contribute to the preservation of sites. UNESCO's list rests on evaluation, defined criteria and binding commitments; the poll rested on votes. Nineteen years later the result is still reproduced in travel coverage as though it were a designation.

The correction we would like readers to carry is this: a destination ranking measures attention, not place. Attention is the one variable in tourism that responds instantly to being measured — which is the trap this page is in. A high Silence Index is exactly the signal that sends the wrong reader toward the most fragile entry here. We have no clean answer, which is why every dossier carries a cost section rather than only a how-to.

How to read a score

85–100. Early. Real friction, low footfall, dark skies. Go carefully or do not go; either is defensible. Thirty-one entries.

65–84. Findable but not found. A season, a permit or a difficult road is still doing protective work. Thirty-eight entries — the true centre of this archive.

40–64. Established. Extraordinary places with an industry around them. The quiet is available, but only at the edges of the day and the season. Twenty-two entries.

Below 40. Fully found. Eight entries. We will still write about them when they deserve it, but we will tell you what the place was, and when.

Scores are reviewed annually, re-dated, and they move — mostly downward. That direction of travel is the whole reason this publication exists. The number is a comparator, it carries no error bar, it was assigned by one desk, and it now reaches into the bottom third of its own scale, which it did not for the archive's first hundred entries.

Sources & further reading — checked 13 August 2026

1. US National Park Service — Natural Sounds and Night Skies Division. That a dedicated federal division exists to collect baseline acoustic and night sky data, provide technical assistance and identify source-specific impacts of anthropogenic noise and light in and around parks — the institutional model our acoustic input imitates without replicating
2. US National Park Service — Understanding Sound. The reference sound levels used in this piece: about 10 dBA in the Haleakalā crater, about 20 dBA on remote Grand Canyon trails and 50–60 dBA in a typical suburb, plus the statement that a 10 dB rise doubles perceived loudness
3. US National Park Service — Mapping Sound / the geospatial sound model. That the NPS models L50 in A-weighted decibels on a 270-metre grid across the contiguous United States and publishes both existing and modelled natural conditions, so that noise impact is the difference between two mapped surfaces
4. US National Park Service — Measuring Night Sky Quality. That NPS night sky assessment uses calibrated mosaic and fisheye camera systems with V-band filters and separates natural light sources from anthropogenic light — an instrumented process with no analogue in our own scoring
5. Science 356(6337):531 (2017), Buxton, McKenna, Mennitt, Fristrup, Crooks, Angeloni and Wittemyer — Noise pollution is pervasive in U.S. protected areas (bibliographic record). The headline findings: anthropogenic noise doubled background sound levels in 63% of US protected areas, produced a tenfold or greater increase in 21%, and affected 14% of critical habitat for endangered species, with stricter regulation associated with lower noise
6. Quiet Parks International — Wilderness Quiet Parks. The five certified Wilderness Quiet Parks — Zabalo River, Glacier National Park, Boundary Waters Canoe Area Wilderness, NamibRand Nature Reserve and American Prairie Reserve — which is how we establish that exactly one entry in our archive carries independent acoustic certification
7. Quiet Parks International — Assessment Criteria. That the certification criteria are offered only as a downloadable document and that the page itself publishes no decibel threshold, no noise-free interval and no listening-hour requirement — the reason we quote no QPI number in this piece
8. NamibRand Nature Reserve — official site. That NamibRand holds both a dark sky reserve designation and status as Africa's first wilderness quiet park — the single archive entry where two external instruments certify what our index only infers
9. DarkSky International — International Dark Sky Places. The six certification categories, 270-plus certified places across 22 countries and six continents, and more than 160,798 km² covered — a pass/fail certification model rather than a 0–100 score
10. Sky & Telescope — Light Pollution and Astronomy: The Bortle Dark-Sky Scale. That John Bortle published the nine-class scale in the February 2001 issue, that class 1 corresponds to a naked-eye limiting magnitude of 7.6–8.0 and class 9 to 4.0 or less, and that the classes are observer-described rather than instrument-defined
11. Science Advances 2(6):e1600377 (2016), Falchi, Cinzano, Duriscoe, Kyba, Elvidge, Baugh, Portnov, Rybnikova and Furgoni — The new world atlas of artificial night sky brightness (bibliographic record). The measured-tradition counterpart to Bortle: the Milky Way hidden from more than a third of humanity, 60% of Europeans and nearly 80% of North Americans, with 88% of Europe and 23% of world land surface under light-polluted skies
12. Science 379(6629):265 (2023), Kyba, Altıntaş, Walker and Newhouse — Citizen scientists report global rapid reductions in the visibility of stars from 2011 to 2022 (bibliographic record). The 51,351 naked-eye observations implying sky brightening of roughly 7–10% a year, and the authors' explicit statement that this outpaces satellite radiance estimates — the disagreement between observer and orbital measurement that we leave standing
