GDAL vs pktools: command comparison

Comparison of the commands in the GDAL programs index (https://gdal.org/en/stable/programs/index.html) and the pktools available tools (https://pktools.nongnu.org/html/md_apps.html). Raster first, then vector.

What was read

• pktools: the full apps page, and the man page for 26 of the 37 tools. These cover the raster tools, the vector tools, pkann, and pkcomposite (via its pktools page).

• GDAL: the full programs index, with the description of every command, plus the full sub pages for gdal raster neighbors and gdal raster zonal-stats.

• Not opened: the other GDAL sub pages, and the pktools pages for pkannogr, pksvm, pksvmogr, pkoptsvm, pkfsann, pkfssvm, pkregann, pkascii2img, pkegcs, pkfilterascii and pkstatascii. For these, the one-line descriptions from the index pages are used. Cells based only on that are marked †.

Notes:

• The new gdal <command> interface (GDAL 3.11+) is provisional. Both the new and the traditional command names are given.

• gdal raster neighbors and gdal raster zonal-stats were added in GDAL 3.12.

• pktools support is now limited, and the author recommends the Python library pyjeo instead.

---

Part 1: Raster

1A. Operations in common

Operation	GDAL (new CLI / traditional)	pktools	How they differ
Dataset info	gdal raster info / gdalinfo	pkinfo	pkinfo prints only the items you request, in a form other pktools commands can take through shell substitution. gdalinfo prints everything at once.
Pixel value at a location	gdal raster pixel-info / gdallocationinfo	pkinfo (-r -x -y)	Same result. pkinfo can also print the filename of images that cover a coordinate (-cover).
Crop, band selection	gdal raster clip, select / gdal_translate	pkcrop	pkcrop crops by corners, by centre plus size, or by a vector extent. It also selects and reorders bands.
Stack bands	gdal raster stack / gdalbuildvrt -separate	pkcrop (several -i)	Both do it. GDAL can write a virtual (VRT) result.
Mosaic	gdal raster mosaic / gdal_merge, gdalbuildvrt	pkcomposite	Mosaicking is common. The compositing rules are pktools-only (see 1B).
Change resolution / resample	gdal raster resize, gdalwarp	pkcrop, pkcomposite (-dx -dy -r)	pkcrop lists only nearest-neighbour and bilinear. GDAL’s resampling options are broader.
Data type change	gdal raster set-type / gdal_translate -ot	-ot on nearly every pk tool	Both do it.
Rescale values	gdal raster scale, unscale / gdal_translate -scale	pkcrop (-scale, -off, -autoscale)	Both do it.
Assign or override CRS	gdal raster edit / gdal_edit	-a_srs on pkcrop, pklas2img, pkkalman	pktools can only override the CRS, not reproject.
Output format	gdal raster convert / gdal_translate	-of on most pk tools	pktools has no separate conversion command.
Compare two rasters	gdal raster compare / gdalcompare	pkdiff	pkdiff does a pixel-by-pixel comparison and can write a map of equal, greater and smaller.
Fill nodata holes	gdal raster fill-nodata / gdal_fillnodata	pkfillnodata	Same approach: a search distance and smoothing passes. pkfillnodata takes a mask raster where 0 marks pixels to fill.
Raster to polygons	gdal raster polygonize / gdal_polygonize	pkpolygonize	Both support a mask band.
Sieve small clumps	gdal raster sieve / gdal_sieve	pksieve	Both support 4- or 8-connectivity. pksieve merges small objects into the largest neighbour.
Reclassify pixel values	gdal raster reclassify (classic GDAL: gdal_calc)	pkreclass	pkreclass takes from/to lists or a two-column recode file.
Masks	gdal raster calc / gdal_calc, gdal raster nodata-to-alpha	pkgetmask, pksetmask	GDAL does it through general raster algebra. pktools has dedicated tools. pksetmask uses <, =, > and ! operators and can apply several masks.
Colour tables	gdalattachpct, pct2rgb, rgb2pct, gdal raster rgb-to-palette	pkcreatect	pkcreatect attaches an ASCII table or a min–max ramp, supports greyscale, writes a PNG legend, and can remove a table.
Neighbourhood filters	gdal raster neighbors	pkfilter	Overlap: window mean, sum, min, max, stdev, median, mode, and edge-detection kernels. GDAL has custom convolution matrices, gaussian, sharpen and unsharp-masking. pktools goes beyond this (see 1B).
Raster to text	gdal2xyz	pkdumpimg	pkdumpimg writes a matrix or x y z lines.
Text to raster	XYZ driver, gdal_grid	pkascii2img †	Via drivers in GDAL, a dedicated tool in pktools.
Statistics at samples or zones	gdal raster zonal-stats, as-features	pkextractimg, pkextractogr	GDAL zonal-stats is the richer one, with a weighting raster, fractional pixel coverage and about 20 statistics. pktools is built for training-sample extraction (see 1B).
Raster statistics	gdal raster info (stats)	pkstat	pkstat computes mean, median, variance, skewness, kurtosis, histograms and KDE, correlation, RMSE and regression between two rasters.
Terrain-related	gdal raster hillshade, slope, aspect, roughness, tpi, tri / gdaldem	pkfilterdem, pkdsm2shadow	Related, not equivalent. pktools filters DEMs and computes cast shadows. GDAL derives terrain indices.

