Day 2: Real NASA terrain analysis

Today I built the part that actually finds the best landing site in an area near the south pole of the moon. I wrote a Python pipeline that processes real NASA 5-meter-per-pixel lunar terrain data to find optimal landing points.

What it does

The script pulls real slope GeoTIFFs from NASA’s Planetary Geodynamics Lab (PGDA). for reference, that’s the same 5m/px Digital Elevation Models used for actual Artemis landing site selection. For each candidate region it:

1. Downloads the real slope raster (40–140 MB each)
2. Scores every location by the mean slope across a 100m lander footprint, because a real landing zone has to be flat across the whole area the lander occupies
3. Finds the optimal point down to around 5m of precision
4. Converts the winning pixel from polar-stereographic coordinates back to lat/lon (this part sucked)
5. Writes everything to a JSON file the app serves instantly

Results

Processing NASA’s actual DEMs, the optimal landing points came out to:

• Nobile Rim 2: 0.45° mean slope over 100m
• Haworth: 0.91°
• Shackleton Crater Rim: 1.15°

The hard part

Two bugs. The first version picked nodata artifact pixels that read as “perfectly flat 0.0°.” The single-pixel approach could also land on a fluke-flat spot surrounded by cliffs. Fixed both: excluded the artifact values, then switched to integral-image windowed averaging so it evaluates the full lander footprint in O(1) per location instead of brute-forcing it.

Next

Wiring this into the frontend so “Search Area” surfaces the precise optimal point, then on to the launch window calculator using JPL Horizons ephemeris.