13. OECD and European Commission Joint Research Centre — Handbook on Constructing Composite Indicators: Methodology and User Guide (2008). The standard reference work for building and criticising weighted composite indicators, jointly produced by the OECD Statistics Directorate and the JRC unit at Ispra — the framework against which this page judges its own instrument
14. Social Indicators Research 141(1):61–94 (2018), Greco, Ishizaka, Tasiou and Torrisi — On the Methodological Framework of Composite Indices (bibliographic record). The contemporary review establishing that weighting and aggregation are where the paramount criticism of composite indices falls, and that robustness analysis after construction is routinely neglected
15. Journal of the Royal Statistical Society Series A 176(3):609–634 (2013), Paruolo, Saisana and Saltelli — Ratings and Rankings: Voodoo or Science? (bibliographic record). That nominal weights rarely equal a component's main effect because component variables are correlated and unequally dispersed, demonstrated across six composite indicators including the Human Development Index — the basis for our admission that we do not know our own effective weights
16. Journal of the Royal Statistical Society Series A 168(2):307–323 (2005), Saisana, Saltelli and Tarantola — Uncertainty and Sensitivity Analysis Techniques as Tools for the Quality Assessment of Composite Indicators (bibliographic record). The argument that uncertainty and sensitivity analysis should accompany composite indicator construction throughout, in order to state the robustness of a ranking — a step we have never performed for the Silence Index
17. UNDP Human Development Reports — Human Development Index. That the HDI aggregates its three dimension indices with a geometric mean and uses the logarithm of income — the standard worked example of an index restraining compensability between dimensions, which our additive score does not do
18. Journal of Development Economics 99(2):201–209 (2012), Martin Ravallion — Troubling tradeoffs in the Human Development Index (bibliographic record). That the 2010 geometric-mean revision of the HDI introduced its own implicit and contestable tradeoffs between dimensions — evidence that changing an aggregation rule relocates the problem rather than removing it
19. Ecological Indicators 23:143–154 (2012), Nicky Rogge — Undesirable specialization in the construction of composite policy indicators: The Environmental Performance Index (bibliographic record). The documented case of a major environmental composite indicator rewarding specialisation in high-scoring components — the failure mode we identify in our own access-friction and night-darkness inputs
20. Journal of Sustainable Tourism 9(5):372–388 (2001), McCool and Lime — Tourism Carrying Capacity: Tempting Fantasy or Useful Reality? (bibliographic record). That the conditions needed to establish a numerical tourism carrying capacity are rarely met in practice, that published capacities often reflect capital investment rather than limits, and that the better question is what conditions are acceptable — the reason we score pressure rather than capacity
21. Tourism Management 7(4):254–258 (1986), A. M. O'Reilly — Tourism carrying capacity (bibliographic record). The originating definitional paper for tourism carrying capacity, cited here as the concept our footfall input borrows from rather than as an endorsement of it
22. Political Analysis 2:131–150 (1990), Barbara Geddes — How the Cases You Choose Affect the Answers You Get: Selection Bias in Comparative Politics (bibliographic record). The general demonstration that selecting cases on the outcome of interest distorts the conclusions drawn from them — the precise error that produces our 22-to-94 distribution
23. Tutorials in Quantitative Methods for Psychology 8(1):23–34 (2012), Kevin Hallgren — Computing Inter-Rater Reliability for Observational Data (bibliographic record). The standard practical account of Cohen's kappa and intraclass correlation for multi-coder ratings — the test we name as missing from our own single-scorer process
24. UNESCO World Heritage Centre — The New 7 Wonders of the World (statement, 9 July 2007). UNESCO's formal position that there is no link whatsoever between the World Heritage programme and the private New7Wonders campaign, that it declined repeated invitations to collaborate, that the poll reflected only the opinions of internet users, and that such a campaign cannot sustainably contribute to site preservation
25. World Bank Open Data — International tourism, number of arrivals (ST.INT.ARVL). The international arrivals series and its attribution to UN Tourism's Yearbook and Compendium of Tourism Statistics — the compiled global series we rely on, since UN Tourism's own pages would not serve to us
26. IATA — Air passenger demand full-year 2024 results (30 January 2025). Full-year 2024 traffic up 10.4% on 2023 and 3.8% above 2019, capacity up 8.7%, and a record annual load factor of 83.5% — the transport backdrop against which every score on this site is expected to fall
27. Icelandic Tourist Board (Ferðamálastofa) — Numbers of foreign visitors. Just under 2.3 million overnight foreign visitors in 2025 at 97.6% of the 2018 record, and the explicit statement that the count is made by manual counting of departing passengers before airport security and should be viewed with caution

Official and scientific bodies are the authority for safety, regulatory and scientific claims. Commercial operators are used only for practical detail such as tour length, meeting points and indicative pricing — never for a safety or scientific statement.