Mike Hann's org
Community member@mikehann
Harvey's returning rainbands: Houston to Beaumont, 2017
Press Play or step through the original radar scans. Watch 129 selected original NOAA KHGX Level II lowest-sweep reflectivity scans from 26 August 12:01 to 30 August 11:58 UTC 2017 over southeast Texas. The fixed geospatial view shows rainbands repeatedly crossing Houston, Galveston and Beaumont while Hurricane Harvey weakened and moved slowly near the Gulf Coast. Scans are approximately 45 minutes apart and the viewer swaps observed volumes without inventing intermediate weather. Radar reflectivity in dBZ depicts precipitation structure, not wind speed, hurricane category, rainfall accumulation or flood inundation. The NHC post-analysis track and intensity are separate evidence: https://www.nhc.noaa.gov/data/tcr/AL092017_Harvey.pdf . Original NOAA radar archive via Unidata: https://registry.opendata.aws/noaa-nexrad/ .
Updated 21 Oct 2026 · 2 views
- showcase
Mexico City Is Sinking — NISAR Radar Interferometry
Mexico City is sinking into the lakebed it was built on. This 84-day NISAR radar series measures motion along the satellite’s line of sight; two GNSS stations provide a limited independent check. WHAT YOU ARE LOOKING AT. Ground velocity along the satellite's line of sight over 84 days (25 October 2025 to 17 January 2026), from eight NISAR radar passes. Blue indicates motion away from the satellite, consistent with sinking where vertical motion dominates. The eastern lobe lies over former Lake Texcoco. Research links the basin’s long-term subsidence to compaction of lake clays as groundwater is withdrawn; this short radar series alone does not establish the cause. The western hills sit within a few mm/yr of zero. DRAG THE TIMELINE across seven dated radar measurements, twelve days apart. The 1,785 points show cumulative displacement since 25 October 2025. Compare the eastern basin with the steadier western hills, and click a point for its value at the selected date. IS IT TRUE? Two GNSS stations recorded continuously inside the same 84-day window, so their rates can be compared against the radar directly — not against a multi-year average, which would be a different measurement: MMX1 (fast) GNSS -197.2 mm/yr radar -176.2 difference 21.0 ICMX (stable) GNSS +12.3 mm/yr radar +13.2 difference 1.0 The 21 mm/yr disagreement at MMX1 is smaller than this product's own stated uncertainty (1-sigma 26.1 mm/yr, from the scatter between interferograms rather than a formal error bar). Those two comparisons are encouraging, but they do not validate every mapped point or establish a long-term rate. THE OTHER ELEVEN STATIONS ARE GREY, and that is the point. Eleven of the thirteen GNSS receivers inside this frame have no data inside the radar window; their long-term rates are shown for context but they cannot validate anything here. Scoring against long-term rates instead produced a flattering agreement that turned out to be two errors cancelling. Click a grey station to see its span. HONEST LIMITS. Amplitude rests on one fast station. The 80 m product is the trustworthy one: the same data at 20 m reads about 1.5x low, because a broad subsidence bowl has steep phase gradients everywhere and finer pixels lose whole radar cycles to them. Values over the fastest-moving ground should be read as a floor. Line-of-sight is not vertical — for near-vertical subsidence divide by about 0.75 to convert. WHAT WAS REMOVED, AND WHY. 1.0% of pixels (1,580 of 157,543) read beyond ±600 mm/yr and are blanked. One full radar fringe over this 84-day span is 530 mm/yr, so a value past 600 cannot be ground motion here — it is a residual whole-cycle unwrapping error, and the run's own phase-closure check independently flagged unwrapping problems in the stack. They are masked rather than displayed because showing them would put the map's most saturated colours on its least trustworthy pixels. Nothing inside ±600 mm/yr has been altered. METHOD. NISAR L-band GUNW products (JPL-unwrapped, 80 m), track 113 descending, 8 acquisitions / 8 interferograms. Corrections: water masking, whole-cycle unwrap-error repair, elevation-correlated and turbulent atmospheric screens — the latter projected orthogonal to the deformation field, because a smooth atmospheric screen otherwise absorbs a smooth subsidence bowl and quietly removed about 68 mm/yr of the real signal. SBAS inversion with weighted least squares, referenced to the scene median. SOURCES. Radar: NISAR (NASA/ISRO) GUNW via Alaska Satellite Facility. Ground truth: Nevada Geodetic Laboratory, MIDAS solutions and daily position series, IGS20 frame.
WGS 84 · Updated 20 Oct 2026 · 5 views
North America Smoke & Active Fires
Regional companion to the fire-scale imagery map. GEOS-CF smoke severity as a paletted 0.25° raster (3-hourly), official BCWS + NIFC perimeters and BC evacuation orders (both daily time series, with people and homes affected), and FIRMS hotspots both coloured and sized by fire radiative power — 88% of detections fall in the two faintest classes, so size is what makes the handful of major fires findable at continental zoom. === READING THE TIMELINE === Every layer rolls its own window, and they do not line up exactly: smoke 2026-09-20 to 2026-10-19; FIRMS 2026-09-21 to 2026-10-20; perimeters 2026-09-20 to 2026-10-20; evacuations 2026-09-20 to 2026-10-20. The agency feeds (perimeters, evacuations) publish a running current state with no history, so a day nobody captured while it was current cannot be recovered later. Where a layer has no frame for the day you have scrubbed to, the map draws that layer's NEAREST available frame rather than leaving it blank — so at the edges of the window, an outline may be from a neighbouring day rather than the one on the scrubber. === LATENCY: NEAR REAL TIME, NOT LIVE === FIRMS refreshes the CURRENT day every 3 hours, so active fire detections are hours old rather than the 24–48 h that publishing only completed days used to force. Today's frame is stamped at its day's 00:00 UTC and fills in as satellite passes arrive, so it grows through the day and is finalised once the day closes. GEOS-CF is the slow one: it publishes roughly 35 h behind real time in once-daily 24-hour batches, so the newest smoke frame is typically 18–36 h old — an upstream limit, not a pipeline delay. Perimeters and evacuations refresh daily. Treat this as a monitor, not a live feed — for an active incident, go to the responsible agency. === WHAT 'SMOKE SEVERITY' ACTUALLY MEASURES === Carbonaceous aerosol — organic plus black carbon PM2.5 — binned to the five US EPA PM2.5 AQI breakpoints (5 / 12 / 35.4 / 55.4 / 150.4 µg/m³). Carbonaceous rather than total PM2.5 on purpose: total folds in dust and sea salt, which makes the Sahara read as extreme smoke. The trade-off is that it still counts urban and industrial combustion, so it over-reads over cities. GEOS-CF is a 0.25° global model, so treat it as regional context, not a local air-quality reading. === SOURCES === NASA FIRMS (MODIS + VIIRS SNPP/NOAA-20/NOAA-21, near-real-time). NASA GEOS-CF v2 analysis. BC Wildfire Service and NIFC official perimeters. BC evacuation orders and alerts. All free and open. Class shares re-measured each run over the 30 published days, 3,675,686 detections.
WGS 84 · Updated 20 Oct 2026 · 111 views
Hurricane Michael: the radar eye reaches Georgia
Press Play or drag the time slider to follow Michael's eye. Watch Hurricane Michael's radar eye move from the Gulf of Mexico into Georgia on 10 October 2018. The slider contains 50 selected observed NOAA KEVX NEXRAD lowest-sweep reflectivity volumes from 07:10 to 23:39 UTC, approximately 20 minutes apart. Each original radar volume was regridded to geographic coordinates using its nearest native gates; the map does not invent in-between weather. Coastal radar cannot follow the hurricane over the whole open Atlantic. Colours show precipitation reflectivity in dBZ, not wind speed or hurricane category. The National Hurricane Center estimated 160 mph maximum sustained wind at landfall separately from these images. NOAA Level II archive via Unidata: https://registry.opendata.aws/noaa-nexrad/ . NHC final report: https://www.nhc.noaa.gov/data/tcr/AL142018_Michael.pdf .
Updated 20 Oct 2026 · 13 views
- showcase
Isaias: radar over the Florida Panhandle
Hurricane Isaias made landfall near Destin around 01:30 UTC on 10 October 2026; NHC estimated the storm's maximum sustained wind at landfall at 105 mph. Scrub the observed NOAA KEVX lowest-sweep reflectivity scans from the Gulf approach through landfall. Colours show precipitation in dBZ, not wind or outage. Pensacola and Destin markers identify the nearby cities, not precise wind stations or the storm centre. NHC reported a 100 mph gust at a station near Pensacola. NASA Black Marble imagery shows markedly reduced nighttime lights along the Pensacola–Destin corridor on October 11, with partial return October 12; August 2026 is a typical-night composite and the later images are single nights. Dark hatched areas lacked valid satellite data because of cloud cover, and the 30 m Black Marble HD detail is modeled/downscaled rather than directly resolved by VIIRS. NASA story: https://science.nasa.gov/earth/earth-observatory/isaias-dims-the-gulf-coast/ ; NHC update: https://www.nhc.noaa.gov/text/refresh/MIATCUAT4+shtml/752822.shtml ; NOAA archive: https://registry.opendata.aws/noaa-nexrad/ .
