Tav ת · Aleph for the Moon

What Tav is

one place in, one folder out

Aleph · Earth
A Python command-line tool. You give it a point or a drawn rectangle and it downloads everything inside: satellite imagery, elevation, 3D mesh, Street View and OpenStreetMap data.
No AI inside
Like Aleph, Tav only collects data. It is the layer that lets a language model “look” at a place.
Tav · Moon
Aleph is the first letter of the Hebrew alphabet; Tav is the last. The same idea with lunar layers: give a latitude and longitude, or a box, and get one folder back.
Easier than Earth
The Earth version depends on undocumented Google services. Lunar data is public NASA and JAXA data, and most of it is served as open map tiles.

What a run would look like

$ tav --lat 14.09 --lon -56.77 --radius 5km --out marius-pit/

marius-pit/
  manifest.json        every file: what it shows, resolution, date, sun angle
  nac/                 every high-resolution frame of the spot, one per sun angle
  wac.png              low-resolution context
  elevation.tif        heights in metres
  thermal_night.tif    night-time surface temperature
  gravity.tif          gravity anomaly
  radar/               sounder tracks and radar images that cross the box
  pits.geojson         known pits inside the box
  sheet.png            all layers side by side, for a quick look

The coordinates are those of the Marius Hills pit. manifest.json is the file the agent reads first.

Layers

Earth layer · lunar equivalent

Aleph Earth	Lunar equivalent	Resolution	Notes	Do we have it?
Satellite imagery	LROC NAC	0.5–2 m per pixel	Many frames of the same spot at different sun angles	Yes, in part: tiles and crops around hardware sites in Xenarch. The rest: ODE · LROC PDS
	LROC WAC, global mosaic	100 m per pixel	Context	Sample · NASA Moon Trek
	Kaguya Terrain Camera	about 10 m per pixel	Near-global coverage	Sample · JAXA DARTS
	Chandrayaan-2 OHRC	about 0.25 m per pixel	The sharpest, but limited coverage	Yes, in part: 300 tiles from three scenes in Xenarch. The rest: ISRO PRADAN portal
Elevation	LOLA / SLDEM	60–120 m per pixel	Global	Sample · LOLA, PDS · SLDEM, PGDA
	Stereo DTMs from NAC	2–5 m per pixel	Chosen sites only	Sample · LROC data products
3D mesh	No ready-made mesh	—	Made by draping the images over the elevation. Good only where a NAC DTM exists	No · nothing to download, we generate it
Street View	Surface photos from Apollo, Chang’e / Yutu, Lunokhod	—	Only at a handful of landing sites	Sample (Apollo only) · Apollo Lunar Surface Journal · Chang’e data, CNSA
OpenStreetMap	IAU names (Gazetteer), crater catalogues, Lunar Pit Atlas (over 200 pits), USGS geologic map, landing sites	—	Vector data spread over different sources, not one database	Sample · Gazetteer · Crater database · Pit Atlas · Geologic map

Extra layers · not in Aleph Earth, useful for caves

Layer	What it is	Resolution	Do we have it?
Diviner	Night-time temperature	about 240 m per pixel	Sample · PDS Geosciences
GRAIL	Gravity	a few km	Sample · PDS Geosciences
Kaguya radar sounder	Radar below the surface	single tracks	Sample · JAXA DARTS
Mini-RF	Radar images	partial coverage	Sample · PDS Geosciences
M³ and Kaguya	Mineral composition maps	—	Sample · M³, PDS Imaging · JAXA DARTS

Samples on disk

downloaded 6 Nov 2026 · test point: Marius Hills pit

We downloaded a sample of every layer above into data/tav/ (not in git). Rule: a source under 1 GB is taken whole, a bigger one gives one product over the Marius Hills pit, 14.09° N, 303.23° E. Every number below was measured on the downloaded files.

