How this was computed

Contact times and sun positions from Skyfield against the DE440s ephemeris, solving where the apparent Sun–Moon separation equals the sum and difference of their apparent radii. The umbral centreline is computed by intersecting the Sun–Moon shadow axis with the WGS84 ellipsoid. Regional screening uses the 1-arcsec AWS terrarium DEM, ray-traced from every grid cell along 62 azimuths out to 40 km with Earth curvature and standard refraction. Every shortlisted site is then re-traced individually against the IGN 5 m LiDAR DTM over the inner 2.5 km, with the Meta/WRI 1 m canopy-height model applied per azimuth on top. Land cover from ESA WorldCover 10 m; roads, buildings and peak names from OpenStreetMap. Every featured site carries an IGN PNOA orthophoto (25-50 cm national aerial survey) with the exact coordinates crosshaired, and every row links out to Google satellite, Bing aerial and the PNOA image itself, plus Windy, Yr and Ventusky for that exact spot. Street View is listed last because it frequently has no coverage on a farm track. Drive-up spots come from scoring points on the road network itself, so parking is guaranteed by construction.

Access is exact point-to-segment distance from the site to the nearest drivable way, roads and farm tracks alike, and the curve is steep: full marks inside 100 m, falling to nothing by 800 m, because carrying chairs and camera gear across open country is not a real option. Tracks are labelled as such so you can tell a lane from a ploughed headland.

Buildings are ray-cast per azimuth from OSM footprints, using the tagged height where there is one and an inferred height by building type otherwise. They are treated like a lone tree: anything within 50 m is ignored, since you step aside, and where a cluster does sit in the way the search looks for a spot within 250 m that clears it. On this path no site's verdict rests on a building — the town-centre pins clear within 60–100 m, and everything else is decided by terrain or trees.

Contact times use the standard two-radius lunar convention, k1 = 0.2725076 for the exterior contacts C1 and C4 and k2 = 0.2722810 for the interior contacts C2 and C3, so totality is not overstated by lunar terrain.

Scoring is one gate then five weighted components out of 100: clearance at totality (28), how far down the sun can be followed toward sunset (20), local relief (16), totality duration (16), access (20). The gate: if the sun is less than 2° above the skyline at totality, the site scores zero regardless of everything else.

• Access ratings are geometric. "Park at the spot" means a drivable way in OSM passes within 40 m — not that there is a passing place, an open gate, or a surface your car will like.
• Many of the tracks come from bulk cartography imports, not survey. In Castilla y León most carry source=ITACyL, were added in a single 2022 edit and have never been revisited, with no surface or access tag. The aerial photo on each card is the check that matters: look for wheel ruts, and for whether the track actually reaches the pin.
• Land access is not checked; some summits will be fenced, private or inside protected areas.
• Canopy and building heights are models. The canopy is a 2020–2024 snapshot that catches isolated trees unreliably, and most building heights are inferred from type rather than tagged. Obstructions closer than 50 m are ignored by design, so a continuous wall or hedge right at the parking spot would not be caught.
• ERA5 cloud figures are a 25 km reanalysis, calibrated here against the airports' own METARs — correlation 0.91, typical error 6 points, but roughly 25 points too optimistic on the Cantabrian coast, where it cannot resolve low stratus. Trust it inland; distrust it within sight of the north coast.
• Weather outranks everything else here. An 8° sun looks through roughly six air masses; haze and distant cloud on the western horizon will matter more than any ranking on this page.
• Slice boundaries are arbitrary — eight equal lengths of centreline. A site near a boundary is equally valid from either side.