0.9.2
- Every layer now records what it was and when it arrived: product,
licence, the resolution it came back at and a UTC retrieval date. A
product name carries an edition but not a date, and a repeat study
needs
the date to tell its own change apart from the provider reissuing a
tile. Basin report prints the list on its provenance page, and the
full
run writes it into the manifest beside the dataset catalogue.
- HydroRIVERS and HydroLAKES did not name themselves at all, only
their
licence. The elevation source named its route key, "cop30", rather
than
the dataset that answered it. Both now take their name and licence
from
the catalogue.
- Needs basinkit 0.9.2 or newer.

0.9.1
- Basin report ran at ten times the budget 0.9.0 had just set for it.
The
package caps the elevation grid at four megapixels because a larger
one
was killed by the operating system, and inside QGIS that takes QGIS
with
it -- but this algorithm passed a budget of its own, and that default
stood at forty. Measured on the Juniata (8,805 km2) in QGIS 3.34: 245
seconds and 5.61 GB at forty, 26 seconds and 1.26 GB at four, for the
same eight pages within six kilobytes of each other. The larger grid
bought nothing, because every raster a page draws is decimated to
1,600
pixels first.
- Fetch basin data layers asked for a hundred megapixels, about 5.6
GB,
past the ceiling the elevation guard enforces -- so a large basin was
refused outright instead of coming back at a coarser resolution. It
now
asks for sixty, as the three other surface algorithms do.
- Data quality reported a layer it could not measure as a layer that
measured badly. A rainfall correlation that came back as not-a-number
was
graded LIMITED, and through the worst-of rule that one layer set the
whole
basin's grade: a basin whose rainfall pair agrees at r = 0.949 was
reported LIMITED overall instead of MODERATE. An unmeasurable layer
now
stays ungraded and says why.
- The notice shown when QGIS runs a Python too old for the package
said to
update to 3.28, which is the release the reader was already on. It now
names 3.34. The minimum stays 3.28, because the bundled Python belongs
to
the installer rather than to the QGIS version: the Windows and macOS
builds of 3.28 shipped Python 3.9, where pip finds no release of
basinkit
at all, while the Debian and Ubuntu packages of the same QGIS run on
3.11,
where it installs normally.
- Needs basinkit 0.9.1 or newer.

0.9.0
- Delineate river basin takes an optional line layer of known channels
and
forces the flow along them. Where a stream runs in a culvert under a
city,
or a road embankment dams a valley in the elevation model, the surface
carries no channel and the water is sent the wrong way; a line drawn
on
the canvas is enough to correct it. Lowering the cells by a fixed
depth
does nothing -- the trench keeps the terrain's own rises -- so each
line
is walked from its lower end and forced to descend the whole way. The
provenance records that the path was forced, how deep and over how
many
cells, because a basin that was told where to go and one that was
found
are not the same kind of answer.
- A basin that disagrees with the river at its outlet is now reported
instead of printed. The check measured the basin only against the
largest
river within the search radius, which beside a confluence is a
different
and far bigger river: a basin at half of what drains to its own stream
passed, and one at nine times it passed too. It now tests both
directions
against the river at the outlet, and where two channels meet within a
metre of the point it says the outlet is ambiguous and names both
catchments rather than choosing one.
- The TDX-Hydro backend took the reach whose recorded point was
nearest.
Beside a confluence that is a stub: at one outlet it returned 122 km2
of
a 228 km2 basin. It now takes the unit catchment that contains the
outlet, and that basin reads 241 km2, in line with the other backends.
- Basin morphometry and the terrain surfaces no longer fetch a second
elevation grid, so one report cannot print two different values for
the
same basin's highest point.
- Every algorithm's help panel now prints the citation and the DOI.
- The plugin is GPL-3.0-or-later. The basinkit Python package it calls
is
Apache-2.0. Neither licence grants the use of the name basinkit; a
fork
must carry its own.
- The plugin now asks for the packages it actually needs. It told
people
to run "pip install basinkit", which does not install pyflwdir -- and
the DEM backend routes with pyflwdir, which auto uses for every
catchment below 2,000 km2. Those basins came back as the coarser
HydroBASINS answer instead, 21.6 percent high on one 259 km2
catchment,
with nothing on the face of the result to say so. It now names
basinkit[delineate] and basinkit[viz], says what each is for, and
quotes
the extras so the command survives a shell. On Debian and Ubuntu,
where
QGIS runs on a system Python the distribution marks as externally
managed, the command now carries --break-system-packages, without
which
it failed outright.
- Basin report no longer takes QGIS down with it. It read elevation at
the 100 Mpx default, which on a large basin reached about 5.9 GB and
was
killed by the operating system -- and inside QGIS that kills QGIS. The
grid is now budgeted to 4 Mpx at the point every caller arrives, which
on a 54,000 km2 basin is about 186 m and on a 235 km2 one is still its
native 31 m, and the rasters drawn on the pages are decimated to 1,600
pixels because an A4 page at 300 dpi cannot show more. The cover
states
the resolution used. Measured on the large basin: 1.33 GB and eighteen
seconds, against a process that did not finish.
- Delineate river basin declines a window it cannot route, naming the
megapixels, the memory estimate and the backend to use instead, rather
than growing the window until the process is killed with nothing
written.
- The log is readable again: a delineation emitted twenty-six copies
of
"Use @ matmul instead of * mul operator", a deprecation inside affine
that names a file nobody here can change.
- Asking for a backend by name is now checked the way auto checks
itself.
Before, a backend chosen explicitly skipped every check, and a basin
at
one percent of what drains to the point came back looking like any
other
answer.
- Needs basinkit 0.9.0 or newer.