On disk5.8 GB · 906 files

Sources sampled19 of 20

NAC frames over the pit57

Marius Hills pit seen from above: a round hole with a sunlit floor and a crescent shadow — LROC NAC · 232 m across · 0.48 m per pixel

Kaguya Terrain Camera image of a sinuous rille with the pit as a small dark dot — Kaguya TC ortho · 3.6 km across

Kaguya elevation of the same area, the rille shows as a dark band — Kaguya TC elevation · 3.6 km across

NAC stereo elevation model, the pit is a small dark dot in the rille floor — NAC stereo DTM · 1.9 km across

Wide Angle Camera mosaic of the Marius Hills region — LROC WAC · 176 km across

LOLA elevation map of western Oceanus Procellarum — LOLA elevation · 30° across

SLDEM elevation with rilles running across the plain — SLDEM elevation · 28 km across

GRAIL gravity anomaly map, light and dark patches — GRAIL gravity · 30° across

Curve of surface temperature over a lunar day: about 100 K at night, close to 390 K at noon — Diviner · temperature in K by local hour

Kaguya radar sounder · depth profile

Mini-RF radar image with bright crater rims; the strip ends at the right edge — Mini-RF radar · 7.1 km across

M3 spectrometer image at 750 nm with rilles and craters — M³ at 750 nm · 42 km across

Kaguya Multiband Imager false-colour image, fresh craters in blue-green — Kaguya MI false colour · 7.1 km across

Map drawn from the catalogues: crater outlines in blue, pits in pink, IAU names in white — Craters, pits, names · 121 km across

NAC crop of the Apollo 11 landing area from the Xenarch dataset — Xenarch crop · Apollo 11 site

Astronaut James Irwin salutes the flag next to the lunar module and rover, Apollo 15 — Surface photo · Apollo 15 · NASA