WGS 84 · Updated 20 Oct 2026 · 2 views
- showcase
Oklahoma's Man-Made Earthquakes
Before 2009 Oklahoma had about one magnitude-3 earthquake a year. In 2015 it had 887. In 2025 it had 6. This map puts the injection wells and the earthquakes on one timeline — and next to a control, so you can check the claim instead of taking it. ▶ NARROW THE TIME RANGE, then drag it year by year. Blue bubbles are Class II saltwater disposal wells, sized by the barrels each one took that year. Warm dots are earthquakes, sized and coloured by magnitude. Watch disposal swell through 2014, the earthquakes arrive in 2015, and both fall away after the state ordered volume cuts. THE CONTROL, AND WHY IT MATTERS. Plot wells beside earthquakes and the obvious objection is that wells get drilled where the geology is complicated, and complicated geology is where earthquakes happen anyway. Oklahoma answers that itself. The state injects two comparable rivers of fluid underground each year, and the same regulator records both in the same file. DISPOSAL goes deep — usually into the Arbuckle, which sits directly on the crystalline basement where the faults are. ENHANCED RECOVERY pushes a similar volume back into shallower producing zones to sweep out oil. In 2015: 1,203 million barrels of disposal against 1,115 million of recovery. From 2011 to 2014 disposal rose 85%; recovery rose 9%. Turn on the enhanced-recovery layer. It is drawn on exactly the same size scale as the disposal layer, so the bubbles are directly comparable. A comparable volume of fluid goes into the same counties in every year of the animation — and it never moves, while the earthquakes come and go. If earthquakes tracked injection as such, both layers would show it. WHAT THE NUMBERS SAY. Across 0.15° cells, the share of cell-years with an M3+ earthquake climbs with disposal volume: 1.2%, 4.6%, 6.9%, 11.2%, 15.8% from the lowest fifth to the highest. The same table for enhanced recovery is flat and near zero — 5.8%, 2.2%, 0.0%, 3.7%, 0.0% — with its top bin still taking a median 1.6 million barrels a year. In the 277 cells hosting BOTH kinds of well (same rock, same faults, same seismometers) disposal correlates with earthquakes at rho +0.22, recovery at −0.04. Inside individual cells over time, where geology is fixed by construction, disposal has a median correlation of +0.42 with the following year's earthquakes and is positive in 79% of cells; recovery, +0.09. THE REVERSAL. Take the 20 highest-disposal cells of 2014–15 and follow those same cells, never reselected: 96 M3+ earthquakes in 2015 — then zero in every year from 2019 through 2025. WHAT THIS IS NOT. This does not attribute any individual earthquake to any individual well. A well's volume is what its operator reported injecting, not an accusation about what it caused. The mechanism — pore pressure diffusing into stressed basement faults — comes from the published literature, not from this map. LIMITS. Every count is M≥3.0, because Oklahoma's seismic network densified after 2009 and lower-magnitude counts partly measure the instruments rather than the ground; the smaller events are drawn for texture and excluded from every statistic. Note the direction of that bias: better detection after 2015 works against the decline shown here, so the fall is if anything understated. About 40% of well-months are operator-calculated rather than metered, and each well carries its own value. Oil prices also collapsed in 2015–16, cutting produced water independently of regulation; this map cannot separate the two effects. Earthquakes shown include southern Kansas, where the same disposal trend continues across the state line — 887 is the Oklahoma-only figure, 945 the whole study box. SOURCES. Oklahoma Corporation Commission Form 1012A injection volumes, 2011–2025 (23,962 wells, 1,995,545 well-months) and USGS ComCat earthquakes, 2000–2026. Both public record.
WGS 84 · Updated 20 Oct 2026 · 9 views
- Weather
Every tropical cyclone
Every tropical cyclone track in IBTrACS v04r01, seasons 1842 to 2026: 13,239 storms drawn as 50,461 stretches, each coloured by its Saffir-Simpson category and drawn wider where the wind was stronger. Drag the two time handles to pick which seasons show; click a track for the storm. Read this before comparing decades. Before routine satellite coverage (from about the 1970s) storms away from shipping lanes and coasts were often missed, and their winds often went unrecorded (grey tracks have no wind on record). More tracks in recent decades is partly better observation, not only more storms. NCEI also notes that some pre-1950 storms were never matched across sources, which inflates early counts, so the number of tracks per decade is not a clean trend. Agencies also measure wind over different averaging periods (see below), and 2026 tracks are provisional real-time data until the agencies reanalyse them. Source: International Best Track Archive for Climate Stewardship (IBTrACS) v04r01, NOAA National Centers for Environmental Information, file dated 02 Oct 2026. Gahtan, Knapp, Schreck, Diamond, Kossin and Kruk (2024), doi:10.25921/82ty-9e16; Knapp et al. (2010), BAMS 91, 363-376. NOAA distributes IBTrACS with full and open access under the World Data Center policy, with WMO Resolution 40 as the guide for commercial use. Wind and category. The Saffir-Simpson scale is defined on 1-minute sustained wind. Where a US agency (NHC, CPHC, JTWC) reported wind we use it as is. Elsewhere we use the official regional centre's wind, which is a 10-minute mean (3-minute at New Delhi), converted to a 1-minute equivalent with the WMO at-sea factors of Harper, Kepert and Ginger (2010): 10-minute = 0.93 x 1-minute. Every stretch names its wind source. Thresholds in 1-minute knots: tropical depression under 34, tropical storm 34 to 63, Category 1 64 to 82, Category 2 83 to 95, Category 3 96 to 112, Category 4 113 to 136, Category 5 137 and up. Stages IBTrACS marks extratropical are drawn in grey-blue whatever their wind, because the scale describes tropical cyclones only. Method. One line per unbroken stretch of one category within a storm; consecutive stretches share a point so the track stays continuous. Wind gaps of up to 24 hours between two reports are filled by interpolating in time (IBTrACS leaves the WMO wind blank at its own 3-hourly interpolated positions). Spur tracks are left out, as IBTrACS advises. Lines are split at the antimeridian. The scrubber places every storm of a season on 1 January of the season year; Southern Hemisphere seasons run July to June and take the year they end in. "Formed in" is the basin of the first track point. Peak wind and category leave out extratropical stages.
WGS 84 · Updated 20 Oct 2026 · 3 views
- History
Ever Given: 35 minutes before the Suez grounding
Follow 11 GPS fixes recorded by Ever Given's voyage data recorder in the final 35 minutes before its grounding in the Suez Canal on 23 March 2021. The orange point is the recorded GPS antenna position at 05:41:35 UTC, 30°00.9913′N, 32°34.8188′E. It is not a measured bow location or the later final orientation of the hull. Straight links between positions are an illustration of the approach, not a continuous observed ship track. Attributed source: Panama Maritime Authority, M/V Ever Given investigation R-026-2021-DIAM, pages 38–57, hosted at https://gcaptain.com/wp-content/uploads/2023/07/Final-Investigation-Report-Ever-Given-23-March-2021.pdf. Each plotted point links to its PDF page. The ship was 400 m long and nearly 60 m wide, according to the Boskalis salvage press release of 29 March 2021: https://boskalis.com/press/press-releases-and-company-news/suez-canal-unblocked-we-pulled-it-off?s=08. The map uses only reported point positions; the vessel and canal are not surveyed here.
WGS 84 · Updated 20 Oct 2026 · 4 views
Rivers to the Sea
Every river on Earth coloured by the sea it finally reaches, with line width following modelled mean discharge. Switch on the basin layer to see each drainage basin in its own colour. Sources and licences: river network, discharge and sub-basins from RiverATLAS and BasinATLAS v1.0 (HydroATLAS; Linke et al. 2019, Scientific Data 6:283), which carry the HydroRIVERS and HydroBASINS v1.0 data (Lehner and Grill 2013) under CC BY 4.0. Coastline and river names from Natural Earth (public domain). Receiving-water boundaries are our own. Caveats: HydroRIVERS is modelled from a 15 arc-second DEM, coarser above 60 N. Discharge is a modelled long-term natural mean (WaterGAP, 1971 to 2000) that ignores dams and diversions, so the Colorado and the Amu Darya are drawn reaching their seas. Rivers are thinned for display: 25,000 km2 upstream when zoomed out, 5,000 km2 closer in; the numbers below use all 8.5 million reaches. Where one sea ends and the next begins is a drawing choice: IHO limits (S-23) almost everywhere, Canada's ocean drainage areas in northern North America (Hudson Strait, Ungava Bay and Foxe Basin count as Hudson's Bay). There is no Southern Ocean class: its draft IHO limit (60 S) reaches no land in this dataset, so south of Australia the coast drains to the Indian Ocean west of Tasmania and the Pacific east of it; the Murray-Darling counts as Indian Ocean. The Red Sea is part of the Indian Ocean; the Sea of Marmara is part of the Mediterranean; the Kattegat is part of the Baltic. Basin names come from Natural Earth rivers matched at the mouth; unnamed basins show where they end. Share of the world's drained land: No outlet 24%, Atlantic 22%, Pacific 14%, Arctic Ocean 13%, Indian Ocean 11%, Gulf of Mexico 4%, Mediterranean 3%, Hudson's Bay 3%, Black Sea 2%, Baltic Sea 1%, Persian Gulf 1%, Gulf of California 1%, Caribbean 1%, North Sea 1%. Share of the water reaching the sea: Atlantic 37%, Pacific 28%, Indian Ocean 13%, Arctic Ocean 9%, Gulf of Mexico 3%, Hudson's Bay 2%, Caribbean 2%, Baltic Sea 1%, Mediterranean 1%, Black Sea 1%, North Sea 1%, Persian Gulf 0%, Gulf of California 0%. The longest path to the sea is 6,846 km (Nile). Click a river for its basin, where it ends, its modelled discharge and how far it still has to go.