0.8.4
- Landscape form no longer fails on a basin that has knickpoints.
It asked the output sink for its field list, which QgsFeatureSink
does not have, so it only survived basins where there was nothing
to write. Found by running all eleven tools over five small
sub-catchments of the Koshi; three of the five hit it.

0.8.3
- A projected basin layer is now reprojected to EPSG:4326 on the way
in.
basinkit works in lon/lat throughout, so a UTM layer used to arrive as
metres and PROJ rejected a northing of 7,793,981 as a latitude. This
failed in Fetch basin data layers and in Basin morphometry; all seven
algorithms share the same entry point and all seven are fixed.
- Delineate river basin now warns below 2,000 km2 rather than 500, and
says what the coarse polygon is: whole HydroBASINS level-12 units of
about 130 km2, not a routed elevation model, so its boundary follows
unit edges rather than the terrain divide.
- Needs basinkit 0.8.3 or newer.

0.8.0
- New algorithm: Landscape form. chi, normalised channel steepness,
concavity and knickpoints -- whether a river is still adjusting or has
settled, which the shape indices cannot tell you. Knickpoints come
back
as a point layer carrying the height of each step and how far its
local
gradient stands above the profile's own median. Channel steepness was
checked cell by cell against TopoToolbox on two basins and agrees to
within 6 percent at every quantile. The knickpoint count is not that
stable and the algorithm says so; it is not an uplift rate either.
- New algorithm: Export for Arc Hydro. Sub-catchments and reaches as
Catchment and DrainageLine layers carrying HydroID, HydroCode,
NextDownID and AreaSqKm, so Arc Hydro's own tools read them without
conversion. A renaming, not a computation: every value written is one
basinkit already holds, and DrainID is deliberately left for Arc Hydro
to populate rather than guessed with a spatial join.
- Eleven algorithms now. Needs basinkit 0.8.2 or newer.

