Welcome to snow_pc¶

A python package for automated processing of point clouds to simplify elevation creation, co-registration and differencing to facilitate the production of snow depth and vegetation products.

Free software: MIT license
Documentation: https://Surfix.github.io/snow_pc

Introduction and statement of Need¶

Light Detection and Ranging (LiDAR) and Structure from Motion (SfM) photogrammetry currently provide the most advanced and accurate approaches for monitoring snow distribution across a range of platforms, scales, and repeat intervals. These techniques generate high-resolution digital elevation models (DEMs) by producing georeferenced point clouds from overlapping imagery in the case of photogrammetry or from high frequency laser pulses in the case of LiDAR. However, post-processing of point clouds for generation of snow depth rasters remains complex compared to many other earth science applications such as topographic mapping, vegetation monitoring, geomorphology and landform analysis. Existing point cloud processing software suite such as Point Data Abstraction Library (PDAL) and LAStools provide general purpose tools for pre-processing, filtering and analyzing large point cloud data. Yet, there is a lack of tool that leveraged these capacities for optimized automated workflows specifically tailored for snow and ice applications. Consequently, complex manual interventions are often required for tasks like merging point cloud files from different flight lines or acquisitions, point clouds filtering, construction of (DTMs), aligning the DTMs and generation of products. This hinders efficient production of snow depth and limits full utilization of rich information in point clouds datasets.

snow_pc addresses this challenge by leveraging pdal and Stereo Pipeline (ASP) to automate point clouds management in a standardized workflow for generating elevation models, snow depths and vegetation products. This allows diverse users of developers, data processors and snow scientists to automate core point clouds processing tasks like merging, coordinate transformations, classification, co-registration and rasterization to simplify effort and facilitate multi-temporal analysis.

Usage¶

To install snow_pc, run this command:

Installation¶

1	`pip install leafmap`

You can also install from sources with this command:

1	`pip install git+https://github.com/Surfix/snow-pc`

Preparing point clouds for processing¶

The core module, snow_pc, features the prepare_pc function, which handles preprocessing tasks like removing whitespace from files, converting LAS to LAZ format, and merging files within a directory.

from snow_pc import prepare_pc
prepare_pc('project_dir')

This function returns the path to the processed point clouds. Additional filtering can be applied using functions from the filtering.py module.

from snow_pc.filtering import return_filtering, elm_filtering, outlier_filtering, dem_filtering, ground_segmentation
return_filtering('unfiltered_merge.laz')
dem_filtering('returns_filtered.laz', 'dem.tif', dem_low= 10, dem_high=50)
elm_filtering('dem_filtered.laz')
outlier_filtering('elm_filtered.laz', multiplier=2.5)
ground_segmentation('outlier_filtered.laz')

Intermittent results of DTM pipeline workflow for a track. The original point cloud is shown in top left. Point clouds after removal of invalid return is shown in top right. Bottom left is applying dem filter while the bottom right is the ground segmented points. Dem filter effectively removes noise in the point cloud such that no point is left for elm and outlier filter.

However, users can move on to generating elevation models and the necessary filtering will be applied as required.

Generating DEMs and DSMs¶

snow_pc module provides a pc2uncorrectedDEM for generating Digital Terrain Models and Digital Surface Models from point cloud file in a single code call.

from snow_pc import pc2uncorrectedDEM
dtm_tif, dtm_laz, dtm_laz, dsm_laz = pc2uncorrectedDEM('project_dir')

Aligning Point Clouds.¶

The laz_align function of the align module can be used to align a point cloud to a reference surface.

from snow_pc.align import laz_align`
aligned_laz = laz_align('project_dir')

The entire workflow can be accomplished in one line of code:

from snow_pc import snowpc_pipeline
snowpc_pipeline(in_dir)

snow_pc builds on many GIS tools particularly leafmap, gdal, pdal and ASP so these packages are automatically available after installing snow_pc and can be used as necessary. For example, the resulting point clouds at any stage can be view using the view_lidar() of leafmap.

import leafmap
leafmap.view_lidar('unfiltered_merge.laz', cmap="terrain")

The key features To learn more about snow_pc, check out the snow_pc api reference on the documentation website- https://Surfix.github.io/snow-pc

Key Features¶

Prepare Module¶

replace_white_spaces(in_dir) : remove white spaces in the point cloud files.
las2laz(in_dir) : Takes a user directory full of las files and convert the files to LAZ files. LAZ is a compressed version of LAS so it provides optimal data transfer and computation efficiency.
merge_laz_files(in_dir): merge all LAZ files in the project directory into one LAZ file. This step is crucial for mosaicking point cloud data from different flight lines to ensure seamless coverage over the area of interest, simplify data management tasks and facilitate processing in subsequent commands that take a single point cloud file.

Filtering module¶

return_filtering(laz_fp) : removes points with invalid returns where the return number or number of returns is zero. This is only required for LiDAR point clouds.
dem_filtering(laz_fp) : extracts only points within a defined elevation range relative to the reference DEM. This filter is important to remove atmospheric or MTA noise from the data thereby eliminating outlier points too far above or below the ground surface.
elm_filtering(laz_fp) : finds isolated low points that are likely errors or noise far below the actual ground level.
outlier_filtering(laz_fp) : removes extreme outlier points in the point cloud that deviate significantly from surrounding points.
ground_segmentation(laz_fp) : Classify the terrain points using the Simple Morphological Filter (SMRF) algorithm
surface_segmentation(laz_fp) : Isolate the surface points from the point clouds

Modeling module¶

terrain_models(laz_fp) : Use filters.dem, filters.mongo, filters.elm, filters.outlier, filters.smrf, and filters.range to filter the point cloud for terrain models [Todo: Refactor to leverage the filtering module]
surface_models(laz_fp) : Use filters.dem, filters.mongo and filters.range to filter the point point cloud for surface points [Todo: Refactor to leverage the filtering module]

Align module¶

clip_pc(in_laz, buff_shp) : Clip the point cloud to a shapefile. [To do]
align(in_laz, dem): Align the point clouds to a reference [To do]

Product module¶

generate_product(dtm_file, dsm_file): Derive snow depth and canopyheight from DTM and DSM files [To do]

Snow_pc module¶

prepare_pc(in_dir) : Steps through all preparing tools in one call
laz2uncorrectedDEM(laz_fp) : steps through all tools in prepare, filtering and modeling modules to generate dtm and dsm in one call
laz2correctedDEM(laz_fp) : steps through all tools in prepare, filtering, modeling and aligning modules to generate co-registered dtm and dsm in one call
snowpc_pipeline(in_dir): steps through all tools in prepare, filtering, modeling, align and product module to derive snow depth and vegetation height products

Common module¶

download_dem(las_fp) : Download DEM within the bounds of the las file.

TODO¶

[x] Add ground segmentation function to the filtering module
[] Refactor the modeling module
[] Refactor the clip_pc module and add pc_align function to align module (ASP does not support windows distribution???)
[] Refactor the snow_pc module (move download_dem function to common module, Merge all steps into one call)
[x] Complete surface segmation function to the filtering module
[] Use a better algorithm than first return isolation for the surface_segmentation (filtering_module)
[] Improve the segmentation accuracy for photogrammetry point clouds
[] Add a novel approach for combining LiDAR and photogrammetry point clouds
[] Add interactive map feature to allow users draw control surface for coregistration
[] Implement coregistration using points