NOAA Phase 2 Hydrological Data Processing

Docker Setup

Run ./scripts/setup to create the local development container.

Run the notebook server with ./scripts/server and then navigate to the URL that is provided on the host machine.

Notebooks

Here is a summary of the Jupyter notebooks in this repo.

nhd_nwm.ipynb: Shows how to get a HUC by id, query NHD for all reaches within the HUC, and then query NWM (in Zarr format) to get gridded and reach-based data. Assumes that a sample of NHDPlus V2 and NHDPlus HR have been loaded locally following the instructions above.
save_nwm_sample.ipynb: Saves a sample of NWM in Zarr and Parquet formats.
benchmark_zarr_parquet.ipynb: We want to see if it's faster to query reach-based data in NWM when it is stored in Parquet since it has more of a tabular flavor than the gridden datasets. This notebook implements a a rudimentary benchmark of the speed of querying NWM in Zarr vs. Parquet format.
save_nhd_extract.ipynb: Saves a GeoJSON file for each HUC in NHD containing the reach geometries and associated COMID fields. This is so that we can perform other workflows without needing an NHD database running.
hydrotools_test.ipynb: Shows how to use the Hydrotools library to compare predicted and observed streamflow levels at sites within a HUC.
archive_nwm_zarr.ipynb: Shows how to append grid-based NWM predictions to a Zarr file in order to archive them.
archive_nwm_parquet.ipynb: Shows how to append point-based NWM predictions to a Parquet file in order to archive them.
rechunker_test.ipynb: Test rechunker library on NWM
save_huc2_comids.ipynb: Save COMIDs for HUC2 02 which is used for testing purposes.
huc8_streamflow_query.ipynb: Run benchmarks using HUC8 streamflow query
rechunk_zarr_subset.ipynb: Save a subset of NWM and rechunk it to use in benchmarking experiments.

Local Database Setup

These instructions are for setting up a local copy of NHDPlus V2 and NHDPlus HR which is used in some notebooks. If a notebook requires this step, it should say so within its documentation.

While the server is running with ./scripts/server, run the psql database console using:

./scripts/psql

Download a sample of NHDPlus V2 which has the nhdflowline table with comid field that is used to reference reaches in NWM. This field is not available (to our knowledge) in NHDPlus HR. Then, uncompress it and load it into the database.

# in psql
CREATE DATABASE nhdplusv2;
\c nhdplusv2;
CREATE EXTENSION postgis;

# on host machine
ogr2ogr -f "PostgreSQL" PG:"host=localhost port=5432 user='postgres' password='password' \
    dbname='nhdplusv2'" data/NHDPlusMA/NHDPlus02/NHDSnapshot/Hydrography -overwrite -progress --config PG_USE_COPY YES -lco GEOMETRY_NAME=wkb_geometry

Then, follow similar steps to load a sample of the NHDPlus HR dataset which has the wbdhu12 table containing HUC 12 polygons. This table is not present in NHDPlus V2.

# in psql
CREATE DATABASE nhdplushr;
\c nhdplushr;
CREATE EXTENSION postgis;

# on host machine
ogr2ogr -f "PostgreSQL" PG:"host=localhost port=5432 user='postgres' password='password' \
    dbname='nhdplushr'" data/NHDPLUS_H_0204_HU4_GDB/NHDPLUS_H_0204_HU4_GDB.gdb -overwrite -progress --config PG_USE_COPY YES -lco GEOMETRY_NAME=wkb_geometry

Other Setup Docs

Setting up DaskHub on Kubernetes

Running on AWS Batch

Make a new ECR repository.
Make a new Batch job definition modeled after lfishgoldNoaaHydroData. This should point to the above ECR repo and use the queueCPU job queue.
Set the NOAA_ECR_IMAGE_NAME environment variable to the ECR repo created above.
Build the Docker image using ./scripts/update and then upload the image to ECR using ./scripts/ecr_publish.
Run commands in the container on Batch using something like the following assuming you have AWS CLI v2:

aws batch submit-job --job-name <a job name> --job-queue queueCPU --job-definition <the batch job definition> \
    --container-overrides '{"command": ["echo", "hello", "world"], "resourceRequirements": [{"value": "<the number of cores to use>", "type": "VCPU"}]}'