Online Resource 1 (ESM_1.zip)
The Uruana de Minas magnetic anomaly within the South Atlantic Anomaly: regional-field control and limits of cross-survey prediction
Studia Geophysica et Geodaetica submission package, 5 November 2026
Lucas Giovani Ribeiro and Marcos André Simonssini
Independent researchers, Uruana de Minas, Minas Gerais, Brazil
Corresponding author: Lucas Giovani Ribeiro, coin26@mail.com

This archive is supplied as Online Resource 1. Internal references to S1
in dated development protocols identify this supplementary archive. The original numerical inputs, analysis scripts, outputs and figures are
preserved. historical_sensitivity supplies the 3x3 depth/damping audit
in Section 6.2. resolution_audit adds the post hoc synthetic recovery
and sign-constrained ridge analyses in Sections 7.1 and 7.2. It includes
a frozen protocol, a documented weak-field amendment, all initial and
amended outcomes, and independent numerical checks. See its README.txt.
The manifest has been regenerated for this archive.

Reproduce the principal numerical comparisons from the extracted directory:
  python local_recovery/recovery.py
  python core_lithosphere/verify.py
  python regional_specificity/check_results.py
  python processing_audit/verify_stages.py
  python distant_transfer/verify.py
  python spatial_sensitivity/verify.py
  python resolution_audit/verify.py

The sensitivity analyses were run with CPython 3.12.14, NumPy 2.3.5, SciPy
1.17.0, pandas 2.2.3, Matplotlib 3.10.8 and Pillow 12.3.0 on Linux. Full
spatial recalculation additionally uses a C99 compiler (tested with GCC 13.3).
See spatial_sensitivity/README.txt for the exact commands. The bundled numerical
checks and data subsets work offline. The source hash files and IBGE data
attribution below identify the origins and documented terms of external inputs.
spatial_sensitivity/analysis_decisions.tsv records choices, timing and the
relevant internal protocols. These records document a retrospective study.

To verify every archived file before running any scripts:
  sha256sum -c MANIFEST.sha256

Generate all ten main-text figures as numbered vector PDFs:
  python render_figures.py

Figure  Subject                                      Source script
  1     Uruana and the weak-field region             figures/fig.py
  2     Northern recovery and the 23 sectors          local_recovery/figure.py
  3     Main-field components on the flight lines    nested_gradients/plots.py
  4     Observed signs across eligible sectors       nested_gradients/plots.py
  5     Geographic substitution at fixed geometry    background_transplant/figure.py
  6     Historical prediction and rotation tests     figures/fig.py
  7     Historical rotations matched offshore        matched_rotation/figure.py
  8     Directional resolution                       directional_resolution/figure.py
  9     Historical profiles and 2011 predictions     figures/fig.py
 10     Conditional sign stability and recovery     resolution_audit/figure.py
 S1     Distant historical transfer in Brazil        distant_transfer/figure.py
 S2     Near-zero raw target slope on line 12350     spatial_sensitivity/figure_s2.py

The individual analysis scripts retain their development filenames. Figure and section numbers throughout this README refer to the current manuscript. The development filenames below are paired with their current figure numbers. regional_specificity/README.txt and processing_audit/README.txt describe how to rerun the full two-epoch sector comparison and processing checks. local_recovery/README.txt describes the line-level sensitivity and the 85 single-bin omissions on the five eastern lines. core_lithosphere/README.txt describes the EMM2017 main-field and lithospheric comparison in Section 3.1, including the optional rebuild from NOAA coefficients. The current MANIFEST.sha256 covers every other file in this archive.

Development history and detailed reproduction instructions
----------------------------------------------------------
Supplementary data and code
The Uruana magnetic anomaly within the South Atlantic Anomaly
Manuscript version 7, 3 November 2026 | L. G. Ribeiro and M. A. Simonssini

New in version 7: background_transplant/README.txt describes the native decomposition, global scan in both survey epochs, comparison along the native parallel and numerical checks. The bundled Brazilian measurements and IGRF coefficients suffice to reproduce these results.

Version 6: directional_resolution/README.txt gives the complete bearing sweep over the historical Brazilian surveys, tests on withheld production lines and geographic blocks, and transfer between regions. See directional_resolution/PROTOCOL.txt for the order of exploratory checks. Earlier workflows for nested gradients, geographic transfer and comparisons outside the contour remain included.

Run the commands below from the root of the extracted archive in a Python environment with numpy, scipy, pandas and matplotlib. The distributed outputs allow the figures and checks to run immediately. No network or original large survey downloads are required to use the included subsets. SOURCE_SHA256.txt and SHA256_FONTES.txt identify original sources used in extraction. MANIFEST.sha256 covers the contents distributed here.

Quick verification and reproduction of figures
  python figures/verify_paper.py
  python figures/fig.py

The first command recomputes the change in point-weighted, within-segment RMSE and its range when each geographic block is removed. Results are written to figures/leave_one_block_out.csv. The second produces figures/fig1.png (Figure 1), figures/fig2.png (Figure 6) and figures/fig3.png (Figure 9) from model tables and survey observations. Figure 1 uses the 1:110 million Natural Earth coastline (public domain). Source GeoJSON: https://github.com/nvkelso/natural-earth-vector/blob/master/geojson/ne_110m_coastline.geojson (Git blob 0dde16b6687ab562b2ca8864bb1b8828b4093c99). The geographic coordinates used by the figure are included as figures/coast.json.