WGS 84 · Updated 20 Oct 2026 · 17 views
- Urban
Manhattan and Brooklyn by year built
Every building in Manhattan and Brooklyn, New York City, coloured by the year it was finished: 371,409 footprints with a known year, plus 3,902 in grey whose year is not recorded. Drag the time handles to see which of today's buildings had been finished by a given year. Click a building for its year, roof height and floors. Two things to know before reading it as history. This shows the buildings standing today by the year they were built, not the city as it stood in that year: everything demolished since is missing, so old neighbourhoods look emptier in the past than they were. And many years are estimates. Across the five boroughs, city records put 46% of buildings on a year ending in 0 (about 10% would if dates were exact), and the biggest single years are 1920 (129,585), 1925 (108,674), 1930 (112,102). The Department of City Planning says these years are generally right to the decade but not to the year, and that many buildings from the 1800s to early 1900s carry 1899 to 1901. On the timeline those round years arrive as bursts. Popups mark buildings on such an over-used year. Sources. Building footprints: NYC Office of Technology and Innovation, BUILDING (Building Footprints), NYC Open Data dataset 5zhs-2jue, version of 2026-10-05. Address and floors: NYC Department of City Planning, PLUTO 26v2, NYC Open Data dataset 64uk-42ks, joined on the MapPLUTO tax lot; floors are for the tallest building on the lot. NYC Open Data is published under NYC Administrative Code 23-502 (Local Law 11 of 2012) with no registration, licence or restriction on use; the City asks users to name the source and version and to describe changes, which this paragraph and the next do. Changes we made. Construction years that are missing or later than 2026 are treated as unknown (10,075 buildings citywide). Roof heights are converted from feet to metres. Footprints are simplified by about half a metre and placeholder triangles and temporary structures are left out. Where a tall tower kept the year of an older building on its lot, we used the newer tax-lot year (13 towers) or the documented completion year (111 West 57th Street 2021, 26 Broadway 1928, The San Remo 1930). Each timeline step is the middle of the year built (1 July).
WGS 84 · Updated 20 Oct 2026 · 11 views
- showcase
Ukraine's Deep Strikes — Russian Refineries, Terminals & Retail Warehouses
Two Ukrainian deep-strike campaigns on one timeline: five months of steady pressure on Russian oil refining and export, then a burst against consumer logistics. Almost every target is inside Russia; one, WB Simferopol, is in Russian-occupied Crimea. Newest strike drawn: 15 September 2026. REFINING (cyan) — 48 strikes on 23 refineries, 2 April to 26 August 2026. Kstovo and Tuapse were hit five times each; Moscow, Perm, Ryazan and Syzran three. It reaches far past the front: Omsk is 2,500 km from Ukrainian-controlled territory, Orsk 1,900 km, Perm 1,485 km. Point size is refining capacity. WAREHOUSES (red/orange/yellow) — 49 strikes on 42 logistics sites, 18 July to 15 September 2026. The campaign opened on WILDBERRIES, which Ukraine accuses of supplying the Russian military with drone components, and on 22 August spread to its rival OZON: "after Ukrainian units targeted 15 of the largest logistics hubs belonging to Wildberries, it was Ozon's turn" (Ukraine's defence ministry). Nine Ozon warehouses struck so far — Belgorod Estate Logistics, Blagoveshchensk, Chapayevsk, Enem, Nevinnomyssk, Orenburg, Saratov, Taganrog and Tyube. Twelve were destroyed outright, including WB Elektrostal (230,000 m2) and WB Koledino north (200,000 m2) (one further unit the source calls an accidental fire is excluded). The tables record 17 killed and 55 injured, but only where a row itemises them — casualties_recorded says which. Colour is the SOURCE's damage class, size is floor area; damage_caveat carries "20% destroyed" and similar qualifiers a three-way class cannot. EXPORT TERMINALS (green) — 20 strikes on 12 terminals, depots and pumping stations, 5 April to 18 July 2026 — Sheskharis (3), Yugnefteprodukt (3), Primorsk (2), Vysotsk (2), Taman (2), Petersburg Oil Terminal (2). ENERGY SITES (grey rings) — repeat strikes sit at the SAME coordinate, so 33 of them are hidden under another point. One ring per site, sized by how many times it was hit (21 of 35 sites more than once); its popup lists every date. Points are never moved to un-stack them. RADAR CHECK (rings on warehouses) — 40 struck sites scored against Sentinel-1 AT THEIR OWN STRIKE DATE. 15 red = major structural change confirmed; 5 amber = confirmed, but the target is smaller than the class the detector was validated on (AUC 0.866, large structures), so read those as leads; 18 blue = no major change seen, which is NOT "undamaged" — partial damage is below C-band physics; 2 grey = no usable pass. This is a SEPARATE assessment from the dot's colour: they agree on 10 of 12 sites the source calls destroyed. Change detected LATER is on post_strike_change, not the ring. Last run 2026-10-20. SOURCES — every strike carries the ORIGINAL report for that date, not just the Wikipedia page it was compiled from: 114 of 117 drawn strikes link a named outlet (14 wire agency, 86 established outlet), 1 rests on a social post and 3 have no citation on their source row — source_tier says which. Compiled from Wikipedia's incident tables ('Ukrainian strikes on Wildberries warehouses', 'Deep strike campaign'), re-read nightly (last 2026-09-21); Wikipedia marks those tables incomplete and they lag reporting by days. Text CC BY-SA 4.0; coordinates partly OpenStreetMap, ODbL. Imagery: Copernicus Sentinel-1/2 via Microsoft Planetary Computer. Not a verified strike record and not an intelligence product. READING IT — TWO DISTANCES: distance_km is to Ukrainian-CONTROLLED ground (the source's measure); border_distance_km is to the internationally recognised boundary and is what the dashed rings are drawn from, so it is the smaller of the two. PRECISION: geocode_precision is on every point — some terminals are town centroids, not the tank farm. WINDOW: strikes from 1 April 2026 only, so the timeline resolves this campaign; 37 earlier refinery and 12 earlier terminal strikes are in the source data but not drawn, and strike_no_at_site counts over the full record.
WGS 84 · Updated 20 Oct 2026 · 64 views
- showcase
Hurricane Ian: Category 4 radar
Hurricane Ian near Category 4 landfall. NOAA NEXRAD lowest-sweep reflectivity scan starting 2022-09-28 19:01:42 UTC, georeferenced from nearest native radar gates. Reflectivity colours show precipitation structure, not wind speed. Same 5–65 dBZ colour scale as the Sarkoa hurricane radar countdown. Category and landfall details: https://www.nhc.noaa.gov/data/tcr/AL092022_Ian.pdf . NHC wind scale: https://www.nhc.noaa.gov/aboutsshws.php . Radar archive: https://registry.opendata.aws/noaa-nexrad/ .
WGS 84 · Updated 20 Oct 2026 · 1 view
- showcase
Hurricane Idalia: Category 3 radar
Hurricane Idalia near Category 3 landfall. NOAA NEXRAD lowest-sweep reflectivity scan starting 2023-08-30 11:47:28 UTC, georeferenced from nearest native radar gates. Reflectivity colours show precipitation structure, not wind speed. Same 5–65 dBZ colour scale as the Sarkoa hurricane radar countdown. Category and landfall details: https://www.nhc.noaa.gov/data/tcr/AL102023_Idalia.pdf . NHC wind scale: https://www.nhc.noaa.gov/aboutsshws.php . Radar archive: https://registry.opendata.aws/noaa-nexrad/ .
WGS 84 · Updated 20 Oct 2026 · 1 view
- showcase
Hurricane Sally: Category 2 radar
Hurricane Sally near Category 2 landfall. NOAA NEXRAD lowest-sweep reflectivity scan starting 2020-09-16 09:46:44 UTC, georeferenced from nearest native radar gates. Reflectivity colours show precipitation structure, not wind speed. Same 5–65 dBZ colour scale as the Sarkoa hurricane radar countdown. Category and landfall details: https://www.nhc.noaa.gov/data/tcr/AL192020_Sally.pdf . NHC wind scale: https://www.nhc.noaa.gov/aboutsshws.php . Radar archive: https://registry.opendata.aws/noaa-nexrad/ .
WGS 84 · Updated 20 Oct 2026 · 1 view
- showcase
Hurricane Florence: Category 1 radar
Hurricane Florence near Category 1 landfall. NOAA NEXRAD lowest-sweep reflectivity scan starting 2018-09-14 11:15:03 UTC, georeferenced from nearest native radar gates. Reflectivity colours show precipitation structure, not wind speed. Same 5–65 dBZ colour scale as the Sarkoa hurricane radar countdown. Category and landfall details: https://www.nhc.noaa.gov/data/tcr/AL062018_Florence.pdf . NHC wind scale: https://www.nhc.noaa.gov/aboutsshws.php . Radar archive: https://registry.opendata.aws/noaa-nexrad/ .
WGS 84 · Updated 20 Oct 2026 · 1 view
- showcase
Hurricane Michael: Category 5 radar
Hurricane Michael near its Category 5 landfall on 10 October 2018. This is the NOAA KEVX NEXRAD lowest-sweep reflectivity volume beginning 17:27:43 UTC, regridded to geographic coordinates with nearest native radar gates. Michael's Category 5, 140 kt landfall estimate comes from the National Hurricane Center's final reanalysis; the radar colours show precipitation structure in dBZ, not wind speed. Colour scale: 5–65 dBZ. Source volume: https://unidata-nexrad-level2.s3.amazonaws.com/2018/10/10/KEVX/KEVX20181010_172743_V06 . NHC final report: https://www.nhc.noaa.gov/data/tcr/AL142018_Michael.pdf . NHC wind scale: https://www.nhc.noaa.gov/aboutsshws.php .
