PGC Imagery Utils is a collection of commercial satellite imagery manipulation tools to handle batch processing of Geoeye and DigitalGlobe/Maxar imagery. The tools can:
- Correct for terrain and radiometry
- Mosaic several images into one set of tiles
- Pansharpen a multispectral image with its panchromatic partner
- Calculate an NDVI raster from a multispectral image.
These tools are build on the GDAL/OGR image processing API using Python. The code is built to run primarily on a Linux HPC cluster running PBS or Slurm for queue management. Some of the tools will work on a Windows platform.
The code is tightly coupled to the systems on which it was developed. You should have no expectation of it running perfectly on another system without some patching.
Files starting with "qsub" and "slurm" are PBS and SLURM submission scripts. See the script-specific documentation for more details on usage.
The orthorectification script can correct for terrain displacement and radiometric settings as well as alter the bit depth of the imagery. Using the --pbs or --slurm options will submit the jobs to a job scheduler. Alternatively, using the --parallel-processes option will instruct the script to run multiple tasks in parallel. Using --threads N will enable threading for gdalwarp, where N is the number of threads (or ALL_CPUS); this option will not work with --pbs/--slurm, and (threads * parallel processes) cannot exceed number of threads available on system.
Example:
python pgc_ortho.py --epsg 3031 --dem DEM.tif --format GTiff --stretch ns --outtype UInt16 input_dir output dir
This example will take all the nitf or tif files in the input_dir and orthorectify them using DEM.tif. The output files will be written to output_dir and be 16 bit (like the original image) GeoTiffs with no stretch applied with a spatial reference of EPSG 3031, or Antarctic Polar Stereographic -71.
The mosaicking toolset mosaics multiple input images into a set of non-overlapping output tile images. It can sort the images according to several factors including cloud cover, sun elevation angle, off-nadir angle, probability of overexposure, and proximity to a specific date. It consists of 3 scripts:
- pgc_mosaic.py - initializes the output mosaic, creates cutlines, and run the subtile processes.
- pgc_mosaic_query_index.py - takes mosaic parameters and a shapefile index and determines which images will contribute to the resulting mosaic. The resulting list can be used to reduce the number of images that are run through the orthorectification script to those that will be eventually used.
- pgc_mosaic_build_tile.py - builds an individual mosaic tile. This script is invoked by pgc_mosaic.
Example:
python pgc_mosaic.py --slurm --bands 1 --tilesize 20000 20000 --resolution 0.5 0.5 input_dir output_mosaic_name
This example will evaluate all the 1-band images in input_dir and sort them according to their quality score. It will submit a job to the cluster queue to build each tile of size 20,000 x 20,000 pixels at 0.5 meters resolution. The output tiles will be Geotiffs named by appending a row and column identifier to the output_mosaic_name.
The pansharpening utility applies the orthorectification process to both the pan and multi image in a pair and then pansharpens them using the GDAL tool gdal_pansharpen. GDAL 2.1+ is required for this tool to function. The --threads flag will apply threading to both gdalwarp and gdal_pansharpen operations.
The NDVI utility calculates NDVI from multispectral image(s). The tool is designed to run on data that have already been run through the pgc_ortho utility.
stack_landsat.py is a command line tool to combine individual Landsat band .tif files into a stacked RGB composite .tif. To run, set the input directory to a folder with the downloaded Landsat imagery you want to combine, with each of the bands as a separate .tiff file. The script will need to be run within the same environment as the other PGC utilities in this repo; it only uses standard python and gdal functionality, so there is nothing further to install.
Show tool help text:
python C:\path\to\stack_landsat.py -h
Example usage with long options:
python C:\path\to\stack_landsat.py --input-dir C:\path\to\landsat\directory --output-dir C:\path\to\output\dir
Example usage with short options:
python C:\path\to\stack_landsat.py -i C:\path\to\landsat\directory -o C:\path\to\output\dir
The script will:
- Verify that the provided input directory exists and is, in fact, a directory
- Create the output directory if it does not already exist
- Find all the Landsat scenes in the input directory
- Attempt to create a composite RGB TIF of the scenes it finds
- Report the scenes it fails to build. For instance, an RGB TIF will not be built if all of bands 4, 3, and 2 do not exist
- Write the console messages to a log file in the input directory (stack_landsat_{date}.log). There is no need to retain the logs long term if the script is operating smoothly
The script will not:
- Know anything about previous runs. If you rerun the script, it will process whatever inputs are present, even if they have been run previously. It will also overwrite any corresponding outputs if pointed to the same output directory.
PGC uses the Miniforge installer to build our Python/GDAL software stack. You can find installers for your OS here: https://github.com/conda-forge/miniforge?tab=readme-ov-file#miniforge3
Users should expect a recent (less than 1-2 years old) version of Python and GDAL to be compatible with tools in this repo. The following conda/mamba environment likely contains more dependencies than are needed for tools in this repo, but should suffice:
mamba create --name pgc -c conda-forge git python=3.11 gdal=3.6.4 globus-sdk globus-cli numpy scipy pandas geopandas rasterio shapely postgresql psycopg2 sqlalchemy configargparse lxml pathlib2 python-dateutil pytest rtree xlsxwriter tqdm alive-progress pyperclip --yes
Most unit tests and functional tests for imagery-utils are written using python's unittest library. They use licensed commercial data that cannot be distributed freely but is available to project contributors.
On Linux systems:
# first time only
ln -s <test_data_location>/tests/testdata tests/
# run the tests
python tests/unit_test_ortho_functions.py
python tests/unit_test_taskhandler.py
etc... One test is written using pytest:
# run the test
pytest tests/test_ortho_functions_nodata.pyTo report any questions or issues, please open a github issue or contact the Polar Geospatial Center: pgc-support@umn.edu