Recalculate the underlying models (single-threaded BLAS recommended)
  export OPENBLAS_NUM_THREADS=1
  python uruana_teste_carta/avaliar.py
  python uruana_teste_carta/falhas.py
  python uruana_teste_carta/checagem_fisica.py
  python uruana_teste_carta/verificar.py
  python amas_test/run.py
  python amas_test/background_swap.py
  python uruana_orbita/teste.py
  python figures/verify_paper.py
  python figures/fig.py

Regional model: uruana_teste_carta
Training grid: ANP 0032, Sao Francisco Norte, 2008, 1.5 km grid, diurnally corrected mag_var channel. Historical observations: SGB/CPRM 1009, Convênio Geofísico Brasil–Alemanha, CGBA_LAT_LONG_PGW.XYZ, digitized residual. Input includes nominal altitude per group, tile assignments are derived from coordinates. Output/trechos.csv contains the counts, historic line IDs, geographic blocks and paired model scores. The field names are lowercase; the actual path is output/trechos.csv. Segment-level changes and errors of the geological gain predictions use different target quantities. The map checks use the Serra da Ilha (CPRM/COMIG, 2002) and Buritis (CODEMIG/UFMG, 2015) sheets. PROTOCOLO.txt is the internal analysis protocol in Portuguese.

Central 2011 test: amas_test
SGB/CPRM project 3087, Formoso–Buritis–Lagoa Grande. The input survey subset (flight_2011.csv) contains per-line bin medians from the original MAGBRU, position and aircraft altitude channels. Original product Mag-3087.XYZ. Lines 12340–12470 in steps of 10. The archived window origin in the 2011 survey coordinate grid is 365408.554 m E, 8224103.341 m N. Script flight_original.py documents extraction from the larger project ZIP. The central equivalent-source fit is determined by the ANP 2008 grid only. Slopes are in nT/km of grid north. Panels a and b of Figure 9 use the regional segment outputs. Panel c uses the original 2011 channel and central-model predictions. Survey height and channel choices are detailed in the scripts.

Orbital scale check: uruana_orbita
The central grid and its locally fitted source layer give a conditional forward-model prediction at 450 km. It is not a measured satellite signal. teste.py reproduces modelos_450km.csv and referencia_regional.csv.

Regional reference model and data portals
IGRF-14 official coefficient source: https://www.ncei.noaa.gov/products/international-geomagnetic-reference-field
SGB: https://geosgb.sgb.gov.br/downloads/
ANP: https://reate.cprm.gov.br/anp/TERRESTRE
Serra da Ilha cartographic record: https://rigeo.sgb.gov.br/handle/doc/2354


Version 2 | Nested northward gradients and geographical comparison
Run these additional commands from the extracted archive root:
  python nested_gradients/nested.py
  python nested_gradients/harmonics.py
  python nested_gradients/controls.py
  python nested_gradients/bands2008.py
  python nested_gradients/window.py
  python nested_gradients/verify.py
  python nested_gradients/plots.py

The extension reproduces Figures 3 and 4 and the associated results in Sections 3 and 4. It reuses the central 2011 bins and IGRF-14 coefficients delivered with the original analyses. Historical decisions, selection rules and follow-up decisions are given in nested_gradients/PROTOCOL.txt. Paired signs on geographic sectors and the effect of a dipole substitution are distinct from RMSE of the 1971–72 predictions. No global priority, event probability or causal attribution to the SAA is inferred.

Regional controls come from the native project 3087 Mag-3087.XYZ production lines. Each comparison sector covers the same line IDs in its target and recovery windows. Translated sectors may share flight lines and neighboring grids. To regenerate the included flight bins from the original 3087 ZIP:
  python nested_gradients/extract_controls.py /path/to/3087.zip
The program applies the geographic window BEFORE binning. The original central 14 lines reproduce the delivered central flight bins. For the 2008 grid, restore the subset from the original native grid with:
  python nested_gradients/bands2008.py /path/to/0032_SAO_FRANCISCO_MAG_grid.asc
This is optional because the native subset, including missing values, is supplied. The measured-field eligibility and signed margins at all 51 nominal origins were also checked against the authors' archived 2008 template implementation at the time of packaging.

Outputs in nested_gradients/results/
  line_gradients.csv: per-line modeled and observed target/return gradients (nT/km) at actual GPS or fixed model altitude. Measured slopes come from actual flight data in both cases.
  harmonic_checks.csv: single-degree vector omissions and dipole-plus-single-degree alternatives. The sign comparison never subtracts harmonic intensities.
  longitude_sweep.csv: one-degree longitude steps, three latitudes, four reference epochs. Field derivatives are at fixed 0.8 km, and 28 µT membership uses surface intensity at the same modeled epoch.
  control_line_gradients.csv and control_groups.csv: eligible translated flight windows at 2011, raw and compensated channels.
  bands2008_detail.csv and joint2008_2011.csv: all band–width–channel combinations, coverage and joined sector counts.
  gradient_windows.csv: exact open-interval bounds for a uniform additive northward gradient. The Uruana joint interval is bounded by two different 2011 lines.
  plots.py regenerates fig4.png (current Figure 3) and fig5.png (current Figure 4) in the nested_gradients directory. The regression checks in verify.py verify all main new results.