WGS 84 · Updated 20 Oct 2026 · 3 views
- showcase
Edmund Fitzgerald: Whitefish Bay comparison (working copy)
Working copy of the published Edmund Fitzgerald–Whitefish Bay geographic comparison. The only visible overlay is the sourced straight-line comparison between the NTSB wreck coordinate and NOAA Whitefish Point Light; it is not a final ship track. A separate NOAA study's approximate route tracing is being checked and is not shown on this map.
WGS 84 · Updated 19 Oct 2026
- showcase
Edmund Fitzgerald: the last 17 miles to Whitefish Bay
The Edmund Fitzgerald's reported wreck position sits about 17 miles from the entrance to Whitefish Bay, according to the NTSB. This map compares the NTSB position (46°59.9′ N, 85°06.6′ W) with Whitefish Point Light on the west side of the bay entrance (46°46′16″ N, 84°57′27″ W, NOAA Coast Pilot 6, 2026). The WGS84 straight-line distance between those published coordinates is 27.806 km / 17.278 statute miles; NTSB rounds the entrance distance to approximately 17 miles. The gold line is a geographic comparison, NOT the ship's route, the distance it could have sailed safely, or a surveyed final track. There were no survivors or witnesses, and no distress call was heard. Bathymetry and historic wreck data in the underlying map remain credited to NOAA NCEI, NOAA Office of Coast Survey, USCG and NTSB; land elevation is Copernicus DEM with its original credit. Sources: https://www.ntsb.gov/investigations/AccidentReports/Reports/MAR7803.pdf ; https://www.nauticalcharts.noaa.gov/publications/coast-pilot/files/cp6/CPB6_WEB.pdf .
WGS 84 · Updated 19 Oct 2026 · 3 views
- storms
Typhoon Champi: real radar of the eye passing Guam
Archived PGUA weather radar scans show Typhoon Champi’s eye move west just south of Guam on 18 October 2026 (ChST). The 28 observed frames, 03:02–06:01 ChST, are lowest-sweep radar reflectivity (dBZ), not measured wind speed. NOAA NEXRAD Level II data came from the public Unidata AWS archive; the coloured raster is a nearest-gate georeferenced visualization, with values below 5 dBZ hidden. Blank radar sectors do not prove clear weather. The white outline is U.S. Census Bureau January 2026 US Landmass. NWS Guam later reported the eye about 10 miles south of Guam’s southern tip around 04:26 ChST and occasional airport gusts around 80 mph; these are NWS report facts, not measurements extracted from the image. Parts of Guam lost power after 05:00; the island did not suffer a complete blackout. Sources: https://registry.opendata.aws/noaa-nexrad/ ; https://forecast.weather.gov/product.php?format=CI&glossary=1&issuedby=PQ2&product=TCU&site=GUM&version=8 ; https://forecast.weather.gov/product.php?format=CI&glossary=1&issuedby=PQ2&product=TCU&site=GUM&version=10 . Historical event, not a live forecast.
WGS 84 · Updated 19 Oct 2026 · 5 views
- showcase
Where Meteorites Are Found | 32,186 Mapped Records
This map shows 32,186 geocoded meteorite records from the NASA Open Data / Meteoritical Society catalogue. Of these mapped records, 22,099 lie south of 60°S. The points are finds, not a measure of where meteorites fall more often. Antarctic blue ice can transport, concentrate, preserve and expose meteorites; dark rocks also stand out against ice. This companion map is the source geometry for Sarkoa's October 2026 Instagram Reel. NASA data: https://data.nasa.gov/dataset/meteorite-landings . Explanation of Antarctic concentration: https://science.nasa.gov/earth/earth-observatory/finding-meteorite-hotspots-in-antarctica-149554/ . The source catalogue does not represent all meteorites that have ever landed.
WGS 84 · Updated 19 Oct 2026 · 2 views
- geography
Where Would You Emerge? City Antipodes
Three city centres and the exact opposite points on a spherical Earth. Orange marks the Natural Earth city centres; blue marks where a straight line through Earth's centre would emerge. New York and Sydney emerge in ocean; Beijing emerges on land in Argentina. Sources: Natural Earth 50m Populated Places v5.1.2 and 50m Land v4.1.0.
WGS 84 · Updated 19 Oct 2026 · 1 view
- showcase
The Meteorite That Hit Ann Hodges | Sylacauga 1954
On 30 November 1954, Ann Hodges was napping in Sylacauga, Alabama when a roughly 3.9 kg meteorite came through her roof, bounced off a radio and struck her. She survived with a bruise. The map opens at Sarkoa's Sylacauga story marker, a town locator rather than a verified house coordinate or recorded flight path. Other markers are curated meteorite stories; this companion map is focused on the Hodges event. Sources for the Hodges account: Smithsonian Institution (In 1954, an Extraterrestrial Bruiser Shocked This Alabama Woman) and NASA Technical Reports Server (Meteorites hitting humans, 2021).
WGS 84 · Updated 19 Oct 2026 · 1 view
- showcase
The Modern Pirate Map — 46 Years of Attacks at Sea
Pirates never went away. This is every recorded act of piracy and armed hostility against ships from May 1978 to June 2024 — 9,131 incidents from the U.S. National Geospatial-Intelligence Agency's Anti-Shipping Activity Messages (ASAM) archive. Press play on the timeline and watch modern piracy move around the planet: harbour theft across Southeast Asia in the 1990s, the Strait of Malacca years, the Somali explosion of 2008–2012 reaching a thousand miles offshore, kidnap-for-ransom in the Gulf of Guinea, and the missile-and-drone era that opened in the Red Sea in late 2023. Colour marks what happened: warm colours are violent — hijackings, kidnappings, missile and drone strikes, ships fired upon. Cool colours are boardings and theft. Click any dot to read the official incident report; many read like thriller paperbacks (sonic cannons, ransoms parachuted onto decks, machetes, tug-of-war rescues). The white gold-ringed markers are the Infamous Thirteen — incidents that made world news, from the Achille Lauro (1985) to Captain Phillips' Maersk Alabama (2009) to the Galaxy Leader helicopter hijacking that started the Red Sea crisis (2023). Positions are as reported by mariners; some historical positions are approximate. Source: NGA ASAM (public domain), via the Esri Living Atlas archive mirror. Not for navigation.
WGS 84 · Updated 19 Oct 2026 · 25 views
- storms
Champi Near-Miss at Guam | NWS Reported Centers
Super Typhoon Champi passed close to Guam on 18 October 2026. The six dots are NWS Guam reported center positions from 17 October 13:00 to 18 October 13:00 ChST. Straight connecting segments are visual interpolation between advisories, not a surveyed best track. NWS Guam reports the eye passed about 10 miles south of Guam's southern tip at approximately 04:26 ChST on 18 October. The coastline is the U.S. Census Bureau January 2026 US Landmass outline. This map is historical and is not a live forecast.
WGS 84 · Updated 19 Oct 2026 · 4 views
Tornado Tracks | Explore the Record
Find tornado paths in the hosted Tornado Archive dataset. The companion short follows U.S. NOAA/NWS Storm Prediction Center paths from 1950 through 2025 and compares the centre of (E)F1-and-stronger paths in the 1950s with the 2020s. This map contains the wider archive, including Canadian and other historical records; it is a static spatial catalogue, not the video's year-by-year playback. Weak tornado reporting changed markedly over time, so do not infer a simple trend from all recorded paths. Source: Tornado Archive, with U.S. paths from NOAA/NWS Storm Prediction Center SVRGIS and Canadian tracks from Environment and Climate Change Canada. Positions and classifications inherit source limitations.
WGS 84 · Updated 19 Oct 2026 · 1 view
- Change detection
Earth at Night: Three Decades of Change
Thirty annual snapshots of Earth's artificial light at night, 1992-2024, with a change layer and twelve highlighted regions. HOW TO READ IT. Press play, or drag the time scrubber, to watch lights spread. Colour runs dark navy (unlit) through violet and orange to warm white (city cores); the same brightness scale is baked into every year, so changes on screen are changes in the data, not in the styling. Turn on 'Where lights changed' for the 1992-96 vs 2020-24 difference: orange gained light, blue lost it. METHOD AND LIMITS. The source splices calibrated DMSP (1992-2013) with simulated VIIRS (2014-2024) on a shared 0-63 brightness scale at ~1 km. Three choices were made to keep the animation honest: 2014, 2015 and 2016 are omitted, because they lose about 27% of their lit pixels versus BOTH 2013 and 2017 - a detection-floor artifact, not a real dimming; pixels below DN 6 are dropped, because only the DMSP half has data there and keeping it would fake a decline at the splice; and every regional statistic uses a DN>=16 floor, where the 2013-2017 instrument seam is -1.0% globally instead of -16.5%. Because of that seam, each highlighted region is described using WITHIN-ERA change (1992-2013, and 2017-2024) rather than a single 1992-2024 figure. Apparent dimming in wealthy, dense regions is the least trustworthy signal on the map: those cities saturated the old DMSP sensor, and neither sensor sees blue-rich LED light well. The Japan polygon is included specifically to show that trap.
