Related work

Name the neighbours yourself. A reviewer who finds them before you do will read the omission as either ignorance or concealment, and both are worse than whatever the comparison costs.

Four packages sit close enough to matter. None of them does what basinkit does, but each one does part of it, and two of them do that part better.

The claim, stated so it can be tested

No existing open-source software package provides, through a single installable API and without requiring any account, login or credential for any of its default sources, the complete chain from an arbitrary global outlet coordinate → a delineated upstream basin polygon → multiple Earth observation datasets returned as polygon-masked gridded arrays.

Every clause in that sentence is load-bearing, and each one excludes a real tool. Drop any of them and the claim is false.

clause	what it excludes
software package	mghydro Global Watersheds: global, account-free, multi-dataset, but a web app
any global outlet coordinate	watershed-workflow, HyRiver, StreamStats, Model My Watershed, all US-only
no account for any default source	rabpro (Earth Engine + MERIT-Hydro logins), Caravan (Earth Engine), eodag (per-provider credentials)
polygon-masked gridded arrays	delineator (geometry only), mghydro and rabpro (zonal statistics), hydromt (extent-based clipping)

The four to cite

hydromt: the closest, and the real threat

Deltares/hydromt. Concede all of this: get_basin_geometry takes a point and returns the upstream basin, with sub-basin and inter-basin modes. It ships a global MERIT Hydro basin index. It auto-downloads to ~/.hydromt_data/ with no account. Its catalog entries carry source_license, source_url and paper_doi.

Where it stops: the account-free catalog is a sample. artifact_data covers the Piave basin in northern Italy and is what you get when no catalog is supplied. The catalog with global coverage is deltares_data, and the documentation states it is "only accessible when connected to the Deltares network." Everyone outside Deltares assembles and hosts their own, which is exactly the work basinkit removes.

Two smaller differences: get_rasterdataset clips by extent with a buffer, not by polygon mask; and hydromt is a model-building toolchain whose purpose is producing wflow and SFINCS inputs, not a library for getting a basin's data.

hydromt has moved on since this page was first written: at 1.4.1 it ships more predefined catalogs than artifact_data and deltares_data -- aws_data, gcs_cmip6_data and earthdatahub_data. Those widen the climate and cloud- hosted coverage rather than replacing the internal global catalog, and Earth Data Hub is credentialed, so the account-free gap still stands. It is the neighbour to keep re-checking, because it is the one that could close it.

rabpro: same scope, two accounts

VeinsOfTheEarth/rabpro, JOSS 2022. Global, and it does mask to the basin. But its docs require a Google Earth Engine account and a MERIT-Hydro username and password, it asks you to upload your basin to Earth Engine as an asset, and it returns zonal statistics rather than the data. (Checked again in August 2026: rabpro is on PyPI at 0.2.2. An earlier draft of this page said it had been pulled, which is no longer true.)

watershed-workflow: the same pipeline, one continent

environmental-modeling-workflows/watershed-workflow. Delineation plus automatic download of DEM, hydrography, land cover and soils, colocated on the watershed, largely without accounts. This is basinkit's pipeline, done well, and explicitly scoped to "the conterminous US (and most of Alaska/Hawaii/Puerto Rico)."

It is the strongest evidence that the design is right, and the clearest proof that nobody had done it globally.

Global Watersheds: the conceptual wound

mghydro.com/watersheds. Click a point anywhere, get a basin and a report drawing on population, land cover, WorldClim, GLEAM, GRACE, irrigation and dams. Global, account-free, multi-dataset, no install at all.

It is not a package, and it returns summary statistics and vector geometry rather than gridded arrays: the boundaries and rivers download as GeoPackage, not the data. But it is the honest answer to "why does this need to be a library?", and the answer is: because a report is not an input to a model.

Disclose the dependency. basinkit's api backend calls this service. It is never the default, it is recorded in provenance, and its MERIT-Hydro lineage makes it non-commercial, but a reviewer will find it, so say it first.

Also worth naming

Caravan (Kratzert et al., Sci. Data 2023; GRDC-Caravan, ESSD 2025) is a dataset and its derivation code, not a tool. It requires catchment boundaries as input, runs on Earth Engine, and yields basin-mean forcing for a fixed set of gauged catchments. Caravan starts where basinkit finishes.

delineator and upstream-delineator (Heberger): global, MERIT-Hydro based, accurate, account-free, and delineation only. No EO data at all.

HyRiver: the model for how this should feel as an API, and US-only.

pysheds, pyflwdir, WhiteboxTools: routing algorithms. You supply the DEM.

AquaFetch (JOSS 2025) and HARBOR (Copernicus preprint, 2026): the same idea, unified retrieval of basin data, published under other names. AquaFetch harmonises around seventy datasets but for pre-defined gauged catchments, not an arbitrary coordinate; HARBOR collates river-basin attributes into one repository with a toolkit. Neither returns polygon-masked rasters for a point you choose. They are the closest published statements of the concept and belong in any introduction that claims it is unaddressed.

