1.1.2 - QGIS 4: the plugin repository's Qt6 checker flagged two
unscoped enums (QgsColorRampShader.Interpolated) in the raster
styling; now resolved through the version-agnostic helper, and the
static compatibility test covers QGIS enum classes as well as Qt ones.
No functional change.
1.1.1 - Release hygiene for the QGIS plugin repository: its Bandit
scan blocked 1.1.0 for eight try/except/pass blocks; every swallowed
exception now leaves a trace in the QGIS message log, unused imports
were pruned, and the same Bandit and Flake8 scans run in CI so this
cannot happen again. metadata.txt no longer contains percent signs,
which the repository's parser rejects. No functional change.
1.1.0 - Graded river crossings. Not every watercourse is impassable:
many trails ford shallow streams or cross on stones. The route now
pays a cost that grows with the size of the stream, using the
contributing area of the channel as a proxy for its size (hydraulic
geometry, Leopold and Maddock 1953; Faustini et al. 2009): headwater
streams below 2 km2 cost 2x, streams 2-10 km2 cost 4x, small rivers up
to the fordable ceiling cost 8x, and above the ceiling (new parameter
'Largest fordable drainage basin', default 50 km2) the river is a
barrier. Zones still exclude the drainage strip. Every crossing the
route makes is written to a new output layer of points with
contributing area, class, cost factor and a field-check warning, and
is loaded into the project with the other products. The DEM knows the
size of the basin, not today's depth or current, so the safety of a
crossing is never asserted -- it is listed to be verified.
1.0.0 - First stable release. End-to-end battery on three regions with
very different relief -- Serra da Mantiqueira (real 1-arcsecond DEM,
the Marins-Marinzinho-Itaguare traverse against the hiker's own GPS
track and summit waypoints), Ceara (real Copernicus GLO-90, field GPS
tracks, park polygon) and a Himalayan slope regime -- plus polar
latitude, Web Mercator and memory-layer cases: 14 of 14 pass, and the
full window was driven as a user would in all six languages. The
battery found one real problem, fixed here: with the window's
defaults, watercourses derived from the DEM entered the route as an
absolute wall, so any traverse that had to cross a valley failed with
'could not connect'. Watercourses are now always a cost for the route
(crossable at 8x) while zones still exclude them in 'avoid' mode; the
user's own constraint layer keeps the chosen mode. The no-path error
now lists the likely causes. Docs: docs/BATERIA_TESTES.md,
validation/bateria_regioes.py.
0.14.0 - Three independent adversarial audits (mathematics, geography,
languages) before publication, and the fixes they demanded.
Mathematics: cells next to NoData were differentiated against the
scene mean, which invented slopes of hundreds of percent along every
reprojected DEM's edge -- derivatives now use one-sided differences at
the data edge (slope, curvature, VRM, TWI); criteria that do not cover
a cell (extra raster, wetness, roughness) no longer count as the worst
score but are left out with per-cell weight renormalisation; the
stream buffer is circular; curvatures are now correctly cited (Moore
et al. 1991 / Mitasova and Hofierka 1993, not Zevenbergen-Thorne);
degenerate rasters get a clear message. Geography: DEMs in a projected
CRS that does not measure in metres (State Plane feet, Web Mercator)
are reprojected to the automatic UTM instead of being read as metres;
rotated or rectangular-pixel grids are resampled to a square north-up
grid; constraint layers come through QGIS features, so GeoPackage
layers opened from the browser and memory layers work; custom output
CRSs (USER:nnnn, no authority) work through WKT; the cell of a point
is found with floor, not round; a point outside the DEM says so; UTM
zone is right for 0-360 longitudes and a warning covers latitudes
above 84; KML explains it is always WGS84; rasters inside a GeoPackage
are accepted. Languages: the Processing help text, the step-4 summary,
the map-pick prompts, file-dialog titles and filters, the layer names
loaded into the project and the not-registered error were still
Portuguese or English in the other languages -- all keyed now; combo
boxes resize with the language; switching language no longer resets
the cost model; locale strings like zh.UTF-8 detect correctly; the
Chinese entry says Simplified; translation fixes in es, fr, zh, ja
from a fluent review. Nineteen regression tests added.