1B. Raster: unique to pktools

Tool	What it does
pkcomposite (compositing)	Resolves overlapping pixels by rule: overwrite, maxndvi, maxband, minband, mean, stdev, median, mode, sum, minallbands, maxallbands. Its own page says GDAL does not support a composite step.
pkfilter (beyond the common part)	Morphological dilate, erode, open and close. Sobel edge detection, Markov random field, discrete wavelets, Savitzky-Golay, percentile, circular kernels, user filter taps, and filtering along the band (spectral or temporal) axis. This includes spectral response functions and nodata interpolation over time.
pkstatprofile	Per-pixel statistics along a temporal or spectral profile (mean, median, var, min, max, mode, percentile, proportion, count of valid observations).
pkkalman	Kalman-filter data assimilation: fills gaps in a fine-resolution time series using a coarse-resolution model series.
pklas2img	Rasterizes LAS/LAZ point clouds (height, intensity, scan angle, return number), with per-cell rules, return and class filters, and a percentile height profile.
pkfilterdem	Progressive morphological filter to derive a terrain model from a surface model.
pkdsm2shadow	Binary sun-shadow mask from a surface model and sun zenith and azimuth angles.
pkdiff (accuracy mode)	Validates a classified raster against reference points and prints a confusion matrix.
pkextractimg, pkextractogr (sampling)	Training-sample extraction with random or grid sampling, per-class thresholds, and polygon rules such as mean, mode, proportion, count and percentile.
pkann, pksvm †, pkoptsvm †, pkfsann †, pkfssvm †, pkregann †	Neural-network and SVM classification, parameter optimisation, feature selection, and neural-network regression. pkann was read in full: it is built on the FANN library, supports cross-validation and bagging, and takes training samples from a vector file.
pkegcs †	Utility for rasters in the European Grid Coordinate System.

