skjalfti.1881.is is an unofficial hobby and experimental site that shows earthquakes in Iceland automatically. It is a one-person project, independent of official institutions. This is not an official warning — in an emergency follow the Icelandic Met Office and Civil Protection.

The data comes from Skjálftalísa, the Icelandic Met Office open API, and are automatic preliminary results that may change on review. Magnitude (m_autmag), location and depth are automatically determined and often uncertain.

Other sources used: GNSS deformation from the Nevada Geodetic Lab; satellite thermal anomalies from NASA FIRMS (VIIRS); independent confirmation of larger events from EMSC and USGS; the weekly volcanic activity report from GVP (Smithsonian/USGS); geology map layers from Náttúrufræðistofnun, Landmælingar Íslands and IMO/FUTUREVOLC (CC BY 4.0).

Raw data download: CSV · GeoJSON (last 365 days).

• Volcanic systems are classified by a rough radius — an approximation, not official boundaries. Real geological boundaries (volcanic systems, central volcanoes + faults) can be toggled on the map as the Geology (NÍ) layer — from Náttúrufræðistofnun and IMO/FUTUREVOLC. The major glaciers (LMÍ) are shown by default. All CC BY 4.0.
• "Activity" colours are based on earthquake counts and are not the official aviation colour code.
• The front page opens with a status banner at the top — a question and answer ("Do I need to worry? — No/Yes …") coloured by severity. The severity level and the banner wording are computed deterministically (not AI): a red/orange VONA aviation warning from the Met Office at the top, then β tiers (unusually high / elevated seismicity), uplift / rapid deformation, and finally quiet. Strong eruption wording comes only from the Met Office VONA, never from the site itself. Below the banner is a structured view: active systems with signal chips (β, M, uplift, migration; no score number, just the chips), recent changes in the last 48 h (chips from journal signatures, no AI prose) and other active areas. One "Status now" AI summary for the whole country (plain language) is logged in the journal and on Telegram. Each system page carries a "Status (automatic)" sentence — an AI rephrasing of deterministically computed facts (microseism level + weather correlation); the AI cannot add forecasts or magma claims.
• The front page uses progressive disclosure: heavy charts — earthquake frequency, uplift (GNSS), InSAR, thermal anomalies (FIRMS), volcanic activity (GVP) and tremor — are tucked into a single expandable section, "Trend, uplift and background". The volcanic-systems table shows only the most active systems by default (≥3 events at M≥0.1, or a larger event); the rest appear under "show all systems". The table counts M≥0.1 events (not micro-quakes) so the count is symmetric with the Activity column and the β value; micro-quakes (M<0.1) are visible on the map. If the AI service fails, the site reuses the last good summary instead of "unavailable" — the status banner and figures stay correct because they are computed.
• b-values, percentiles and trend charts are computed from the data (preliminary).
• Earthquakes alone do not show magma movement — deformation (GPS) and gas are needed to assess eruption hazard.

• β (swarm signal) was backtested over 2 years: β≥2 flags an ongoing swarm at about 5× the baseline rate, but predicts large earthquakes poorly — an activity measure, not a forecast. The β number itself is hidden in a tooltip; the Activity column shows plain language ("elevated", "unusually high") so the reader need not interpret the number.
• Historical probabilities (e.g. jökulhlaup) are phrased as "~1 in N" rather than an exact percentage, and b-values are shown with an uncertainty range, to avoid overstating precision.
• Cumulative moment (Mo) shows real energy release and is barely affected by older quakes lacking a magnitude.
• Hypocenter migration is shown only when activity moves clearly in one direction (a possible dike intrusion).
• GNSS uplift comes from the Nevada Geodetic Lab — key station per volcanic system (88 stations total; one key station shown per system) with about a three-week lag, not the Met Office real-time solution. The trend is computed robustly (Theil-Sen) so single artificial steps do not bias it; stations with an obvious equipment step and no real trend are hidden.
• Declustering (swarm character): Zaliapin–Ben-Zion declustering does not work nationwide (all activity pooled) nor for individual systems, but per volcanic belt on the homogeneous catalogue (after Feb 2026) a weak separation of swarm vs background emerges. From it a "swarm character" is computed — the share of independent (background) events in the recent activity of each belt — distinguishing a short-lived swarm from a more diffuse background increase. This is an indicator, not exact declustering; the background rate turns out about 3× steadier than the raw count.
• Tremor (RSAM): IMO tremor plots (3 frequency bands: 0.5–1, 1–2, 2–4 Hz) are fetched and embedded on the 8 main volcanic systems. Automatic extraction turns the plot into numbers (CSV per station with ~520 points over 11 days), separating the baseline (microseism) from spikes (earthquakes). A model of baseline ~ wind (IMO, nearest weather station) + waves (Open-Meteo) explains 54–89% of variance at coastal stations. A tripwire flags when the baseline rises above what the weather explains (z > 3 standard deviations). The status sentence under each plot is numerically grounded: the LLM only rephrases deterministically-computed facts and cannot add eruption or magma claims.
• API delay on recent earthquakes: the Skjálftalísa API returns events only after initial review. The newest automatic events (quality 50 on the Met Office web) typically appear 1-6 hours later in the API than on the IMO web table — so counts in the last 6 hours may be lower here than on vedur.is (typically 2-4 events with lowest quality). Older counts (24h+) are accurate.
• Catalogue not time-homogeneous: before February 2026 about half of events lacked an automatic magnitude, so magnitude-dependent figures over longer periods under-count the earlier part. This coincides with the Met Office switching from the SIL system to SeisComP in Feb 2026.