0.13.1 - Interface pass. Qt silently ignores fractional pixel font
sizes, and seventeen of the window's type rules used them, so most
text was rendering at the QGIS default instead of the designed scale;
every size is now a whole pixel and the hierarchy finally reads as
drawn. Header is one row shorter (About and version join the step
row). Included products no longer look disabled: accent icon, dashed
frame and an "Included" pill. Four form labels and one group subtitle
that were written once at build time now follow the language switch,
and the Qt6 harness checks that nothing stays behind in another
language.
0.13.0 - Zero external dependencies. Every remaining use of geopandas
and shapely (zone vectorisation, vector export, route and corridor
construction) now goes through GDAL/OGR, which QGIS itself ships on
every platform: install the plugin and it runs, with nothing to pip
install. Verified end to end in a headless QGIS with geopandas,
shapely, pandas, fiona and pyproj blocked from import, in GeoPackage,
Shapefile and KML, on projected and geographic DEMs. Along the way,
KML export switched to the LIBKML driver, because the classic KML
driver silently dropped every attribute but the first two. A new test
forbids any module-level import that QGIS does not guarantee, so this
cannot regress.
0.12.2 - Publication-readiness audit, and the start of its most
serious finding: the plugin imported geopandas at module level, which
QGIS does not ship on any platform, so it failed to load on a clean
QGIS install before showing a single window. The constraint layer now
goes through OGR, which QGIS does guarantee; the remaining call sites
follow. Also removed a 902-line module no import ever reached.
0.12.1 - The window is now actually built and exercised under Qt6 --
the toolkit QGIS 4 uses -- in all six languages, and that found a
break the static check had missed: QPainter.Antialiasing is also a
scoped enum in Qt6. The harness runs in continuous integration from
now on, so QGIS 4 support stops being an assumption.
0.12.0 - QGIS 4 compatibility, verified rather than assumed. The
plugin already declared support for QGIS 4, but Qt6 removed unscoped
enum access -- Qt.PointingHandCursor no longer exists -- and 33 such
accesses would have raised as soon as a user clicked. All of them now
go through the version-agnostic helpers the project already had, and a
test resolves every enum name against a real PyQt6 install, which
catches a mistyped group that Qt5 would silently tolerate. The footer
now credits the author alongside the institutional logos, in all six
languages.
0.11.2 - Layout fixes found by measuring every label in all six
languages: no text is clipped any more. The clipping was in the Latin
languages, not the Asian ones -- "Destination" needed 98 px in a slot
fixed at 64, and the active step tab reserved its width in the normal
font weight while rendering in semibold, so it cut the step the user
was actually on. A test now checks all six languages at the minimum
window size, because this defect is silent.
0.11.1 - Audit fixes. The transitability class labels are now
translated as well: they are written into the output raster's legend,
so a Japanese user was opening the map in QGIS and reading a
Portuguese legend -- and the file kept it when passed on. Also removed
an interface file the rewritten window no longer loads, an unused
widget class, and a stray run log, all of which were shipping inside
the plugin package.
0.11.0 - The plugin now speaks six languages: Portuguese, English,
Spanish, French, Chinese and Japanese. It opens in the language QGIS
is running in and remembers a different choice. The strings moved out
of the Python source into one JSON file per language, so a wrong term
can be fixed by whoever speaks the language rather than by whoever
maintains the code, and the Processing toolbox labels went through the
same mechanism -- they previously stayed in Portuguese in every
language, because the Qt .qm files they relied on were never
generated. Spanish, French, Chinese and Japanese are draft
translations and say so in the interface; corrections are welcome.
0.10.0 - The window was restructured around a horizontal stepper
instead of a sidebar, freeing the full width for content, and it now
follows the QGIS theme in light or dark with a manual override.
Choosing a DEM shows what was actually read from it -- format, cell
size, CRS and pixel count -- so a wrong file is caught before the
analysis runs rather than after. Outputs that are always produced now
say so with the word "included" instead of a ticked-and-greyed
checkbox that read as switched off, option cards answer Space and
Enter, and the footer carries a contextual line saying what is missing
or what is selected.
