dataflow-geobeam: A Python repository from tjwebb

geobeam adds GIS capabilities to your Apache Beam pipelines.

What does geobeam do?

geobeam enables you to ingest and analyze massive amounts of geospatial data in parallel using Dataflow. geobeam provides a set of FileBasedSource classes that make it easy to read, process, and write geospatial data, and provides a set of helpful Apache Beam transforms and utilities that make it easier to process GIS data in your Dataflow pipelines.

See the Full Documentation for complete API specification.

Requirements

Apache Beam 2.27+
Python 3.7+

Note: Make sure the Python version used to run the pipeline matches the version in the built container.

Supported input types

File format	Data type	Geobeam class
`tiff`	raster	`GeotiffSource`
`shp`	vector	`ShapefileSource`
`gdb`	vector	`GeodatabaseSource`

Included libraries

geobeam includes several python modules that allow you to perform a wide variety of operations and analyses on your geospatial data.

Module	Version	Description
gdal	3.2.1	python bindings for GDAL
rasterio	1.1.8	reads and writes geospatial raster data
fiona	1.8.18	reads and writes geospatial vector data
shapely	1.7.1	manipulation and analysis of geometric objects in the cartesian plane

How to Use

1. Install the module

pip install geobeam

2. Write your pipeline

Write a normal Apache Beam pipeline using one of geobeams file sources. See geobeam/examples for inspiration.

3. Run

Run locally

python -m geobeam.examples.geotiff_dem \
  --gcs_url gs://geobeam/examples/dem-clipped-test.tif \
  --dataset=examples \
  --table=dem \
  --band_column=elev \
  --centroid_only=true \
  --runner=DirectRunner \
  --temp_location <temp gs://> \
  --project <project_id>

You can also run "locally" in Cloud Shell using the py-37 container variants

Note: Some of the provided examples may take a very long time to run locally...

Run in Dataflow

Write a Dockerfile

This will run in Dataflow as a custom container based on the dataflow-geobeam/base image. See [geobeam/examples/Dockerfile] for an example that installed the latest geobeam from source.

FROM gcr.io/dataflow-geobeam/base
# FROM gcr.io/dataflow-geobeam/base-py37

RUN pip install geobeam

COPY requirements.txt .
RUN pip install -r requirements.txt

COPY . .

# build locally with docker
docker build -t gcr.io/<project_id>/example
docker push gcr.io/<project_id>/example

# or build with Cloud Build
gcloud builds submit --tag gcr.io/<project_id>/<name> --timeout=3600s --machine-type=n1-highcpu-8

Start the Dataflow job

Note on Python versions

If you are starting a Dataflow job on a machine running Python 3.7, you must use the images suffixed with py-37. (Cloud Shell runs Python 3.7 by default, as of Feb 2021). A separate version of the base image is built for Python 3.7, and is available at gcr.io/dataflow-geobeam/base-py37. The Python 3.7-compatible examples image is similarly-named gcr.io/dataflow-geobeam/example-py37

# run the geotiff_soilgrid example in dataflow
python -m geobeam.examples.geotiff_soilgrid \
  --gcs_url gs://geobeam/examples/AWCh3_M_sl1_250m_ll.tif \
  --dataset=examples \
  --table=soilgrid \
  --band_column=h3 \
  --runner=DataflowRunner \
  --worker_harness_container_image=gcr.io/dataflow-geobeam/example \
  --experiment=use_runner_v2 \
  --temp_location=<temp bucket> \
  --service_account_email <service account> \
  --region us-central1 \
  --max_num_workers 2 \
  --machine_type c2-standard-30 \
  --merge_blocks 64

Examples

Polygonize Raster

def run(options):
  from geobeam.io import GeotiffSource
  from geobeam.fn import format_record

  with beam.Pipeline(options) as p:
    (p  | 'ReadRaster' >> beam.io.Read(GeotiffSource(gcs_url))
        | 'FormatRecord' >> beam.Map(format_record, 'elev', 'float')
        | 'WriteToBigquery' >> beam.io.WriteToBigQuery('geo.dem'))

Validate and Simplify Shapefile

def run(options):
  from geobeam.io import ShapefileSource
  from geobeam.fn import make_valid, filter_invalid, format_record

  with beam.Pipeline(options) as p:
    (p  | 'ReadShapefile' >> beam.io.Read(ShapefileSource(gcs_url))
        | 'Validate' >> beam.Map(make_valid)
        | 'FilterInvalid' >> beam.Filter(filter_invalid)
        | 'FormatRecord' >> beam.Map(format_record)
        | 'WriteToBigquery' >> beam.io.WriteToBigQuery('geo.parcel'))

See geobeam/examples/ for complete examples.

A number of example pipelines are available in the geobeam/examples/ folder. To run them in your Google Cloud project, run the included terraform file to set up the Bigquery dataset and tables used by the example pipelines.

Open up Bigquery GeoViz to visualize your data.

Shapefile Example

The National Flood Hazard Layer loaded from a shapefile. Example pipeline at geobeam/examples/shapefile_nfhl.py

Raster Example

The Digital Elevation Model is a high-resolution model of elevation measurements at 1-meter resolution. (Values converted to centimeters). Example pipeline: geobeam/examples/geotiff_dem.py.

Included Transforms

The geobeam.fn module includes several Beam Transforms that you can use in your pipelines.

Module	Description
`geobeam.fn.make_valid`	Attempt to make all geometries valid.
`geobeam.fn.filter_invalid`	Filter out invalid geometries that cannot be made valid
`geobeam.fn.format_record`	Format the (props, geom) tuple received from a FileSource into a `dict` that can be inserted into the destination table

Execution parameters

Each FileSource accepts several parameters that you can use to configure how your data is loaded and processed. These can be parsed as pipeline arguments and passed into the respective FileSources as seen in the examples pipelines.

Parameter	Input type	Description	Default	Required?
`skip_reproject`	All	True to skip reprojection during read	`False`	No
`in_epsg`	All	An EPSG integer to override the input source CRS to reproject from		No
`band_number`	Raster	The raster band to read from	`1`	No
`include_nodata`	Raster	True to include `nodata` values	`False`	No
`centroid_only`	Raster	True to only read pixel centroids	`False`	No
`merge_blocks`	Raster	Number of block windows to combine during read. Larger values will generate larger, better-connected polygons.		No
`layer_name`	Vector	Name of layer to read		Yes
`gdb_name`	Vector	Name of geodatabase directory in a gdb zip archive		Yes, for GDB files

License

This is not an officially supported Google product, though support will be provided on a best-effort basis.

Copyright 2021 Google LLC

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    https://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.

tjwebb/dataflow-geobeam