The site received significant expansions in May 2026 after review by a 4-person panel (geologist · scientific graphics · UX · science communicator):

• Mogi model on Svartsengi page — magma-chamber volume change (ΔV) time series from a 5-station cluster (SENG, HS02, NBIO, ORFC, SKSH). Fixed source at 4.5 km depth (Sigmundsson et al. 2024); linear-LS for ΔV. Daily fit on a 30-day window.
• Svartsengi cycle overlay — 9 Sundhnúkur eruption cycles stacked over a common x-axis (days since cycle start). Shows how the current cycle compares to the median of prior peaks.
• Baseline lengths between SENG and 4 nearest stations — common-mode filtering removes atm/geomagnetic noise and makes the chamber signal stronger than at single stations.
• Depth cross-sections (E-W and N-S) on every system page — show rising magma columns during active intrusion.
• Bárðarbunga reloading since Holuhraun end 2015 — cumulative seismic moment (M₀ = 10^1.5M+9.1 N·m) compared to the Holuhraun moment (~2×10²⁰ N·m, Gudmundsson 2016).
• Migration-rate time series on every system page — rolling 24-hour window of hypocenter migration past 14 days. Colored by intensity (red > 5 km/day = possible dike intrusion).
• VONA Aviation Colour Code automatically scraped from the Met Office and shown on relevant system pages (Reykjanes, Grímsvötn, Bárðarbunga, Katla).
• Jökulhlaup history on Bárðarbunga/Grímsvötn/Katla — historical renewal model with days-since-last + median interval + probability over next 6 months. Not realtime monitoring (Met Office does not publish water levels openly).
• GNSS step detection on glacial stations (GFUM, KISA, AUST) — detects sudden Up drops that may indicate ice-shell subsidence and thus jökulhlaup. Winter snow filtered out (May-Oct only).
• "Last eruption" history box at the top of each system page — timeline of the last event and historical context.
• "Did you know?" story card with historical context (Holuhraun, Mýrdalssand 1918, Hekla = "gateway to hell", Eyjafjallajökull 2010 flight closure, etc.).
• "What if..." scenario popovers with civil-protection-style answers for 8 systems — timeline, jökulhlaup discharge, settlement impact, historical comparison.
• Tremor band heatmap — 3 frequency bands (0.5-1, 1-2, 2-4 Hz) as colored strip image over 9 days; high 2-4 Hz + low 0.5-1 Hz = possible magma signal.
• InSAR card on Reykjanes systems with link to COMET LiCS portal (Sentinel-1 radar deformation).
• GNSS: stations increased from 13 to 88 (all Iceland stations in NGL llh.out). New N/E/Up 3-panel stack on Svartsengi like Met Office time series. New gnss_stations table with (lat,lon,elev).
• Scientific errors fixed: Mogi caveat (assumptions stated), 9 dike intrusions vs 7 eruptions at Sundhnúkur, Holuhraun moment 5×10¹⁹ → 2×10²⁰ N·m, Katla probability minimum 2% (was 0%).