Version 3 | Distant geographic transfer and historical datum audit
All paths below are relative to the supplement root. The original historical PGW product is a digitized IGRF-reduced residual. Its geographic coordinates are in SAD69. The alternative WMC product has an undocumented arbitrary addition and was excluded. See Monteiro, A.C. (2002), Relatório das Áreas Digitalizadas pela PGW e WMC, pp. 1–2, Leia-me.pdf in the project 1009 XYZ archive. Heights are the archived barometric levels, not point-specific measured radar heights. 1971.5 is the representative epoch of an acquisition spanning 1971–1972.

The original version-2 frozen local score (32.4%) can still be reproduced by the previous regional transfer workflow. Version 3 also transforms the historical positions using IBGE SAD69 to SIRGAS2000 offsets and re-evaluates identical local segment memberships (32.1%). It reports the new area using transformed positions (3.6%). The two percentages refer to different historical locations. The local segment set is 8–66 km from Uruana; it is distinct from the 14 target lines in 2011. The previous untransformed-coordinate score for the new distant membership is supplied as a sensitivity.

  export OPENBLAS_NUM_THREADS=1
  python distant_transfer/model.py
  python distant_transfer/score.py
  python distant_transfer/uncertainty.py
  python distant_transfer/verify.py
  python distant_transfer/figure.py

distant_transfer/results/segments.csv and forecasts.csv.gz contain the segment geometry, centered observations and all competing point predictions. The local 'run' code -1 is a placeholder for the legacy segment ID. Historical line IDs are strings. In scores.csv the raw score means profile-shape RMSE without extra slope calibration, not an original total-field magnetic observation. The calibrated score uses slopes trained only on the 75 nearby profiles; for those local profiles themselves it is an in-sample diagnostic and must not be presented as local validation. The curvature score removes a separate linear east-coordinate term within each segment as an explanatory diagnostic. Block uncertainty is descriptive, depends on origin and grid size, and is not a prospective survey-level significance test.

Supplementary Figure S1. Panel A shows the Uruana center (star), the nearby 2008 training tiles (gray), and the distant test tiles colored by the pooled reduction in RMSE of their 24 km block. Panel B shows the reductions in all 74 distant blocks. 2008 training disks have 24 km radii. The closest distant disk center is 48 km from the closest earlier disk center. All 539 distant production line identifiers differ from the ones previously scored.

Supplied inputs suffice to rerun all commands. Optional source extraction:
  python distant_transfer/extract.py /path/to/1009_XYZ.zip /path/to/0032_SAO_FRANCISCO_MAG_grid.asc
  python distant_transfer/prior.py
The previous north-sector line-ID inventory is included. The original 2008 grid and CGBA archive checksums appear in amas_test/SOURCE_SHA256.txt. The extract script retains nearby WMC rows for a coverage audit when rerun; that file is not required for analysis.
  
Coordinate-grid attribution. br_ibge_SAD69_003.tif is from the Instituto Brasileiro de Geografia e Estatística (IBGE), NTv2 converted to GeoTIFF by OSGeo/PROJ-data. https://cdn.proj.org/br_ibge_SAD69_003.tif . CC BY-SA 4.0, https://creativecommons.org/licenses/by-sa/4.0/ . The accompanying ibge_offsets.npz is a lossless repackaging of the latitude and east-positive longitude offset bands by the manuscript authors, released under the same license. IBGE_README.txt is included. The geographically applicable grid applies to the entire area tested; geographic SIRGAS2000 longitude and latitude are projected on the WGS84-compatible 2008 UTM 23S geometry. Optional re-extraction of offset bands (requires Pillow and the system libtiff library):
  python distant_transfer/datum.py distant_transfer/br_ibge_SAD69_003.tif

Version 4 — Alaskan archive, epoch transfer and Brazilian rotation overlap
See outside_saa/HISTORICO_ALASCA.txt for the Portuguese inventory of three survey pairs, archived endpoints and the 60-pair catalog. Follow all six commands in outside_saa/README.txt. The folder includes the two USGS products for the 1977/2008 comparison, both older state survey datasets, their S16 result package, scores, block sensitivity, a bound on the change in RMSE, catalog rotations and the post hoc Brazilian overlap restriction.

Version 5 — Global geographic matching and the direction of field rotation
Reproduction commands and methods are in matched_rotation/README.txt. That directory also contains results/Figure_6.png (current Figure 7), the independent IGRF check, the record of candidate external surveys in Portuguese (CANDIDATOS.txt), and the source files for the modeled global geometry and scoring on Brazilian data.