0.7.0
- New algorithm: Data quality. A grade for every layer, not only for
the
elevation model, each measured against something produced
independently
of it: land cover as the agreement between ESA WorldCover and the ESRI
annual series, rainfall as CHIRPS against TerraClimate, soil as the
width
of SoilGrids' own published interval, surface water as the permanent
to
seasonal split, delineation as the network check and the accuracy
regime
for a basin of this size, and elevation as the suitability grade split
by
terrain class. Suggested by Prof. Dr. B. Mishra.
- Soil and surface water are reported without a grade, because no
published
threshold exists for either; the output says so rather than inventing
a
cut-off. Every threshold that does exist is printed beside the
measurement it judged.
- The ESRI land cover reader now writes an integer nodata value, so a
comparison against another class raster no longer counts nodata as a
class.
0.6.3
- On QGIS builds older than 3.28 the plugin now says so. basinkit
needs
Python 3.10, which those builds do not ship, so pip cannot install it
there at all and the old message ('no module named basinkit') sent
people looking for an installation fault that was not there.
- The install instructions now lead with the QGIS Python Console
route,
which installs into the interpreter QGIS imports from on every
platform, and note that '!pip install' works only in Jupyter.
0.6.2
- Found by running every analysis on the Ganga above Farakka (934,516
square kilometres) and on the Potomac, then checking the results
against GRDC, 190 Nepal gauges, IMD rainfall and USGS records.
- Sentinel-2, Landsat, Sentinel-1 and land cover change failed on very
large basins: the basin outline was too big for the catalogue search.
The search now sends a simplified outline that still contains the
basin; results are clipped to the full outline as before.
- The 3D page ran out of memory on large basins and showed a black
screen
for basins wider than 900 km. Both fixed; the page also declares its
text encoding so symbols render correctly.
- The river profile reads a capped elevation model; on a 30,000 km2
basin it used to need more than 6 GB of memory.
- The 3D page sets its colour balance from the scene, so forested
basins
no longer render nearly black.
- The report's first page names the elevation product it used.
- Needs basinkit 0.6.2.
0.6.1
- 'Basin statistics' reported mean slope over the whole bounding box,
with
the ground outside the basin filled in as a flat plain. On a catchment
filling 43 percent of its box that read 7 degrees where the basin's
own
slopes average 16. It now measures inside the basin only.
- A failed delineation now says why, instead of reporting an
AttributeError about a missing helper.
- The check against the river network measures distance on the ground,
so
it no longer misses rivers inside its search radius at high latitude.
- Needs basinkit 0.6.1, which also requires geopandas 1.0: on installs
with
an older geopandas the first delineation used to fail.
0.6.0
- Four new Processing algorithms. 'Terrain surfaces' computes slope,
aspect, hillshade, curvature, topographic position, ruggedness, local
relief, slope-position classes, flow accumulation, the channel
network,
the wetness index and height above nearest drainage, all from one
elevation download rather than one per surface. 'Sub-catchments and
their
routing' returns the pieces a basin is assembled from, each carrying
NEXT_DOWN, which is the routing graph a distributed model needs
without
inferring anything from geometry. 'Elevation data suitability' grades
whether the elevation model can carry the terrain analysis at all.
'Basin report' writes eight A4 pages ready for a thesis or a
manuscript.
- The elevation data suitability grade is the one worth reading first.
Five measurements against stated thresholds -- how much surface
depression filling invented, how many slopes fall below the angle at
which a gradient is only the model's own error, relief against that
error, the largest level surface, and cells with no elevation --
combined
into HIGH, MODERATE or LIMITED with a per-cell map of which ground the
answer rests on. Tested on 133 gauges against an independently
produced
elevation model: where it says HIGH the two models agree about the
slope
field at r = 0.94, where it says LIMITED at r = 0.76.
- A twelve-metre delineation backend. 'tdx' walks a reach-level graph
built
from TanDEM-X, one catchment per stream reach instead of one per 130
square kilometre unit. On 360 gauges between 100 and 500 square
kilometres it cut the median area error from 32.5 percent to 10.5
percent
and was the better answer on two thirds of them. It is never chosen
automatically because TDX-Hydro is CC BY-SA, and ShareAlike travels
into
anything derived from it and redistributed.
- Cell size is taken per row from that row's latitude everywhere in
the
terrain code. Dividing by one assumed width reports a basin at 60
degrees
north as about twice as steep as it is.

0.5.0
- Compatibility across the full supported QGIS range. The provider now
resolves the Processing source-type enum by asking the API for it, so
it
loads identically on QGIS 3.28 through 4.x. QGIS 3.34, the current
long-term release, is covered by a continuous integration job that
installs QGIS and loads every algorithm.
- Accuracy is now reported by catchment size, from blind validation
against
2,550 gauges in 99 countries whose catchment areas are published by
the
agencies that operate them: 0.3 percent median error above
100,000 square kilometres, 1.3 percent from 10,000 to 100,000, 1.7
percent from 2,000 to 10,000. Below about 2,000 square kilometres the
'dem' backend is the better instrument, and the plugin now says so at
the
moment it matters.
- Every answer is checked against the river network. When the basin
exceeds
twice the area draining to the largest river within two kilometres of
the
outlet, the plugin reports it: 85 percent of the outlets it flags need
attention, and it stays quiet on 97 percent of the ones that do not.
Precision runs from 75 percent in Europe to 93 percent in Africa.
- The 'auto' backend refines on the elevation model when the river
network
and a 30 metre routing of the same point agree on a smaller catchment.
Across 300 gauges drawn after the method was fixed and used nowhere
else
in its design, it improved 19 results, left 279 unchanged and reduced
none. Where it acts, the median error falls from 1,731 percent to 4.3
percent, at a median cost of 1.6 seconds.
- Refinement acts only when the river it is judging by lies within a
kilometre of the outlet. Beyond that distance the point is not on that
river, so the two sources stop being independent. Every refinement
that
improved a result had its river within 0.91 kilometres.
- The plugin now names what kind of place the outlet is in. An
endorheic
system, a coastal strip draining straight to the ocean, and an outlet
with nothing draining into it are each reported, the last being what a
lake surface and a sub-grid headwater have in common.
- Messages are matched to the cause and say what to do next: a
coordinate
outside the network's coverage, one beyond the snap distance, and one
on
a hillslope rather than in a channel each get their own guidance.
- The consistency line now compares the result against the river
network
rather than against the source dataset's own upstream-area field,
which
confirms the traversal rather than the choice of outlet.
- The river-network check is a checkbox, on by default, and is the
only
thing that downloads the regional rivers file.
- Against pysheds and WhiteboxTools on identical 30 metre rasters
across 59
catchments under 2,000 square kilometres, the 'dem' backend returned a
basin for every one, matched the agency figure within 20 percent on 68
percent of them against 56 and 51, and had the lowest median error.
- Requires basinkit 0.5 or newer.