0.9.2 - A visual design pass. The plugin now has its own icon
language: eighteen glyphs stroked with QPainter at the exact size and
colour requested, so they stay sharp at any screen density and
recolour with the state of the control, with no image files and no new
dependency. Each output is presented as a selectable card with icon,
title and one-line description instead of a checkbox in a list, so the
screen can be scanned rather than read; the longer caveats moved into
tooltips. Type scale, spacing and states were rebuilt around the deep
forest green of the plugin's own logo.
0.9.1 - Visual redesign of the assistant, around a sidebar in the deep
forest green of the plugin's own logo, with the project and
institutional logos restored to permanent view rather than hidden in
an About box. Help texts no longer illustrate features with one
country's place names: a tool published for worldwide use has to
explain intermediate destinations as "one summit, then another", not
by naming local peaks. Tests now fail the build if a regional example
or a missing credit creeps back in.
0.9.0 - The interface was rewritten as a four-step assistant. It had
fallen behind the engine: sixteen parameters added since 0.6 --
deriving slope and curvature from the DEM, drainage, Tobler walking
time, transitability breaks, intermediate destinations -- could not be
reached from the window at all, and the transitability map was never
loaded into the project because the dialog looked for an output name
that never existed. Now only the DEM is required, every control
carries a sentence explaining what it changes, options appear only
when asked for, and the run happens in the background so QGIS stays
usable. A contract test now fails the build whenever the window and
the algorithm drift apart again.
0.8.0 - Routes can now visit several destinations in order -- climb
one summit, then the next, then descend -- instead of only a start and
an end point, with optional exact optimisation of the visiting order.
This closed the one case where the plugin lost to a straight line:
declaring the three summits of the Marins-Itaguare traverse moved
agreement with the real GPS track from 2.7 percent to 73.0 percent and
median deviation from 1860 m to 114 m. Route constants were also
recalibrated against route geometry rather than walking speed, which
is what they actually govern: Tobler's published decay of 3.5 is
confirmed, the terrain term is worth 12 points of agreement, and
leave-one-out cross-validation showed the apparently better values do
not generalise, so the defaults stand -- now on evidence.
0.7.0 - First empirical validation against real field GPS: 110 h and
224 km of surveyed tracks in two biomes. Tobler's walking speed is
optimistic by a factor of 1.7-3.1 for field survey work, but the route
is provably unaffected -- only the estimated duration changes -- while
the decay, which does move the route, could not be separated from its
published value and is kept. Transitability labels no longer claim a
walker cannot pass: real teams walked 115.8 percent slope. The class
map now records the cell size and warns when it is coarse, because the
same terrain reads as 48 percent or 76 percent walkable depending on
resolution. Full record in docs/VALIDACAO.md.
0.6.3 - Any raster the user has can now enter the model as an extra
weighted criterion -- stoniness, vegetation cover, a cost surface --
aligned to the DEM grid and normalised by scene percentiles. Weighted
zero by default.
0.6.2 - New output: a transitability class map with absolute,
field-meaningful classes and a colour palette written into the
GeoTIFF, so it opens legible in QGIS with its legend. Slope sets the
class; high ruggedness or high wetness push a cell one class worse;
constraints marked as avoided become class 5.
0.6.1 - Route cost can be estimated walking time using Tobler's
anisotropic hiking function, so uphill and downhill no longer cost the
same and the accumulated cost is in hours. Two optional criteria
added, both derived from the DEM and both weighted zero by default:
topographic wetness index and terrain ruggedness index. Fixes the
reprojection fill entering the analysis as sea-level terrain.
0.6.0 - Only the DEM is required: slope and both curvatures are
derived from it by default, which removes the unit, sign-convention,
grid-alignment and coverage problems in one step. Slope unit and DEM
vertical unit are now explicit parameters. Watercourses can be
extracted from the DEM and used as a constraint, and any vector layer
can be used the same way. Route cost model is selectable. Potential
zones now use the multicriteria model by default. Temporary
directories are cleaned up.
0.5.1 - QGIS 4 / Qt6 compatibility while preserving QGIS 3.22+
support; responsive interface for small screens and high-DPI scaling;
theme-aware styling; experimental flag removed; interface text
encoding corrected.
0.5.0 - English documentation promoted to primary; Portuguese
reference copies kept in docs/pt_BR; PT-BR | ENG language switch;
legacy helper code removed; suitability, relative-risk, route and
corridor workflow preserved.
0.4.0 - Initial public beta.

None