1C. Raster: unique to GDAL

Family	Commands
Reprojection and georeferencing	gdal raster reproject, gdalwarp, gdalmove, gdaltransform, gdalsrsinfo
Create and edit	gdal raster create, gdal_create, gdal raster update
Overviews	gdal raster overview add / delete / refresh, gdaladdo
Tiling	gdal raster tile, gdal2tiles, gdal_retile
Indexing, virtual datasets	gdal raster index, gdaltindex, gdal driver gti create, gdalbuildvrt, gdal raster materialize
Terrain and visibility	gdal raster viewshed, gdal_viewshed, gdaldem
Distance and extent	gdal raster proximity, gdal_proximity, gdal raster footprint, gdal_footprint
Image fusion and display	gdal raster pansharpen, gdal_pansharpen, gdal raster blend, color-map, gdalenhance
Border cleaning	gdal raster clean-collar, nearblack
Raster algebra	gdal raster calc, gdal_calc
Vector and raster crossing	gdal vector rasterize, gdal_rasterize, gdal vector grid, gdal_grid, gdal raster contour, gdal_contour
Pipelines	gdal pipeline, gdal raster pipeline (with read and write), gdal external, .gdalg files
Multidimensional data	gdal mdim info / convert / mosaic, gdalmdiminfo, gdalmdimtranslate
File and system management	gdal dataset *, gdal vsi *, gdalmanage, gdal-config, sozip, and driver tools (COG, GPKG, OpenFileGDB, Parquet, PDF)

---

Part 2: Vector

pktools has only seven vector-oriented tools. GDAL has about 48 gdal vector commands plus the ogr* programs.

2A. Vector: in common

Operation	GDAL	pktools	How they differ
Dump vector to text	gdal vector info, convert (CSV), ogrinfo, ogr2ogr	pkdumpogr	pkdumpogr dumps all or selected attributes, optionally with x and y, and can transpose the output.
Text to vector	gdal vector make-point plus the CSV/VRT drivers, ogr2ogr	pkascii2ogr	pkascii2ogr makes points or a single polygon from text columns. Its own page says virtual vector datasets are a better alternative.
Recode attribute values	gdal vector sql, ogr2ogr -sql	pkreclassogr	pktools has a dedicated from/to recode. GDAL does it through SQL.
Attribute statistics	gdal vector sql (aggregate SQL)	pkstatogr	pkstatogr gives min, max, mean, median, stdev, histogram and KDE for a field. GDAL has no dedicated command.
Raster values at features	gdal raster zonal-stats, gdal raster pixel-info	pkextractogr	See 1A.
Raster to vector	gdal raster polygonize	pkpolygonize	See 1A.

2B. Vector: unique to pktools

Tool	What it does
pkannogr †, pksvmogr †	Classify features in a vector dataset using a neural network or an SVM.
pkdiff (vector reference)	Accuracy assessment of a raster map against reference points.

2C. Vector: unique to GDAL

Family	Commands
Info, convert, create	gdal vector info, convert, create, edit, export-schema, ogrinfo, ogr2ogr
Geometry processing	gdal vector buffer, concave-hull, convex-hull, simplify, simplify-coverage, segmentize, swap-xy, combine, explode-collections
Validity and topology	gdal vector check-geometry, make-valid, check-coverage, clean-coverage
Overlay and clip	gdal vector clip, layer-algebra, dissolve, ogr_layer_algebra
Reproject	gdal vector reproject
Fields and layers	gdal vector select, filter, sort, sql, set-field-type, set-geom-type, rename-layer
Combine and update	gdal vector concat, update, partition, ogrmerge
Indexing	gdal vector index, ogrtindex, gdal vector materialize
Pipelines	gdal vector pipeline (with read and write)
Linear referencing and networks	ogrlineref, gnmmanage, gnmanalyse

---

Summary

• Raster: about 23 operations overlap. Most are straightforward equivalents (info, crop, fill, polygonize, sieve, reclassify, compare). The mosaic, filtering, statistics and extraction rows overlap only in part.

• pktools-only raster work: compositing rules, spectral and temporal filtering, Kalman assimilation, LAS rasterization, DEM and shadow tools, accuracy assessment, and machine-learning classification.

• GDAL-only raster work: reprojection, tiling, overviews, terrain derivatives, pansharpening, pipelines, multidimensional data, and file management.

• Vector: pktools covers a small attribute- and sample-oriented subset. Everything about geometry, topology, reprojection, merging and SQL is GDAL-only, and the only pktools-only vector work is classification and accuracy assessment.