Format and statistics · one row per source

Layer	Format	Sample on disk	What the numbers say
LROC NAC	PDS3 `.IMG`, label inside the file. 16-bit integers, brightness = value × 3.05e-5. Not map-projected	1 frame of 57, M122584310LC · 5,064 × 52,224 px · 529 MB	0.48 m per pixel, sun 25° from vertical. The 57 frames span sun angles from 12° to 160°; 13 were taken at night. A frame has no coordinates per pixel: we found the pit by its shadow, at line 30,792
LROC WAC	PDS3 `.IMG`, 32-bit float, plus a GeoTIFF copy	Whole Moon at 948 m per pixel · 5,760 × 11,520 px · 321 MB	The 100 m mosaic is 8 tiles of 2.9 GB each, so we took the 32 pixel-per-degree version. Mean brightness 0.046, at the pit 0.019: dark mare lava
Kaguya TC	PDS3 `.img` + `.lbl`, 16-bit big-endian	1 tile of 3° × 3°, image and elevation · 12,288 × 12,288 px · 577 MB	7.4 m per pixel. Tile heights from −2,218 to −598 m. The pit shows as a 28 m dip (real depth 40 m)
Xenarch	PNG crops 2,048 × 2,048, 8-bit, with CSV and COCO JSON catalogues	123 of 11,593 files · 280 MB of 4.0 GB	4,633 catalogued frames (4,133 Moon, 500 Mars) over 100 hardware sites. Moon median 0.98 m per pixel. Needs a Hugging Face login
Chandrayaan-2 OHRC	PNG tiles 1,024 × 1,024, 8-bit, with a CSV giving latitude and longitude per tile	All 300 tiles in Xenarch · 139 MB	Three scenes at 62–74° S, 0.23–0.26 m per pixel: the sharpest images we hold. Each tile is a dot in the Tav map. The full archive at ISRO PRADAN needs an account
LOLA	PDS3 `.img` + `.lbl`, 16-bit, height = value × 0.5 m. JPEG 2000 copy	Whole Moon at 4, 16 and 64 pixels per degree · up to 11,520 × 23,040 px · 789 MB	474 m per pixel at best. Lowest point −8,982 m at 70.3° S 187.5° E, highest +10,686 m at 5.4° N 201.3° E. Pit rim at −1,566 m
SLDEM	JPEG 2000, signed 16-bit, height = value × 0.5 m	1 tile of 32, 0–30° N 270–315° E · 23,040 × 15,360 px · 177 MB	59 m per pixel. Gives −1,569 m at the pit, 3 m from LOLA. The scaling is not in the label: we inferred it from LOLA. The float version of the tile is 1.6 GB
NAC stereo DTM	GeoTIFF, 32-bit float, metres	MARIUSPIT01 elevation, shaded relief and 4 m orthophotos · 3,331 × 12,529 px · 519 MB	4 m per pixel, 67% of the grid has data. Heights from −1,780 to −1,276 m. The pit shows as an 18 m dip: stereo cannot see the shadowed floor
Surface photos	JPEG scans of film, 4,000–5,400 px wide	8 Apollo photos, missions 11 to 17 · 28 MB	The image links in the Apollo Lunar Surface Journal are broken, so these come from the NASA Image Library. Chang’e data needs an account: not downloaded
Gazetteer	Shapefile, points, 19 fields	Complete · 9,087 names · 46 MB	7,064 satellite craters, 1,615 craters, 112 rilles. Nearest name to the pit: Galilaei M, 25 km away
Crater database	Shapefile, points, 19 fields	Complete · 1,296,796 craters of 1 km or more · 746 MB	Median diameter 1.6 km; 83,061 craters over 5 km. 4 craters within 10 km of the pit, 30 within 50 km
Pit Atlas	Web table, saved as CSV with 13 columns, plus the atlas PDF	Complete · 278 pits · 145 MB	Median depth 9 m, deepest 105 m. Marius Hills pit: 55 × 49 m opening, 40 m deep. 2 pits within 100 km of it
Geologic map	Shapefiles and geodatabase in metres, plus raster	Complete · 12,247 polygons, 49 units · 606 MB	The pit sits in unit EIp, the Eratosthenian-Imbrian plateau unit
Diviner	Text table, 11 columns: position, local hour, 8 temperatures	1 band of 18, latitudes 10–20° N · 1,382,400 rows · 161 MB	Cells of 0.5°, about 15 km, in 96 time steps. At the pit: 392 K just after noon, 97 K before dawn, 105 K night average. The 240 m maps are 6.5 GB each
GRAIL	PDS3 `.img` + `.lbl`, 32-bit float, milligals. GeoTIFF copy	2 global maps of 35: free-air and Bouguer · 2,880 × 5,760 px · 145 MB	1.9 km per pixel. Free-air gravity from −1,330 to +1,782 mGal over the Moon, +16 mGal at the pit, Bouguer +194 mGal
Kaguya radar sounder	PDS3 `.img` + `.lbl`, 8-bit, each column starts with a 55-byte header	1 track, 22 Dec 2007, 34° S to 78° N · 44,636 × 1,000 px · 50 MB	The track passes 59 km east of the pit. The ODE location search does not work for this instrument: tracks must be picked by their start and stop longitude
Mini-RF	PDS3 `.img` + `.lbl`, 32-bit float	1 strip, two products, and the global mosaic at 32 pixels per degree · 419 MB	Strip at 14.8 m per pixel, 1,116 px wide; the pit lies at its edge. Polarisation ratio 0.47 in the strip, 0.53 over the Moon. The mosaic covers 64% of the surface
M³	ENVI cube + `.HDR`, 32-bit float, 85 bands from 461 to 2,976 nm	300 lines of 27,795, cut over the pit with partial downloads · 31 MB of 2.7 GB	140 m per pixel. Reflectance at the pit rises from 0.028 at 750 nm to 0.059 at 2,000 nm. The server sends any byte range, so Tav can cut a box without the full file
Kaguya MI	PDS3 `.img` + `.lbl`, 9 bands, 16-bit big-endian, reflectance = value × 2e-5	1 tile of 1° × 1° · 2,048 × 2,048 px · 97 MB	14.8 m per pixel, bands from 414 to 1,548 nm. Median reflectance at 749 nm: 0.047

