Data deposit for "Short communication: Continental-scale drainage divide analysis with pydivide"

Superseded by version 4 (https://doi.org/10.5281/zenodo.23953881). The files of this version include brazil_crop_4000.tif, a window of FathomDEM v1-0 Americas (Fathom, 2025, https://doi.org/10.5281/zenodo.14523356, CC BY-NC-SA 4.0), and are therefore no longer publicly available. Version 4 does not include the window and describes how to rebuild it.Data behind the paper "Short communication: Continental-scale drainage divide analysis with pydivide" (Earth Surface Dynamics): the 4000 x 4000 Brazil validation window (30 m), the MATLAB TopoToolbox 3 reference exports for the two validation DEMs, every benchmark and experiment script with its outputs, the MATLAB scripts of the study, the four patches applied to DIVIDEobj to obtain a runnable reference, the complete divide network of south-eastern Brazil (1.2 x 10^10 cells) as a GeoPackage, and the signed across-divide tables of the Swabian Alb and south-eastern Brazil. README.md describes every file; MANIFEST.json gives the size and SHA-256 of each. The four GeoPackages are stored as numbered parts of 512 MiB (name.gpkg.001, .002, ...; a few parts are further split into .001 to .004) so that the upload could be completed. To rebuild them, download all parts of a file together with REASSEMBLY_MANIFEST.json and reassemble.py into one folder and run "python reassemble.py REASSEMBLY_MANIFEST.json" (Python 3, no other package): the script checks the SHA-256 of every part and of each rebuilt file against the manifest, and MANIFEST.json gives the same SHA-256 values. On a Unix shell, "cat name.gpkg.0* > name.gpkg" followed by a SHA-256 check gives the same result. The last sentence of HOW_TO_REASSEMBLE.txt, about a 7z archive, does not apply to this record. The software is a separate Zenodo record created from https://github.com/ATLAS-Topo/pydivide; cite both. Data: CC BY 4.0. Scripts: GPL-3.0-or-later.Version 2 adds brazil_full_bl5_divides.gpkg, the divide network of south-eastern Brazil built on the river network trimmed at 5 m, which Fig. 1 of the paper shows; external_reference_exports.zip, the MATLAB TopoToolbox 3 reference exports of the seven external validation landscapes, one folder per landscape; and additions_2026-10.zip, with the production logs of the continental runs, the corrected benchmark timing scripts and the timing records of Table 1 of the paper, and the rerun of the 20,000 x 20,000 window. README.md describes every file and lists the corrections in version 2.Version 3 holds the same files as version 2 with README.md corrected: it no longer lists files that are not part of additions_2026-10.zip.