WGS 84 · Updated 19 Oct 2026 · 6 views
- showcase
Rocks From Space — Every Meteorite Known to Humanity
Earth has been hit before. It is still being hit. This map is the record — read it in five acts. 1 · THE SCARS. The red rings are 87 confirmed impact structures drawn at their TRUE size, coloured bright (young) to dark (ancient). Find Chicxulub, the 150 km dinosaur killer under the Yucatán, and Vredefort in South Africa — a two-billion-year-old wound 160 km across. 2 · THE SURVIVORS. The glowing dots are meteorites that made it to the ground and were found — NASA and the Meteoritical Society's full catalogue. To keep things clean, only rocks over ~10 kg are shown at first: drag the slider to zero to bury the map in all 32,186, down to gram-sized gravel. The 60-tonne Hoba iron in Namibia has still never been moved. 3 · THE WITNESSES. Only 1,096 meteorites in all of history were actually SEEN falling — the gold dots. Press ▶ on the timeline and watch the eyewitness record unfold, from a stone that fell on Japan in the year 860 to today. 4 · THE PRESENT. The cyan flashes are fireballs — asteroids exploding in the atmosphere, logged by US government early-warning sensors since 1988. Chelyabinsk (2013) hit with 441 kilotons and injured ~1,500 people. It is still happening: the latest event here is from this month. 5 · THE CHARACTERS. The white gold-ringed markers are thirteen true stories — click them. The woman a meteorite bruised in her sleep. The $300 Chevy a rock made priceless. The 'thunderstone' an emperor chained up in a church in 1492. The missing crater of Tunguska. And one secret: switch on 'Pieces of the Moon & Mars' — 146 of the catalogued rocks aren't asteroid fragments at all, but debris blasted off other worlds and delivered to Earth for free. The empty spaces tell their own story: meteorites fall evenly across the planet, but they are only FOUND where people look — deserts, farmland, Antarctic ice. Sources: NASA Open Data — Meteorite Landings (Meteoritical Society); NASA/JPL CNEOS Fireball Database; Wikipedia — List of impact structures on Earth (CC BY-SA). All data free.
WGS 84 · Updated 18 Oct 2026 · 13 views
Water Quality Now — North America
Current water quality status from every open real-time and near-real-time source we could unify: continuous USGS river and lake sensors across the United States, beach bacterial sampling in Toronto, Ottawa, York Region and Manitoba, and active long-term drinking water advisories — First Nations systems drawn as the reserve areas they affect, and Manitoba's municipal, cooperative, campground and school systems as points. Each location is graded good / fair / poor / unsafe against the published guideline for its own measurement — Health Canada recreational water, CCME aquatic life, or the issuing health unit's own posting. Beaches are sampled weekly to daily and advisories run for years, so every record carries both when it was measured and when this pipeline last confirmed it still stands, plus the rainfall since — the one thing about a beach that is current to the hour. Every popup links back to the issuing agency. Rebuilt hourly.
Updated 14 Oct 2026 · 13 views
Global Lightning Strikes — Last 24 Hours
Every GOES GLM lightning flash detected in the trailing 24 hours, coloured by age. Rebuilt hourly from NOAA GOES-19 (East) and GOES-18 (West); strikes older than 24 h drop off automatically.
Updated 14 Oct 2026 · 5 views
Three rivers meet at Ottawa
Three rivers meet in the middle of Canada's capital, and they are nothing like the same size. The Ottawa River arrives from the west carrying about 1,237 cubic metres a second. The Gatineau comes down from the Quebec side and adds about 354. The Rideau, the river that names the canal, the falls, the street and the hall, arrives over Rideau Falls with about 40, which is roughly two per cent of the three. Each line follows one path through the water and is coloured by which river that water came from: amber for the Ottawa, green for the Gatineau, pink for the Rideau. Downstream of the confluences the model keeps them in lanes, with Gatineau water along the Quebec bank and Rideau water along the Ontario bank. What the lines are, and are not: an idealised model drawn from the shape of the channels and each river's average flow (potential flow on a 10 m grid). It has no depth, dams, wind, tides or mixing, so it is not a measured current and the speeds are relative only. The lanes downstream are a property of the model, not evidence about how far the real rivers run before they mix. Mean flows are the mean of every monthly mean ECCC holds for each gauge, so each covers its own period: Ottawa River at Britannia 1,237 m3/s (772 months, 1960-2024), Gatineau at the Rapides Farmers generating station 354 m3/s (171 months, 1980-1994), Rideau River at Ottawa 40 m3/s (1,044 months, 1933-2025). Sources and licences: water from ESA WorldCover 2021 (CC BY 4.0); ground picture from the WorldCover Sentinel-2 2021 composite (contains modified Copernicus Sentinel data, processed by ESA WorldCover consortium); terrain from the Copernicus DEM GLO-30 (ESA); flows from ECCC HYDAT (Open Government Licence - Canada). Click a line for its water and relative speed.
WGS 84 · Updated 13 Oct 2026
- environment
Aurora Tonight: where the aurora is overhead now
Where the aurora is likely to be overhead right now, from NOAA's OVATION model, drawn over the night lights so you can see where it is dark enough to look. Drag the time slider to step through the last 24 hours. Reading it: the green glow is the chance of aurora directly overhead at the time on the slider, from a faint veil at 5 % to white-green near 90 %. The outlines mark 10, 30, 50 and 70 %; click one for its times. Each frame is the Space Weather Prediction Center's nowcast, valid 30 to 90 minutes after the solar wind was measured. It is not a forecast for later tonight. Aurora low on the northern horizon can be seen several hundred kilometres south of the green when the sky is clear and dark. Sources and licences. Aurora probability: NOAA Space Weather Prediction Center OVATION aurora nowcast (public domain, credit NOAA SWPC). The 1 degree grid is lightly smoothed and resampled for display, so these frames are derived from NOAA data and are not the original product; the stripe of low values SWPC's grid carries along the equator is removed as a grid artefact. Kp: NOAA SWPC (public domain). Night lights: NASA Black Marble VNP46A4 annual composite for 2025 (NASA, no restrictions), log-scaled, with radiance under 1.5 nW/cm2/sr hidden to suppress airglow. Dark sky layer (off by default): Sarkoa Dark Sky Atlas 2025, Bortle class modelled from Black Marble and calibrated on Globe at Night readings (CC BY 4.0). Caveats. Clouds and moonlight decide what you actually see and neither is on this map. OVATION is driven by solar wind measured about an hour upstream of Earth, so sudden changes can arrive before the map shows them. The number is a chance of aurora overhead, not how bright or colourful it will be, and cameras pick up aurora the eye misses. Frames older than 24 hours are removed as new ones arrive, and the feed only adds frames while it is running. The dark-sky layer is a model: read it as the glow around towns, not as a measurement at a single site.
WGS 84 · Updated 12 Oct 2026 · 3 views
- environment
The Great Lakes, drained
All five Great Lakes with the water taken away: the lakebeds at their true height against the shore, and the wrecks that lie on them. Sources and licences. Lakebeds: NOAA NCEI Great Lakes Bathymetry, 3 arc-second grids (about 90 m), public domain. Their depths are below each lake's Low Water Datum, so each lake was shifted by its own datum (IGLD 1985, NOAA CO-OPS): Superior 183.2 m, Michigan and Huron 176.0 m, St. Clair 174.4 m, Erie 173.5 m, Ontario 74.2 m. Land: Copernicus DEM GLO-90, © DLR e.V. 2010-2014 and © Airbus Defence and Space GmbH 2014-2018 provided under COPERNICUS by the European Union and ESA; all rights reserved. Charted wrecks: NOAA Office of Coast Survey, from today's US electronic navigational charts (ENC Direct) and the retired AWOIS wreck list (last edition, February 2024, read from the Internet Archive), public domain. Famous wrecks: positions and depths from the sources named in each popup (US Coast Guard and NTSB reports for the Edmund Fitzgerald, Michigan Underwater Preserves, Ohio Sea Grant, Ontario Regulation 11/06). Caveats. The bathymetry was compiled from soundings collected over more than a century and contoured by hand in the 1990s; Lake Superior's grid is a draft NOAA never finished, so small features there are smoothed (the grid has 150 m of water over the Fitzgerald, which lies in 530 ft, 161.5 m). Chart wrecks cover US charts only, and most are in harbours and shallow water; wrecks in Canadian waters appear only where added by hand. Positions of old wrecks are approximate, and many charted wrecks have no name. The relief layer is shaded with 6x vertical exaggeration; the elevation layer holds plain metres (300 m Web Mercator cells, about 200 m on the ground). Copernicus' flattened lake surfaces sit 0.7 to 3.5 m below the datums, which is invisible at this scale. The deepest point, in Lake Superior, is 405 m below the lake's datum: 222 m below sea level. Lake Ontario's floor reaches 170 m below sea level, Michigan's 100 m. Of the five, Lake Erie, 64 m at its deepest, would empty first. Built by Sarkoa as part of the social maps series. Not for navigation.