0.4.0
- New algorithm: Basin morphometry. The full Horton-Strahler-Schumm
set
from one run, with the per-order table showing Strahler streams and
dataset reaches side by side, and the counts tested against what
Strahler
ordering allows. A bifurcation ratio below 2 and more than one stream
of
the basin's highest order are both impossible, and are now reported
instead of being handed back as numbers to interpret. Until now
morphometry was in the Python package but not reachable from QGIS.
- The plugin description leads with the data package rather than with
"delineate". QGIS already has
several delineation plugins, and the old wording put this one in that
queue while morphometry, which is the part with no equivalent
elsewhere,
went unmentioned.
- Requires basinkit 0.4 or newer.

0.3.3
- Pairs with basinkit 0.3.3, which corrects channel_gradient_m_per_km.
Up to
0.3.0 it divided total basin relief by main channel length, which
reports
a fall the river never makes, because the highest point in a basin is
a
ridge top and not the head of the main stem. It is now read from the
bed:
the DEM is sampled along the main channel and the drop between its
ends is
divided by the channel length. On the Koshi that is 9.6 m/km, not
15.1.
The Basin statistics algorithm never reported this figure, so nothing
in
the plugin's own output changes; it matters if you called
morphometry()
from the Python Console or a script, where the old value should be
recomputed.
- Basin.morphometry() now checks its own stream counts. Two of the
checks
are impossibilities rather than oddities: an order-u+1 stream is
formed
where two order-u streams meet, so N(u) >= 2*N(u+1) and the
bifurcation
ratio can never fall below 2, and a basin with one outlet has exactly
one
stream of its highest order. Counts that break either are reported
rather
than returned as numbers.
- The plugin description now says plainly what this does that a
delineate-from-a-click plugin offers, rather than leaving a reader to
work it out.

0.3.2
- Declares its pip dependency in the external_deps metadata field, so
it
shows in the External dependencies box on the plugin page.

0.3.1
- Fixed 16 Qt6 compatibility issues found by the plugin repository's
own
checker: every Qt and QGIS enum is now spelled with its full scope
(QgsWkbTypes.Type.Polygon rather than QgsWkbTypes.Polygon, and so on).
The scoped spelling works on both PyQt5 and PyQt6, so QGIS 3.28
onwards
and QGIS 4 are both covered by one code path, with no version guard.
- A test now scans the plugin for unscoped enums, so they cannot come
back.

0.3.0
The plugin version now tracks the basinkit package version, so it is
clear
which plugin goes with which package. No algorithm has changed since
0.1.0.
- Fixed: on macOS the "package missing" notice printed the QGIS
application
binary instead of an interpreter, so the suggested pip command opened
a
second QGIS window and installed nothing. A path is now offered only
once
it is verified to be a python executable, and there is a Python
Console
snippet that needs no path at all
- New icon: the b is drawn as a river basin with a dendritic stream
network
draining to a single outlet
- Pairs with basinkit 0.3.0, which adds Basin.morphometry() -- the
classical
Horton-Strahler-Schumm parameters -- to the Python API
0.1.0
First release.
- Delineate a basin from a canvas click, globally, via HydroBASINS
graph traversal or DEM flow routing
- Fetch DEM, land cover, soil, surface water, rivers and lakes clipped
to the basin
- Basin statistics: area, relief, mean slope, land cover fractions,
bounding-box efficiency
- Every algorithm runs in the Processing framework, so batch mode and
the Model Builder work

None