Reading the notebook archive safely¶
The notebook pages preserve the existing tutorials. They are useful procedural examples, but their prose, saved execution state, and current source can differ. The maintained workflow guides distinguish those differences without silently editing the originals.
Known interpretation traps¶
| Historical wording or pattern | How to interpret it |
|---|---|
| “Fits use full flightlines” | Check sample_region() and SAMPLE_REGION; defaults can supply 2,000-row subsets |
| “Nothing is read; opening is instant” | Lazy cubes still require metadata and sometimes geometry I/O; describe() reads samples |
| “Satellite topographic correction is not applicable” | An implementation boundary, not a universal physical statement |
| “No satellite can offer overlap checks” | Not a general scientific rule; observations can overlap across acquisitions |
force_topo=True |
A deliberate demonstration that bypasses the global diagnostic gate |
| “Everything here takes a corrected cube” | Inspect the actual variable: AVIRIS-3 spectral examples use provider win_ds |
| “GeoTIFF cannot carry spectral/bad-band metadata” | The current writer's standardized fields are limited; the file format can carry custom metadata |
| “36 entry points exercised” | A manually written checklist is not measured test coverage; listed calls may not be executed |
| “Clear 100%” | Can reflect absent flags or disabled derivation; not independent clear-sky proof |
| A very short atmospheric runtime | Inspect logs for reused inputs or a completed retrieval |
Parameter-table drift¶
Some notebook tables summarize defaults differently from executable source. For example, bad-band masking should not be assumed to remove coordinates, and an ortho option may be variant-specific. Use the source-derived API signatures and current reader code as the implementation reference.
Saved-product helpers¶
Notebook-local load_product() helpers reconstruct only part of the dataset contract. Confirm band widths, good-band flags, units, geometry, quality, and original band indexing before treating reconstructed products as interchangeable with a reader output.
Execution precautions¶
Read every control panel before running all cells. Change paths, reduce windows, limit worker counts, separate work directories, and inspect external asset requirements. Notebook execution can write or overwrite output files and trigger downloads. This site does not run that code for you.
Evidence status¶
Saved figures and logs are presented as archived evidence, not new test results. A notebook with no error output may simply not have been run. The tutorial index explicitly distinguishes notebooks with saved outputs from those without them.
Maps on this website¶
Map cells have browser-only Leaflet previews that support pan, zoom, and reset. The site does not start Jupyter or execute notebook code. Map tiles come from OpenStreetMap and require a network connection.
If saved widget state is present, the generator recovers the map view and supported polygon, rectangle, line, point, and GeoJSON layers. For the archive tutorial, recorded provider metadata is matched to each saved result identifier and acquisition minute to recover the same bounding boxes used by the Jupyter map. These recordings are stored under docs-site/map-recordings/ and checked against the saved result table before use. NEON monthly deliveries are shown at site coordinates. If saved geometry and matching recordings are both absent, only the saved center is shown with an explicit caption. New archive matches are never substituted for the saved result list.
To retain geometry, enable saving widget state in Jupyter before saving the notebook; see Jupyter's widget-state documentation. Browser previews provide map navigation; Python-backed search and download controls still belong in the original notebook.
Synchronizing notebook changes¶
Edit the originals in tests/0_src_code/, save them, then rebuild from the repository root:
Refresh the tutorial page afterward. Markdown and code reflect the saved notebook; figures, maps, and text outputs reflect its saved output/state. Rerun the relevant cells and save if you want new outputs. The build itself does not run those cells. Changes are not automatically watched by the static preview.
Saved text outputs remain expanded by default; click a Saved output heading to collapse a log.