Load and merge test · one grid, 1° × 1° around the pit, 15 m per pixel

Sources that load18 of 18

Sources that merge on the map14 of 18

Layers in the merged file15

Fifteen layers resampled to the same grid and shown side by side, each with a pink mark on the pit. Black areas are where a layer has no data — All 15 layers on the same grid · pink mark = the pit · black = no data in that layer

Group	Sources	How it loads	Does it merge?
Map rasters · 9	WAC, Kaguya TC, LOLA, SLDEM, NAC stereo DTM, GRAIL, Mini-RF, Kaguya MI, geologic map raster	GDAL opens every file directly and reads size, projection and scale from the label. No custom code	Yes. One reprojection call per layer. Only SLDEM needed a fix: its longitudes run 0–360°
Rasters without a map · 2	Diviner, M³	Diviner is a text table, M³ an ENVI cube with pixel positions in a separate file	Yes, with about 15 lines of code each to place the values on a grid first
Vector · 3	Gazetteer, crater database, Pit Atlas (and the geologic map polygons)	Any shapefile or CSV reader	Yes, as points and outlines drawn on top. Longitudes come as 0–360°, ±180° or metres
Not on the map · 4	NAC raw frame, Kaguya radar sounder, Xenarch crops, surface photos	All open (PDS3, PNG, JPEG)	Not as they are. A NAC frame needs the spacecraft geometry (SPICE, ISIS) to be projected. A sounder track is a depth profile along a line. Crops and photos carry only a site position

The result is data/tav/merged/marius_stack.tif: one GeoTIFF of 2,048 × 2,048 pixels with 15 named bands, 99 MB. The layers agree with each other: the NAC stereo DTM and SLDEM differ by 0.4 m on average (3.8 m spread), the hole in the NAC DTM falls 14 m from the Pit Atlas position, and the pit shadow in Kaguya TC 33 m from it. Kaguya TC elevation sits 14 m higher than SLDEM everywhere, so it needs a constant correction.

What does not merge · shown as points on the Moon

A photo or a crop cannot become a layer, but it has a place. Each dot is one item on disk: click it to see the picture. The violet line is the radar sounder track.

Map of the whole Moon with dots at the sites where we hold photos and crops

8 surface photos73 Xenarch cropsNAC raw frameRadar sounder track

The two that need more work

The whole NAC frame as a long thin strip of cratered mare, with the pit circled near the upper edge — NAC raw frame M122584310LC, whole · 25 km long, 2.4 km wide · pink circle = the pit

NAC raw frame. The camera has a single row of 5,064 pixels. The spacecraft flies forward and the row is read 52,224 times: the file is those rows stacked in time order, a strip 25 km long. Nothing in the file says where a pixel is on the Moon. The label gives only the four corners, good to about 300 m. To put it on the map you need the position and pointing of the spacecraft at each instant (the SPICE files) and software that uses them (ISIS, from USGS). Until then it is the sharpest picture we hold, 0.48 m per pixel, with a dot on the map.

Kaguya radar sounder, one track · 34° S to 78° N, 3,400 km · down = 25 km of echo delay

Radar sounder track. Not a picture from above: a vertical cut through the ground. The radar sends pulses straight down as the orbiter flies north. Each column is one position along the orbit; going down a column is the time the echo took to come back, 25 m per pixel. The bright line is the surface: flat over the mare in the middle, rough over the highlands at both ends. Echoes below it can be buried lava layers, or reflections from craters to the side. It merges with the map only as a line, the violet one above, and this track passes 59 km east of the pit.

Five different containers (PDS3, GeoTIFF, JPEG 2000, ENVI, shapefile) and three longitude conventions (0–360°, ±180°, metres). GDAL already reads all of them, so Tav does not need its own readers.