Inventory of the collection
Lists layers, formats, CRS, encoding, count and extent with ogrinfo, gdalinfo -json and pyogrio, without touching the files.
Octo for GDAL, GeoPandas, Shapely and rasterio
Every format comes in. What comes out checked is what counts as data.
GDAL 3.13 (3.13.3, from Aug 18, 2026) is the layer that QGIS, map servers and geospatial Python use to read and write. Octo converts entire collections to GeoPackage and GeoParquet, generates COGs with overviews, audits geometry and CRS and calculates indices per field plot, all in batch and with a log. Declaring a CRS, accepting a geometry fix and signing the map stay with the technical lead.
GDAL3.13.3Aug 18, 2026, MIT license
GeoPandas1.2.0Sep 28, 2026, BSD-3-Clause
COG driver3.1since GDAL 3.1; GeoParquet 1.1 since 3.9
Sources11listed at the end of the page
01
What Octo does with GDAL and Python.
Six possible deliverables, taken from the survey. All of them start by reading the collection without changing the original, and all go through the technical lead before becoming official data.
Batch conversion
Shapefile, FileGDB and CSV to GeoPackage and GeoParquet with ogr2ogr, explicit encoding, PROMOTE_TO_MULTI and a log per file.
Geometry and CRS audit
Flags layers with no CRS, invalid, empty and duplicate geometries and mixed types with vectorized Shapely 2, across millions of features.
Deliverable: table by layer to decide what to fixCOG with overviews
Generates COGs with chosen blocks, compression and overviews, and checks the generated file: structure, CRS, band statistics against the original.
Deliverable: checked COG + report of the options usedReprojection with judgment
Chooses the transformation between SIRGAS 2000 and older datums with pyproj and shows the expected shift before applying it.
Deliverable: reprojected layer + note on the transformation usedIndices per field plot
Clips the scene by field plots with rasterio, applies a cloud mask, calculates NDVI and summarizes by polygon with the fraction of valid pixels.
Deliverable: index COG + CSV per plot for the agronomist
02
Eight arms, data checked before moving on.
The same Octo cycle, applied to conversion and processing. Each arm has a narrow role; people decide the datum, the fixes and the use of the data.
-
Discovery
Maps the collection
GDAL version, available drivers, folders, formats, layers, sizes and what each in-house system reads.
-
Data
Looks at CRS, fields and geometry
Each layer's datum, encoding, field types, invalid geometries and what is lost in conversion.
-
Security
Separates what is sensitive
Layers with personal data (LGPD), cloud storage credentials kept out of scripts and the license of each source.
-
Architecture
Chooses the format
GeoPackage for delivery and editing, GeoParquet for analysis, COG for remote raster; compression and blocks.
-
Development
Writes the batch
Conversion, audit and index scripts with a dry-run mode, a log per file and the library version recorded.
-
Review
Checks by content
Count before and after, extent, area, COG structure, band statistics. Exit code zero is not a check.
-
Release
Only with your OK
Replacing a collection, publishing a COG or a shared GeoParquet happens with the lead's approval.
-
Documentation
Records the lineage
Where each file came from, with which version and options it was converted, what was fixed and who approved.
Octo works with read access and as support. It does not declare a CRS, replace the original or publish without approval, and it does not sign maps or reports.
03
Each task, its own path.
The criteria Octo uses to choose, with the documentation for the installed version.
| Task | Path | Note |
|---|---|---|
| Inventory a collection | ogrinfo -so -al, gdalinfo -json | Read-only |
| Shapefile to GeoPackage | ogr2ogr -f GPKG -nlt PROMOTE_TO_MULTI | Explicit source encoding |
| Vector to GeoParquet | ogr2ogr -f Parquet or to_parquet | The driver depends on Arrow; check the specification version |
| Reproject and validate in one pass | gdal vector pipeline | Provisional command since 3.11 |
| Raster to COG | gdal raster convert --of COG | Lossy compression is a human decision |
| Clip raster by polygon | gdalwarp -cutline or rasterio.mask | Same CRS first |
| Audit geometry | Shapely 2 (is_valid, is_empty) | Flag, don't fix without sign-off |
| Batch spatial join | geopandas.sjoin | Projected CRS for measuring |
| Transform between datums | pyproj.Transformer | Check the chosen operation |
| Check a COG | rasterio: blocks, overviews, statistics | By the file, not by the command |
04
From one-off conversion to a collection with lineage.
Three steps. Each one only makes sense after the previous one has shown results on your team.
A proposal, not a promise
Inventory of the collection
Octo shows what exists, in which format and datum, without changing anything.
- Layers, formats and sizes
- Files with no CRS and doubtful encoding
- List of questions for the lead
Batch conversion and audit
GeoPackage, GeoParquet and COG generated by script, each file checked against the source.
- Log per file with count and hash
- Report of fixed geometries
- COG with checked structure
Routine with lineage
When new data arrives, the same batch runs, compares with the previous version and records what changed.
- Incremental collection update
- Indices by harvest or by campaign
- History of versions, options and approvals
05
Confirmed in a source.
Each point below has a source published by the project or by the organization responsible for the standard. The sources are in the next section.
4
points with a published source, each with its reference shown alongside.
- ConfirmedGDAL 3.13.3
- Stable release from Aug 18, 2026; the project maintains only the latest release branch, with fixes roughly every two months.Source: gdal.org/download and OSGeo/gdal releases
- ConfirmedProvisional
gdalcommand - Since 3.11, offered as an alternative to the traditional tools; names and behavior may change until a major release.Source: GDAL programs index
- ConfirmedCOG and GeoParquet
- COG driver since 3.1, with
BLOCKSIZE512 and LZW by default; GeoParquet 1.0 since 3.8 and 1.1 since 3.9.Source: COG and Parquet driver pages - ConfirmedGeoPandas 1.2.0
- From Sep 28, 2026, with GeoParquet 2.0 and reading over HTTP; pyogrio has been the default engine since 1.0.Source: GeoPandas changelog
06
What Octo does and what stays with you.
Octo prepares everything, but it does not sign. The document of record stays with whoever is accountable for it.
Octo prepares
Surveys, calculates, organizes and produces the documents: spreadsheets, reports, maps and drafts.
Deliverable: material ready to checkYour team checks
The licensed professional checks what Octo delivered before signing. If something doesn't add up, it goes back to Octo with the correction.
Step: review by the signerThe signer
Octo does not sign. Projects, technical reports, ART/RRT, accountability reports and any official document are still signed by the company's or agency's licensed professional: here, the technical lead for the map or the survey.
Signature: technical lead for the map or survey
07
Sources.
GDAL documentation, GeoPandas changelog, the Python package index and project repositories, consulted on Oct 31, 2026.
- GDAL, download3.13.3, Aug 18, 2026
- GDAL, licenseMIT for most of the code
- GDAL, programsProvisional gdal command
- gdal vector pipelineSteps and examples
- COG driverCreation options
- (Geo)Parquet driverSpecification versions
- GeoPandas, changelog1.0, 1.1.0, 1.2.0
- Shapely on PyPI2.2.0, Oct 7, 2026
- rasterio on PyPI1.5.2, Sep 30, 2026
- GeoParquet, releases1.1.0 and 2.0.0-rc.1
- OpenStreetMap, licenseODbL and attribution
Your geographic collection, checked file by file.
The assessment starts with an inventory of the collection, without changing anything: formats, datums, problem geometries and what would be lost in a conversion. You get the list and the decisions only your team makes, and you decide with it in hand.