WGS 84 · Updated 12 Oct 2026 · 8 views
North American Fall Colour Timing
The slider opens on today. Pick any date from August 15 to November 30 and each area shows how that day compares with its usual peak fall colour. Green runs from about five days before the usual peak to six days after it, since when colour looks its best is partly a matter of taste. Teal is a little early and blue means the leaves are still mostly green. Yellow is past peak with colour still on the trees, and red means most of it is gone. The five colours were picked to stay distinct for people with colour blindness. Tap an area for its full schedule. The timing comes from ten years of NASA MODIS land surface phenology (MCD12Q2, 2016 to 2025) across Canada and the United States. Only deciduous forest counts. Satellite land cover at 500 m misses much of it where hardwoods grow among pine or spruce, so each area's share of broadleaf and mixed forest also comes from 30 m maps by the Commission for Environmental Cooperation, Natural Resources Canada and the USGS. That is what brings in the pine and hardwood forests of the Southeast, the woods along rivers in farm country, the aspen of the Rockies and the birch and poplar of the southern boreal. For each 5 km cell we take the date the satellite sees the canopy nearly bare, as a median over the decade, count back about 25 days, and add a day for every four degrees further south. That rule was picked by testing it against Harvard Forest's weekly survey of tagged trees and 45 places with a published typical peak, from Fairbanks and Prince Albert to Maroon Bells, the Ozarks, Seattle and Nevada City. On places held out of the fit it lands about 5 days from the published date on average. Algonquin's 50-year record puts the typical sugar maple peak on September 27. This map says September 28. It shows a typical year, not a forecast, which is why it works from one autumn to the next without being redrawn. Weather shifts the peak: at Harvard Forest half of the years since 1991 peaked within 3 days of the usual date, while the earliest came 9 days early and the latest 16 days late. Colour fades over about two weeks after peak as later species turn, so yellow lasts longer than teal. Mountains colour from the top down over a few weeks and each cell gets a single date, so ridgelines turn before the map says and valleys after. Eastern Newfoundland comes out later than the province's usual September and October season, so treat its dates as rough. Blank land has too little deciduous forest to rate. Across the prairies and most of the West, that is simply what grows there. North of about 50 degrees in Ontario and Quebec the forest is almost all black spruce and jack pine, much of it on bog. Most of California's oak woodland is left out because it browns through the dry summer, which a satellite cannot tell apart from autumn. Banff and the main ranges of the Canadian Rockies stay blank too, since most of their gold is alpine larch, a conifer that drops its needles. No continent-wide land cover map tells larch apart from spruce and fir, and the same goes for the tamarack that turns the northern bogs yellow.
WGS 84 · Updated 11 Oct 2026 · 126 views
Sunset Clock: sunset by the minute, in local clock time, 2026-27
Sunset in local clock time, on land everywhere on the globe, as one band per minute: every place inside a band sees the sun set at the same minute on the clock. Press the locate button, then click the band you are in for that day's sunset time and clock. Move the time slider to change the week (Sundays, 4 October 2026 to 3 October 2027). The bands break where clocks jump, at time zone borders, and the white lines mark borders where the clocks on the two sides differ that week. Click a city for its solstice times, earliest sunset and latest sunrise. Why now: British Columbia, Alberta, the Northwest Territories and Manitoba are not falling back on 1 November 2026, while the United States and the rest of Canada are. From 1 November to 14 March the 49th parallel is a one-hour step in clock time. On 21 December the sun sets at 5:16 pm in Vancouver and 4:20 pm in Seattle. Europe falls back on 25 October; Russia, Belarus, Turkey, Iceland and Morocco do not change their clocks. Sources and licences: time zone boundaries from timezone-boundary-builder release 2026d (data © OpenStreetMap contributors, ODbL 1.0). The sunset bands and the clock-border lines are databases derived from it and are offered under the same ODbL 1.0 licence. Clock rules from the IANA time zone database 2026e (public domain). Cities from GeoNames cities15000 (CC BY 4.0). Solar position from the NOAA Solar Calculator equations (NOAA Global Monitoring Laboratory). Method: for each date and each UTC offset in use that day, sunset is computed on a 0.05° grid with the NOAA solar equations at the standard -0.833° altitude (upper limb of the Sun, mean refraction), refined at the event's own instant. Within one offset local sunset is a smooth field, so it is contoured at every minute (a band holds sunsets that round to its minute) and clipped to the zones keeping that offset, then simplified. Polar night (no sunrise) and the midnight sun (no sunset) are their own areas. Points sampled inside the bands agree with a direct calculation to within a minute; the solar equations agree with the US Naval Observatory to within a minute for 15 places on 5 dates. Caveats: times are geometric with standard refraction. Terrain, elevation and weather are ignored, so a valley or a mountain horizon loses the Sun earlier. The bands cover land plus a margin of up to about 10 km offshore, so every coastal town is inside one; the open ocean is left out. Zone borders are simplified to about 2 km and band edges to about 8 km of a smooth curve (a fraction of a minute), so within a couple of km of a border a place can take the neighbour's clock; city popups use each city's own zone. Western Canada follows the 2026 provincial decisions as recorded in tzdb 2026b to 2026e; tzdb assumes northwestern Ontario follows Manitoba, which was not confirmed when it was released. Xinjiang is drawn on official Beijing time (UTC+8); many residents keep Xinjiang time (UTC+6), two hours earlier. Near the midnight sun a sunset can fall after midnight (shown as next day). Dates are Sundays, so each clock change in the window shows on its own week.
WGS 84 · Updated 11 Oct 2026 · 8 views
- Society
City growth 1975 to 2030 (GHSL population)
How 18 cities grew from 1975, with projections to 2030, from the European Commission's Global Human Settlement Layer (GHSL). Sources and licences: GHS-WUP-POP R2025A (Schiavina et al. 2025, doi:10.2905/adba95af-db56-4569-acd3-9513201eba30) and GHS-POP R2023A (Schiavina et al. 2023, doi:10.2905/2FF68A52-5B5B-4A22-8F40-C41DA8332CFE), European Commission, Joint Research Centre, CC BY 4.0. Reclassified and aggregated by Sarkoa. Caveats: GHSL population is modelled, not counted. Census totals for each administrative area are spread across the built-up area mapped from satellite imagery. 2025 and 2030 are projections, and 1975 is the least reliable year: Luanda is left off because GHSL puts only 50,000 people in its circle in 1975. A circle is not a city. It can take in neighbouring towns (Kinshasa's includes Brazzaville across the river) and miss outer suburbs. What you see: the coloured layer is people per square kilometre on GHSL's 1 km grid, one frame per five-year epoch. Cells under 300 people per km2, GHSL's own density threshold for an urban cluster, are left clear. Each green circle is a city's measuring area: everyone within 35 km of a fixed centre point, the same size everywhere so cities compare fairly. The dots grow with that population; click one for the figure at the selected year. Method: city totals use the newest GHSL population release (R2025A, aligned with the UN World Urbanization Prospects 2025) at 1 km, spread inside each 1 km cell with the 100 m pattern of GHS-POP R2023A. For most cities and years the two releases agree within 5%; Kinshasa (16.2 million in 2030 against 20.7 million in R2023A) and Dar es Salaam are the largest revisions.
WGS 84 · Updated 11 Oct 2026 · 135 views
- Nature
Follow one bird: Leo the turkey vulture
Leo is a turkey vulture that Hawk Mountain Sanctuary's Acopian Center tagged as an adult in central Saskatchewan in June 2007. Its GPS tag logged 85,517 positions over 14 years: 14 autumn migrations to the llanos of Venezuela and 13 trips back north, about 8,000 km each way. Nine other turkey vultures tagged in Saskatchewan during those years are drawn alongside. Press play or drag the time slider: each bird is a short comet of its last three days of fixes (brightest = last 24 hours), over the full route of every bird in its own colour (Leo in gold, drawn wider). The timeline bands mark Leo's autumn migrations. Click a point or a line for details. Source and licence: Bildstein KL, Barber D, Bechard MJ, Graña Grilli M, Therrien J. 2021. Data from: Study "Vultures Acopian Center USA GPS" (2003-2021). Movebank Data Repository. https://doi.org/10.5441/001/1.f3qt46r2. Licensed CC BY 4.0 (the repository record; the package README says CC0, we follow the stricter one). Bird names are the study's own animal IDs. Leo's sex is not recorded. Read this first. GPS fixes are samples, not the flight path: Leo's tags logged hourly, around the clock until 2014 and about 18 fixes a day after that, so distances (summed between consecutive fixes) miss the circling a soaring bird does and are lower than what it really flew. A gap of more than 24 hours between fixes breaks the line rather than joining it: the old tag was failing in 2013 and 2014, so those seasons have holes. Ten tagged birds are a tiny sample of a species that numbers in the millions. The published data stop on 1 June 2021, with Leo still on its wintering grounds in Venezuela after not flying north that spring; the last week of fixes barely moves. The study does not record what happened. Method. Fixes Movebank flagged as outliers were dropped, then a speed filter removed any fix that implied more than 110 km/h to its neighbours (Leo's fastest hourly steps are just under 100 km/h, on 3 October 2014). Migrations are found from daily positions: runs of days with more than 350 km of net movement across a week, merged across stopovers of under three weeks, starting on the last day near the summer or winter range and ending on the first day near the next one. Median full autumn trip: 8,044 km in 47 days. Biggest day of flying (local solar day): 739 km on 2015-04-14. Autumn 2014, the trip in the video: 8,044 km in 41 days. Points with nothing moving (a night at the roost) are thinned to one every six hours; every moving fix is kept. Three Saskatchewan birds tracked only in 2005 and 2006 (Sill, Blizzard, Duck Lake) are left out so the timeline is Leo's years.
