How To Run FEDS Locally

Author

Zeb Becker

Published

June 25, 2026

Running FEDS locally

This doc provides quick instructions for setting up the required file structure and computational environment to run FEDS on your local machine, without any need for MAAP S3 credentials at runtime.

Start by cloning the fireatlas git repo and setting up the required conda environment by following the directions on the Contributing page. If you are trying to test changes you’ve made on a specific branch, make sure that branch is checked out.

Unless otherwise specified, all commands are intended to be run from the root fireatlas directory that you just cloned and all filepaths are relative to it. Note the /data/ folder in this directory- this is both where we will need to download some input datasets to, and where your outputs will end up.

1. Creek fire using local test data

First, create a file called run_config.yaml. FEDS looks for this file at the repo root. The quickest way to start is to copy the example config:

cp run_config.example.yaml run_config.yaml

The example file is already set up for this small example run of the Creek fire.

# run_config.yaml  (in the repo root)
settings:
  READ_LOCATION: local
  LOCAL_PATH: tests/data
  FIRE_SOURCE: SNPP
  FIRE_NRT: False
  EPSG_CODE: 9311
  FTYP_OPT: CA
  CONT_OPT: CA
  remove_static_sources: False

This will override the default settings in FireConsts.py for the specified values. In this example, we are going to read our input data from fireatlas/tests/data, which includes a small subset of input data for the Creek fire. This data is included in the tests/ folder because it is used in the automated test suite, so this is a nice way to test small changes without needing to download any additional data.

Next, we are going to run FEDS from the command line. Make sure you have installed and activated the conda environment (again, see Contributing for directions if you haven’t).

python fireatlas/FireRunDaskCoordinator.py --regnm=v3_test_data_for_Creek_SNPP \
    --bbox="[-119.5,36.8,-118.9,37.7]" \
    --tst="[2020,9,5,\"AM\"]" \
    --ted="[2020,11,5,\"PM\"]" \
    --no-veda-copy

2. CONUS for 2025

If the Creek fire example worked, you may want to move on to other regions and time periods. However, we don’t have all possible input data checked into the repository. In order to access this data without being connected to MAAP S3 storage, we will have to download the required input data ourselves. For this example, we will run FEDS for CONUS for all of 2025, using NOAA20 monthly files. Following this model, you can use any date range, bbox, or satellite. Please reach out if you have issues accessing the required data.

Required Downloads:

  1. VIIRS active fire detections (standard product)- download from the UMD FTP server following the instructions in the notebook fireatlas/notebooks/21_Download_VIIRS_SP_SFTP.ipynb
  2. Landcover from MAAP: Download landcover from s3://maap-ops-workspace/shared/gsfc_landslides/FEDSpreprocessed/nlcd_export_510m_simplified_latlon.tif to our local data folder fireatlas/data/FEDSpreprocessed/nlcd_export_510m_simplified_latlon.tif. You can use the ADE or MAAP Hub interfaces to do this manually.
  3. Static flaring sources: Download the static source filter from s3://maap-ops-workspace/shared/gsfc_landslides/FEDSinput/static_sources/VIIRS_Global_flaring_d.7_slope_0.029353_2017_web_v1.csv to our local data folder at fireatlas/data/FEDSinput/static_sources/VIIRS_Global_flaring_d.7_slope_0.029353_2017_web_v1.csv

Once those are all downloaded, lets change a few of the settings in run_config.yaml. Paste this text into the existing file and don’t forget to hit save. You can also override any other settings from FireConsts.py that you would like. This yaml file will then serve as a neat encapsulation of the settings you use for a particular run.

settings:  
  READ_LOCATION: local
  # notice that we deleted the local path override from above! 
  # it is now set to fireatlas/data by default, not fireatlas/tests/data
  FIRE_SOURCE: NOAA20
  FIRE_NRT: False # still using the standard product and monthly files
  EPSG_CODE: 9311
  FTYP_OPT: CA
  CONT_OPT: CA

Finally, run this command to give it a try! Check fireatlas/data/FEDSoutputs-v3 for results when you are finished.

python fireatlas/FireRunDaskCoordinator.py --regnm="example_CONUS2025" \
    --bbox="[-126,24,-61,49]" \
    --tst="[2025,1,1,\"AM\"]" \
    --ted="[2025,12,31,\"PM\"]" \
    --no-veda-copy

3. CONUS NRT

Ok, last example. Let’s try doing just a few days of NRT outputs for CONUS. If you want them to match the latest production FEDS NRT outputs, you’d have to start the annual run on January 1st. However, it would take a while to download all of that data one day at a time from the FIRMS API. If you need a longer range of NRT data, use their bulk download tool and fireatlas/notebooks/20_FIRMS_input_backfill. For this example, we are going to just grab the past few days, so we will skip that part.

Your turn to edit run_config.yaml- change FIRE_NRT to True, making sure to save when you are done. This will tell FEDS to use NRT data. With FIRE_NRT=True, FEDS will automatically try to download any missing input data, one day at a time, from the FIRMS API.

Then, use the same command as above to run FireRunDaskCoordinator.py, but change the dates so that TST is a few days ago, and TED is today. The fireatlas package does not currently support time travel, so be sure not to set TED to “PM” if it is still currently “AM” in the region you are looking at.

Once again, outputs for your run can be found at fireatlas/data/FEDSoutputs-v3. Good job!