Subsurface SAR Doppler Tomography — Controlled Reproduction & Refutation

An independent, controls-first reproduction of the single-pass SAR Doppler "micro-motion tomography" method of F. Biondi & C. Malanga — the method behind the disputed 2025 claims of an "underground city" beneath the Giza plateau.

Result, in one line: rebuilt faithfully from the authors' own papers and patent, with the elementary controls the original omits, the method yields no reproducible evidence of deep subsurface structure — and the confident outputs it does produce are reproducible as artifacts. The surface-vibration measurement front-end is legitimate; the deep inference is not.

What the controls show, step by step:

• The patent's depth-focusing "steering matrix" is, by its own wording, a discrete Fourier transform, so depth is a Fourier bin with a free metre label. (A DFT alone does not create a peak from white noise; the systematic shallow peak comes from accumulating and detrending the tracked displacements — paper v6 §3.1, §4.2. Earlier wording "structured output from any input" is withdrawn.)
• The reported depths follow from a physically impossible ~22 kHz investigation frequency and are an arbitrary relabelling of the depth axis, not a measurement.
• A faithful reproduction on Butte, MT (a densely mapped mining district) yields a confident high-contrast "detection" that the guards expose as a surface-pinned artifact matching none of the documented workings.
• Five sites across two independent sensors (Umbra ×4, Capella ×1) — including the authors' own Vesuvius — are statistically indistinguishable from their nulls, while in-data positive controls pass.
• Stacking five passes of a known-empty pit reinforces the surface artifact rather than revealing structure, so "200+ scans agree" reflects the shared operator, not corroboration.

This is a critique of method and mathematics, openly reproducible and falsifiable — not an allegation of intent.

The paper

paper/Giza_SAR_Doppler_Reproduction_and_Refutation.pdf — the merged, definitive preprint (reproduction + refutation). Build it from source with:

python3 paper/build_merged.py

Reproduce the analysis

# 1. install
pip install -r requirements.txt

# 2. fetch a free scene (example: Butte, MT) — see data/README.md for IDs
python3 src/fetch_umbra.py --task "Butte" --collect 2024-03-07-04-48-26_UMBRA-04 --products SICD

# 3. run the end-to-end pipeline + all controls on that scene
python3 src/tomogram.py --sicd data/2024-03-07-04-48-26_UMBRA-04_SICD.nitf \
        --ground-truth --stability --f-compare

# 4. reproduce the artifact at the authors' aggressive settings
python3 src/tomogram.py --sicd data/<butte>.nitf --n-sub 256 --n-chirp 3 --lrsd --f-investigation 22000

# 5. the stacking experiment (five same-geometry passes)
python3 src/stack.py --sicds data/<pass1>.nitf data/<pass2>.nitf ... --n-sub 128

Every stage also has a pure-numpy self-test that needs no data and validates the code against known truth:

python3 src/tomogram.py --selftest      # also: subaperture.py, micromotion.py, stack.py

Repository layout

subsurface-sar-tomo/
├── paper/        the preprint + reportlab build scripts (build_merged.py is canonical)
├── src/          the pipeline: subaperture, micromotion, tomogram, stack, steering stress test
├── docs/         method analysis, patent decode, ground-truth dossiers, technical reference
├── reference/    pinned scene metadata
├── data/         raw SAR scenes land here (gitignored — fetched from open buckets)
└── runs/         output figures (gitignored — regenerated by the pipeline)

Data availability

All scenes are free, openly licensed third-party data — Umbra Open Data and Capella Open Data (CC-BY 4.0), from the AWS Registry of Open Data. The raw products (hundreds of MB to several GB each) are not committed; src/fetch_umbra.py and src/fetch_capella.py retrieve them from source, and data/README.md lists the exact scene identifiers used in the paper.

Falsifiability

The conclusion would be withdrawn if the method, run with these controls on a site of independently documented subsurface structure, produced an above-null, leakage-clean, depth-stable band matching the known structure (under a physically defensible velocity model, not a 22 kHz relabelling). The pipeline, controls, and scene IDs are open for exactly that test.

Citation

See CITATION.cff. Code is released under the MIT License (LICENSE); the third-party SAR data retains its own CC-BY 4.0 license from Umbra/Capella.