WGS 84 · Updated 9 Oct 2026 · 2 views
Earthquake Depths
Every earthquake of magnitude 5 or more recorded worldwide from 1 January 1973 to 5 October 2026: 89,394 events from the USGS Comprehensive Catalog (ComCat). Colour is the depth where the rupture started: shallow 0 to 70 km (orange), intermediate 70 to 300 km (amber), deep 300 to 700 km (blue). Dots grow with magnitude on a square-root scale, so area tracks magnitude above 5. Drag the timeline handles to pick a window of years; the map opens on the whole period. Click a dot for its place, magnitude, depth, date and the USGS event page. SOURCE AND LICENCE. U.S. Geological Survey, Earthquake Hazards Program, ComCat FDSN event service (earthquake.usgs.gov/fdsnws/event/1), public domain. Fetched one calendar year per request (the service caps a query at 20,000 events) on 6 October 2026. Event type earthquake only: 546 nuclear explosions and 99 other non-earthquake events of M4.5+ (volcanic eruptions, quarry and mine events, landslides) were removed. CAVEATS. The catalogue is not uniform over time. Seismic networks improved, so later decades record more events: M4.5+ averaged about 3,300 a year in 1973 to 1982 and 7,400 a year in 2016 to 2025, and part of that rise is better detection, not more earthquakes. Magnitudes mix types (mb, Ms, Mw and variants); 47% of these M5+ events carry a body-wave mb, which saturates for large quakes, so dot sizes compare only roughly. Depths of older and remote events are less certain: 24% sit at exactly 10 km and 16% at exactly 33 km, default depths assigned when the depth could not be resolved. Locations are rounded to 0.001 degrees. WHAT TO LOOK FOR. 70,962 of the events are shallow, 14,192 intermediate and 4,240 deep. Deep earthquakes happen almost only where one plate sinks under another: follow the colours inland from the trenches off Tonga, Japan, the Kurils, the Andes and Indonesia and they turn from orange to amber to blue as the sinking slab goes deeper. Around Tonga, Fiji and the Kermadec Islands lie 71% of all deep M4.5+ earthquakes in the catalogue. WHY M5 AND UP. The full M4.5+ catalogue holds 293,154 events, too dense to serve as one lossless tile set at global zoom, so the hosted layer starts at M5. The 3D clips of individual subduction zones use the full M4.5+ set.
WGS 84 · Updated 9 Oct 2026 · 2 views
- Climate
Arctic refreeze 2026: daily sea ice vs every September since 1979
Arctic sea ice is freezing back after the September minimum. The time slider steps through NSIDC's daily sea-ice concentration from the 2026 minimum (September 12, 2026) to the latest day published (October 6, 2026). The dashed orange line is the typical September ice edge for 1981-2010; the year slider shows the September ice edge of any year since 1979 (pink). Source: NSIDC Sea Ice Index, Version 4 (G02135), daily 25 km concentration and monthly September extent outlines. Credit: Sea Ice Index, National Snow and Ice Data Center. Citation: Fetterer et al. (2025), doi:10.7265/a98x-0f50. NSIDC data are free to use with this citation. Every extent figure here is NSIDC's own published value, not re-derived. Caveats: the grid is 25 km, and cells under 15 % concentration are not drawn (the standard extent threshold). The input sensor changed to AMSR2 on 1 January 2025, so 2025 and 2026 come from a different instrument than 1979-2024 (SMMR, SSM/I, SSMIS). The satellites never see the pole itself: that hole is counted as ice, as NSIDC does, and this web map stops at 85 N in any case. Coast cells take the value of their nearest ocean cell; land is left blank. The record has gaps: one value every other day before August 1987 and no data from 3 December 1987 to 13 January 1988 (no September is affected). A few coastal cells far to the south (James Bay, Sea of Okhotsk) carry ice in the product in early October; they are shown as published. Extent is not volume: it says nothing about how thick the ice is. The September lines are the monthly-average ice edge against open water, lightly smoothed from 25 km cell edges. Numbers: the 2026 minimum was 4.60 million km2 on September 12, 2026 (5-day average; tied with 2008, 2010 and 2025 for 10th lowest since 1979, per NSIDC and NASA). The September 2026 average was 4.81 million km2, 13th lowest of 48 Septembers, against 7.05 in 1979 and the 2012 record of 3.57. By October 6, 2026 the ice had grown to 5.96 million km2, still 1.44 million km2 below the 1981-2010 average for that date.
WGS 84 · Updated 9 Oct 2026 · 5 views
- Society
Festival lights from space
Satellites measure how much light leaves the ground every night. This map asks a narrower question: when a city keeps a festival, do the streets where people live get brighter than the streets where they work? You are looking at Dallas, the largest effect measured here. The other 15 cities are one zoom out, or in the search bar. Prove it yourself in two clicks. The Land use control switches the coloured blocks between homes and the industrial ground they are measured against. Warm blocks are the ones that got brighter. Switch to industry and the warmth largely goes, in the same city on the same nights. Drag the time scrubber. The bottom layer is the measurement itself: VIIRS radiance on the nights either side of each city's festival, on one fixed brightness scale so the frames stay comparable. Each city flips on its own calendar, which is why Ramadan slides eleven days earlier each year. Each city was only measured around its own festival, so it holds its most recent composite until the next one: at a December date Chicago shows December while Delhi still shows Diwali. The scrubber gives the last thing measured in each place, not one night worldwide. Homes are compared against industrial estates, ports and airports in the same city, over the festival nights against the nearest stretch of ordinary nights its calendar leaves alone. Ground lit for work is lit all year and decorated by nobody, so weather, snow and night length hit both groups and cancel. What is left only happens where people live. In plain units, before any statistics, Dallas, homes 35 to 42 nW (+22%), industry 75 to 74 nW (-1%); Chicago, homes 26 to 33 nW (+27%), industry 56 to 61 nW (+8%). In Dallas, 73% of residential blocks were more than a tenth brighter during the festival, against 19% of industrial ones. That gap is the finding; the rest is how hard it was tested. 8 of 15 cities show a festival effect that stands out from what the same measurement does in that city on ordinary nights, against thresholds written down before the data was looked at. The rest are on the map with their numbers and the reason they fall short: a festival that does not show up is a result too. Atlanta +9%, Mexico City +7%, Riyadh +6%, Shanghai +4% clear their own noise but come in under 10%: small effects measured well, not doubtful ones. That line is about what is worth telling, not about the measurement. 6 more sit inside the range this method produces there when nothing is happening, so nothing is claimed for them. Every city is measured the same way on ordinary nights with no festival in them, and that is what an effect has to stand out from. A fixed bound instead passed Cairo on an average placebo under 2% while its own placebo windows ran +16% to -18%. Those placebos average -1.7%, leaning against the festival effect rather than with it. Read Mexico City (41 points against an effect of +7%), Riyadh (14 points against an effect of +6%), Shanghai (11 points against an effect of +4%) with extra care: they clear the bar, but the spread of their ordinary nights is wide enough that the interval flatters the headline. Every window and every year is in the popups. Festival dates are resolved per city per year, and the comparison windows are searched to avoid every festival that city keeps, not just the one being measured. 3 festival frames are left out of the animation: cloud hid more than half the city, so they would have shown the lights going out. No estimate uses them either way. A pixel only counts when both windows saw it, so a cloudy fortnight cannot substitute one part of a city for another. VIIRS is nearly blind to blue light, so white and blue LED decorations are under-counted: every number here is a floor. Tokyo is the negative control: the same Christmas windows in a city that does not decorate its homes for it. It reads +1.0%. Source: NASA Black Marble VNP46A2 (VIIRS day-night band), Collection 2. Land use from OpenStreetMap (ODbL).