• Volcanic system list expanded from 35 to 40 tuples after a 2-expert review (geologist + GIS expert). New systems added: Hrómundartindur (east of Hengill), Hágöngur (SW of Tungnafellsjökull), Snæfell in Múlaþing (NOT Snæfellsjökull), Helgrindur on Snæfellsnes, Eldey/Reykjanestá (active 2022–23 swarm cluster), Eyjafjarðaráll (TFZ-graben). Ten coordinate fixes — including Ljósufjöll (28 km error) and Kolbeinseyjarhryggur (110 km east of the actual ridge).
• Three-expert content review of every kerfi page (geologist, Icelandic geology historian, UX expert) yielded 22 concrete issues. 19 fixed: 10 scientific factual errors corrected (e.g. Katla "12 km" → ~4 km, Reykjanes "1–200 m/s" → 1–20 m/h, Hengill "30% of electricity" → ~15%, Öræfajökull glacial flood "Skeiðará" → Virkisá/Kotá/Svínafellsá), 8 historical enrichments added (Lakagígar 1783–84/Móðuharðindi famine, Eldfell 1973/seawater cooling operation, Vatnaöldur 877/Settlement tephra, Stöng/Hekla 1104, 1784 Skálholt earthquake → Reykjavík capital, Berserkjahraun/Eyrbyggja saga, 1907 von Knebel disappearance, Öræfi place-name).
• β "apples-to-apples" magnitude floor: after the ingest API filter was lowered to M≥-3, small events were only in the recent window, not the 2-year baseline. This inflated β by ~70% (Hengill falsely classified as "unusually high β6" when the apples-to-apples value is ~2.6 "elevated"). β now applies the same M≥0.1 floor on both sides (mag IS NULL still allowed) so the comparison is symmetric.
• Lava WMS overlays on the map: three new toggleable Leaflet layers — Lava (Sundhnúkur 2023–25 + Fagradalsfjall 2021–23 from Náttúrufræðistofnun WMS), Grindavík lava barriers, and Glacial cauldrons (IMO WMS). On the Reykjanes-Svartsengi system page, G1–G9 Sundhnúkur lava flows are shown individually with a viridis colour scale (cql_filter on vinnuheiti).
• UX reordering of system pages: "Status (automatic)" and the VONA flag are now at the top; historical context (KERFI_DESC, Did-you-know, timeline, What-if) follows. Previously the Status was buried below 10+ blocks. Repeated journal entries collapse into a single "Repeated × N since <earliest> — last <latest>" row when consecutive entries have near-identical text.
• "What if..." scenarios added to 5 systems that were flagged active on the front page but had an empty system page (false-alarm trap): Hengill, Eldey/Reykjanestá, Eyjafjarðaráll (explicitly clarified as not a volcano but a graben), Suðurlandsbrotabelti, Tjörnes/Grímsey.
• Qwen number-word fix: the LLM occasionally slipped Scandinavian number words into Icelandic (e.g. "Elleve" instead of "Ellefu", "Tolv" instead of "Tólf"). A deterministic post-processing pass corrects them.
• Region label "Annað" → "Önnur svæði": Qwen read "Annað" as the dative of the female name Anna and hallucinated "M2.9 á Önnu". Renamed to "Önnur svæði" (neuter plural) so the LLM is less likely to grammatically reinterpret it as a person.

The journal (annáll) logs "Status now" whenever the situation materially changes — at most once every 2 hours, unless a new M≥4.0 quake overrides the wait. The same rule applies to the per-system journal on each volcanic-system page. It is posted automatically to the Telegram channel t.me/skjalftavakt and as an RSS feed, so you can get updates without opening the site.

Automatic push notifications (ntfy) about large/nearby earthquakes and swarms are unofficial and do not replace official warnings. Always confirm at vedur.is.

Subscribe: open ntfy.1881.is/skjalftar in a browser, or in the ntfy app add a subscription with server https://ntfy.1881.is and topic skjalftar. The channel is self-hosted and write-protected — only this site can publish to it, so alerts cannot be spoofed.

Feedback, corrections and questions: skjalfti@1881.is.

Icelandic Met Office — vedur.is
Civil Protection — almannavarnir.is
Safetravel — safetravel.is
Emergency line: 112