The points2grid application was extended with MPI to allow the application to be run in parallel on a compute cluster. The goal was an application that would scale to arbitrarily large input, limited only by the amount of disk space needed to store the input and output files. No intermediate files are generated but the size of the grid output is limited by the sum of the memory of the writer processes. This code was extended from the https://github.com/CRREL/points2grid/ repository which is the most actively developed fork of the original work by OpenTopography.
The algorithm_and_test_results.pdf slide presentation and lidar_parallel_processing.pdf paper contain a description of the implementation and test results on large LAS files run on the Stampede cluster at the University of Texas-Austin.
An MPI implementation must be installed. OpenMPI 1.6 and 1.8 are known to work and were used in development and testing. mpic++ must be found in your PATH. The following must also be installed:
- GDAL 1.9.2 or later with internal BIGTIFF.
- Boost 1.55 or later (program-options and iostreams modules only).
- GEOS 3.4.2 or later.
- shapelib 1.3.0 or later.
$ sudo apt install -y libmpich-dev libgdal-dev libgeos++-dev libshp-dev \
libboost-program-options-dev libboost-iostream-dev
$ git clone https://github.com/IllinoisStateGeologicalSurvey/p_point2grid
$ cd p_points2grid
$ make
$ bin/p_points2grid --helpAn alternative approach to installing natively is to build a docker image containing all the dependencies, the MPI runtime, and executable.
$ git clone https://github.com/IllinoisStateGeologicalSurvey/p_point2grid
$ cd p_points2grid
$ make buildp_points2grid works only with LAS version 1.0, 1.1, 1.2, and 1.4 input. Version 1.2 was tested most extensively with up to the 117 GB file size input, producing grids up to 27000X27000 cells. This cooresponds to 1 Meter resolution over a nominal 15 minute X 15 minute geographic extent.
Input flags specific to p_points2grid are:
-m [--interpolation_mode]
-c [--reader_count]
-b [--buffer_size]
-t [--mpi_times]
-f [--input_data_file_name ]
--output_bbox
--output_shape
--keep_class
--first_returns
--last_returns
--epsg
--bigtiff
--fill_empty
The output type of the grid is controlled with --min, --max, --mean, --idw, --std, -den, or --all. The default is all types. The oupput format is controlled with --output_format [all|arc|grid]. The default is all formats.
Here is a run on a cluster on a 117 GB LAS 1.2 file over the Grand Canyon area that produces all grids of all types at 1 meter resolution:
mpirun -n 1024 ./p_points2grid -i grand_canyon.las -o gc -m parallel
--reader_count 256 --mpi_time --resolution 1 --buffer_size 400000
--fill_window 7Here is a run on a cluster on 67 LAS 1.4 input files whose paths are listed in the file elwha using only points with classification 2, producing a 1 meter resolution BIGTIFF with an epsg projection code of 26910. Cell values are the mean of point elevation values within the default search radius.
mpirun -n 200 ./p_points2grid -f elwha -o elwha_kc2 -m parallel -c 80 --mean
--keep_class 2 --output_format tif -b 1000000 --resolution 1 --epsg 26910
--mpi_timesHere is a run on a cluster on 67 LAS 1.4 input files whose paths are listed in the file elwha using only points with classification 2, producing a 1 meter resolution BIGTIFF with an epsg projection code of 26910. Cell values are the mean of point elevation values within the default search radius. The output is clipped to the extent of elwha_urban.shp and empty cells are filled using row wise linear interpolation.
mpirun -n 200 ./p_points2grid -f elwha -o elwha_kc2_fill_shape -m parallel
-c 80 --mean --keep_class 2 --output_format tif -b 1000000 --resolution 1
--epsg 26910 --mpi_times --fill_empty --output_shape elwha_urban.shp