WGS 84 · Updated 7 Oct 2026 · 9 views
The Cryptid Atlas
Where the monsters are supposed to live. 240 cryptids, each drawn as the extent its own sources actually describe. Where a tradition is genuinely continental it is drawn that way — the Thunderbird is central to Indigenous traditions from the Northwest Coast through the Plains and Great Lakes to the Northeast, so it spans the continent. Where a tradition is one pool, it gets one pool: Nessie gets 37 km of loch and not a metre more, and the Hairy Hands of Dartmoor get a single stretch of the B3212. Every shape is classified and coloured by what is claimed: lake and sea monsters (cyan, 56), reptilian and saurian animals (green, 41), beasts and predators (rose, 37), humanoids and folk spirits (fuchsia, 36), wild hominids (amber, 30), the anomalous and unclassifiable (pale, 22), and winged cryptids (violet, 18). Use the 'Cryptid type' dropdown at bottom-left to show one class at a time — with 240 overlapping translucent extents the colours blend, and filtering to a single type is the way to actually read the distribution. Wild hominids alone make a striking map: they cluster on mountain forest worldwide. Click any shape for the full record — the locus the tradition is anchored to and the incident that fixed it there, an 'extent basis' line stating exactly what the polygon was traced to, and a source you can click through to. Illustrations, and how to read them. All 240 entries carry a picture, from two very different places. 131 are real: public-domain or Creative Commons images from Wikimedia Commons — the 1937 Naden Harbour carcass photograph, the mušḫuššu relief from the Ishtar Gate, a Māori taniwha carving on Lake Taupō, a blemmye from the 1513 Piri Reis map. The other 109 are AI-generated paintings, made for this map from a written brief describing what the creature is claimed to look like, the real place it is reported from, and the moment of the encounter. Every one says AI-GENERATED in capitals in its credit line, and they share a single painted look so a generated picture is never mistakeable for an archival one. Nothing generated is offered as evidence. Earlier image passes were rejected rather than shipped: a York shop called Barghest, a Sochi roller coaster called Zmey Gorynych, the São Paulo flyover nicknamed Minhocão and a photograph of a musician trading as Hafgufa — things that share a cryptid's name and nothing else. The honest ones are the best ones. The Hodag was confessed a hoax in 1902 and Rhinelander put it on the high school badge anyway. The Jackalope was built by two taxidermist brothers in 1932 and is Wyoming's official state mythological creature; the Skvader and the Wolpertinger are its Swedish and Bavarian cousins, and the Dahu and the Elwetritsch exist mainly so locals can send visitors out at night with a sack. Arkansas passed a law in 1973 making it illegal to molest the White River Monster. Sweden protected the Storsjö monster as an endangered species in 1986. Argentina's president is godfather to every seventh son, because seventh sons were abandoned as werewolves. Some were solved, and the solutions are better than the stories. The dragon skull of Klagenfurt is a woolly rhinoceros, and you can go and look at it. The Bloop was an iceberg cracking off Antarctica. Trunko was rotted whale blubber and the Montauk Monster was a raccoon. The Feejee Mermaid was a monkey sewn to a fish — a Japanese ningyo relic, sold to Barnum. The 1895 Patagonian sloth skin was real and several thousand years old. The 2018 environmental-DNA survey of Loch Ness found no reptile at all, but a very great deal of eel. And some stopped being cryptids. The Kellas cat was folklore until specimens were examined in 1984 and turned out to be a real wildcat hybrid, now in the National Museums of Scotland. Haast's eagle was a Māori story about a bird that carried off people until the bones were described — a genuine 15 kg raptor that hunted moa and died out within human memory. The Kraken was a bishop's tall tale for a centur
WGS 84 · Updated 4 Sept 2026 · 52 views
Athletes Birth Cities of North American Top 4 Sports
49,863 professional athletes, plotted at the town where each one was born. The timeline scrubs by estimated playing career. NHL, NBA, NFL and MLB, born 1842–2007, across 10,502 towns. Every dot is one player: 20,164 NFL, 16,482 MLB, 8,263 NHL, 5,013 NBA, plus 59 who played in two of the four leagues (gold). ▶ WHO WAS PLAYING? Drag the timeline to any range and the map keeps every player whose career OVERLAPS it — not who happened to be born then. Each player is an interval rather than an instant, so they accumulate across the range instead of blinking past. Set 1920–1925 and 2,563 remain; 1985–1990 gives 11,246; 2000–2005 gives 13,958. ⚠️ HOW MUCH OF THIS IS RECORDED. Wikidata holds real career dates for only a fraction of players, and coverage is wildly uneven: NBA 85%, MLB 18%, NFL 8%, NHL 4%. So 9,056 players (18%) carry a RECORDED span, and the other 40,807 (82%) carry an ESTIMATE derived from their real birth year, assuming a career from age 21 to 35. Every dot says which it is. The popup shows 'span source', and the 'Career dates' dropdown filters to recorded-only so you can see the difference for yourself. Recorded spans are exact — Babe Ruth 1914–1935, Wilt Chamberlain 1959–1973, Kareem Abdul-Jabbar 1969–1989, Michael Jordan 1984–2003, Tom Brady 2000–2022. An estimated span can be several years wrong for any one player; the aggregate shape holds, the individual claim is an inference. 🎂 THE BIRTHDAY BIAS — the strangest thing here, and it uses birth dates, which ARE recorded for everyone. Filter by birth quarter and watch the NHL break. Of all NHL players, 30.0% are born January–March and only 20.5% October–December: a 1.47× edge for being the oldest kid in your age group. It did not always exist: NHL players born 1900–1950 → Q1 24.8%, Q4 25.9% (no effect at all) NHL players born 1950–1980 → Q1 30.7%, Q4 19.6% (1.57×) NHL players born 1980–2010 → Q1 32.2%, Q4 18.0% (1.80×) The bias is absent in the pre-war game and grows as organised youth hockey with a January 1 age cut-off takes over. Basketball and football are nearly flat (1.07× and 1.04×). Baseball runs the OPPOSITE way — Q3 peaks at 26.9% — matching its own historical July 31 cut-off. Four leagues, four cut-off regimes, all written in birthdays. ⭐ LEGENDS are labelled on the map so you can actually find someone, ranked by how many language Wikipedias carry them. Michael Jordan tops it at 138. WHERE IS MICHAEL JORDAN? Brooklyn, born 1963, orange (NBA), career 1984–2003. He is not gold, because 'two-sport' here means appearing in two of these four leagues, and his baseball was Double-A with the Birmingham Barons. There is also a second Michael Jordan: an NFL lineman born in Cincinnati in 1998. 🔥 TALENT FACTORIES: switch on birth-city totals, where circle AREA is the number of players born there. Drag 'Minimum players per city': at 100, 57 places survive out of 10,502; at 300 you are down to 14 — Chicago (803), Los Angeles (708), Toronto (468), Houston (456), Philadelphia (453), Detroit (423), St. Louis (398), Miami (356). Chicago's circle covers about 77× the area of a one-player town, because area — not radius — tracks the count. HONEST LIMITS. Dots are spread on a small spiral around their true city so individuals stay clickable; a single dot is not a street address, and the birth-city layer carries the exact coordinates. The fame ranking surfaces people better known for other work (Terry Crews, Ed O'Neill, O. J. Simpson) because Wikipedia counts all of their fame. 87 players whose Wikidata birthplace is a whole country rather than a town were removed, which is why Roberto Clemente is absent. Coverage is what Wikidata knows, not a complete league register: treat city counts as a floor. Source: Wikidata (CC0), queried August 2026.
WGS 84 · Updated 28 Aug 2026 · 69 views
2050 Climate Projections on Arabica Viability
WGS 84 · Updated 28 Aug 2026 · 69 views
- Analysis
Russian Reserve Equipment Watch
Did the war empty Russia's reserve equipment depots? Free satellite imagery says the yards are emptier — but the decline is continuous and predates the February 2022 invasion, not a war-date collapse. Occupancy of 21 Russian long-term vehicle storage bases, 2019-2026, from Sentinel-2 with Sentinel-1 corroboration, each season shown relative to the site's own 2021 state (control-adjusted against 22 displaced placebo polygons with no storage yard). The bases are measurably emptier than pre-war and than the controls (Hodges-Lehmann shift -0.10 [-0.19, -0.02]), but a trend-adjusted test finds no step at the invasion (p=0.86), so the invasion line is a time reference, not a cause. This is a compressed proxy for how full a yard is, NOT a vehicle count (an index-to-count mapping failed against published hand counts). Drag the scrubber to watch the progressive drawdown.
WGS 84 · Updated 21 Aug 2026 · 13 views
- showcase
The Sunken Treasure Map
TOP 10 TREASURES STILL ON THE SEAFLOOR — ranked by estimated value. Roughly $27 billion, none of it recovered. #1 San José (1708) — $17 billion · #2 Flor de la Mar (1511) — $2.6 billion · #3 Notre Dame de Deliverance (1755) — $2–3 billion · #4 Merchant Royal (1641) — $1.5 billion · #5 Las Cinco Chagas (1594) — over $1 billion · #6 HMS Victory (1744) — £500–800 million · #7 Nuestra Señora del Juncal (1631) · #8 HMS Sussex (1694) · #9 RMS Republic (1909) · #10 I-52 (1944). WHY THE SEARCH AREAS ARE SMEARS, NOT CIRCLES. Six of these ten have been found and never touched — those are precise dots. The other four have never been located, and they are drawn the shape the mathematics actually makes. Before the marine chronometer, a navigator could fix latitude from the noon sun to within a few miles. There was no way to measure longitude at sea at all — you guessed it from your speed and a clock that did not work on a moving deck. So every pre-1770 wreck position is sharp north-to-south and nearly worthless east-to-west. All four of the never-found wrecks here sank between 1511 and 1641. Their search areas run 150 to 400 km east-west and only 30 to 80 km north-south. The Merchant Royal's smear covers 3,534 km² — almost exactly the area of Cornwall. Brighter gold = bigger fortune. Click anything for the value, the story, who would legally own it, and exactly why nobody has brought it up. Filter by rank, by whether it has ever been found, or by the reason it is still down there. More layers in the panel: 15 more famous treasures including the ones that DID pay out (the Atocha, the Whydah, $500M of Gairsoppa silver); 37,399 charted wrecks from nautical charts and NOAA surveys — 16,697 of which have no name attached; 1,518 dated wrecks you can play through seven centuries; the 14 deadliest sinkings in history; and 11 legendary graveyards of the sea. Sources: Wikidata (CC0), OpenStreetMap (© OpenStreetMap contributors, ODbL), NOAA AWOIS surveys (public domain), published historical reporting. Values are widely reported estimates that vary between sources — treat them as legend-adjacent, not appraisals. Search smears are indicative uncertainty envelopes modelled from period navigation limits, not official search boxes. Ownership notes summarise reported claims and are not legal advice. Not for navigation.
WGS 84 · Updated 20 Aug 2026 · 11 views
Lebreton Flats
WGS 84 / Pseudo-Mercator · Updated 17 Aug 2026 · 19 views
Tornado Tracks
WGS 84 · Updated 30 Jul 2026 · 13 views
Tornado Tracks
WGS 84 · Updated 16 Jul 2026 · 4 views
Mt Fuji Trail
WGS 84 · Updated 14 Jul 2026 · 138 views
UTM Grid
WGS 84 · Updated 20 Jun 2026 · 8 views