Sen Hydro - Watershed Delineation (QGIS plugin 5713): the neighbour of basinkit's plugin, not of the library. It delineates from a clicked point using the same mghydro Global Watersheds API that backs basinkit's api backend, and adds river styling and a Senegal boundary layer. Anyone comparing the two QGIS plugins will see the delineation overlap immediately, so the plugin description has to lead with what Sen Hydro does not do: no EO layers, no morphometry, no offline backend.

Morphometry has its own neighbourhood, and it is older

The claim above is about the data chain. morphometry() is a separate feature and it has separate incumbents, older and better established than the ones listed so far. Saying morphometry is unpackaged would be false.

GRASS GIS r.stream.stats (Jasiewicz & Metz, Computers & Geosciences 2011) is the real incumbent. It has computed the Horton set (stream counts and lengths by order, bifurcation and length ratios, drainage density) since 2011, and it counts Strahler streams rather than dataset reaches, which is the part most re-implementations get wrong. It is a raster tool: you give it a flow-direction raster and a stream network, not a coordinate.

QGIS ships two in its official repository. ArcGeek Calculator includes watershed morphometric analysis with Strahler ordering. Drainage Basin Geomorphology (plugin 4004, at 2.1.0 in August 2026) covers forty-odd parameters. A separate Hypsometric Curve plugin (3659) computes the curve and the integral from a DEM and a basin polygon.

ArcGIS has RivEX, and several published toolboxes besides.

So the honest statement is narrower than "nobody does this":

Morphometry is well served in the desktop-GIS and raster-tool world and has been for over a decade. What has no implementation is a Python library that returns the parameters from a coordinate, with the network and the basin fetched for you.

That is worth having, and it is not a scientific contribution; it is a packaging one. Claim it as such.

The one part that is not packaging

Strahler ordering constrains the counts it produces, and the constraints are checkable. An order-u+1 stream exists only where two order-u streams meet, so N(u) ≥ 2·N(u+1), and therefore Rb ≥ 2: always, for every basin, with no exceptions and no tuning. A single-outlet basin has exactly one stream of its highest order for the same reason.

Neither GRASS, nor the QGIS plugins, nor any Python code found so far reports these as checks. They compute Rb and hand it over. basinkit computes Rb and then tells you when the number it just produced cannot be true.

This matters because the error is in print. A PLOS One paper from September 2025 reports a mean bifurcation ratio of 1.85, individual sub-watershed values of 1.0 and 1.5, and five third-order streams in a third-order basin. All of those are impossible, all of them are consistent with counting reaches, and all of them would have been caught by a two-line check.

One caution to state before anyone else does. Kirchner (1993, Geology) showed that Horton's laws hold for essentially any branching network, random ones included, so bifurcation ratios carry far less geomorphic information than the literature built on them assumes. That weakens the case for interpreting Rb. It strengthens the case for screening it: if the ratio is near-inevitable, a value below the floor is not an unusual basin, it is a broken calculation.

The attack, and the honest answer

"This is glue. Delineation from a point is solved. Masking is one call to rioxarray.clip. Fetching is HTTP. watershed-workflow already ships this for the US and hydromt already ships delineate-from-point with a licence-annotated catalog. You combined five existing pieces and called it novel."

The attack is right about the components and wrong about the claim. No component here is new and none is claimed to be. What is new is that the combination has never been shipped under the account-free constraint at global scope, and that constraint is not cosmetic. It is what rules out Earth Engine, Earthdata Login and the Climate Data Store, and therefore forces a different dataset selection and a different fetcher design than every prior tool. The nineteen sources basinkit can reach are nineteen because of it.

Claim integration and the removal of access barriers. Do not claim an algorithm.

One claim to narrow, not drop

Machine-readable licence metadata is prior art: hydromt's catalog carries source_license, and STAC mandates a license field on every collection. "basinkit tracks licences" is not defensible.

What is defensible is that the metadata gates behaviour rather than documenting it:

>>> bk.Basin.check_license("merit_hydro", commercial=True)
LicenseError: MERIT Hydro v1.0.1 is licensed CC BY-NC 4.0 or ODbL 1.0,
which forbids commercial use.

>>> basinkit.catalog.require("glofas")
NotImplementedSource: GloFAS v4 (CEMS-Floods) is catalogued but basinkit
cannot fetch it yet.
    Access : CEMS Early Warning Data Store (ewds.climate.copernicus.eu/api)
    Licence: CEMS-Floods licence

A test asserts that every layer in the default stack is anonymous, commercially usable, redistributable and implemented. That is the defensible sentence: the licence table is executable, and it fails the build rather than the user.