Workflow

Presented here is a generalized workflow for executing STILT simulations and applying flux inventories. In addition to the theoretical points below, several Tutorials provide workable examples to get you up and running.

  1. Collect meteorological data for simulations

    NOAA publishes High Resolution Rapid Refresh (HRRR) mesoscale model data in the ARL packed format required for STILT at ftp://arlftp.arlhq.noaa.gov/pub/archives/hrrr/. This is often the easiest place to start but is only available after June 15, 2015. The coupling of the popular Weather Research and Forecasting (WRF) model with STILT is well documented by Nehrkorn, 2010.

  2. Subset meteorological data for model domain (optional)

    Many meteorological data products are offered at a global, continental, or regional scale. STILT loads all meteorological data files that encompass the temporal domain of the simulation into memory. Reading these large files is often the largest bottleneck of simulations and is highly memory intensive. Recent versions of HYSPLIT provide a spatial grid extraction routine (xtrct_grid) for this purpose and documentation can be found in the HYSPLIT User’s Guide.

  3. Initialize a new STILT project with Rscript -e "uataq::stilt_init('myproject')"

    If the UATAQ package is not installed, see Installation.

  4. Define simulation controls in run_stilt.r
  5. Coordinates for receptor(s)
  6. Footprint grid domain and resolution
  7. Meteorological data path and file naming conventions in run_stilt.r
  8. Adjust parallel execution, transport, and dispersion settings (optional)
  9. Execute the model with Rscript run_stilt.r

    If not dispatching multi-node simulations with SLURM, using a job scheduler or Linux screen to execute the simulations in the background can be useful to avoid server disconnects cancelling the program. More information can be found in Execution.

  10. Convolve footprints with flux inventories to estimate contribution of near-field fluxes on the receptor

    Footprint units give a ppm contribution from the near-field when multiplied by the flux field. For a more detailed look, the Tutorials provide workable examples for convolving footprints with flux inventories.

  11. Add background signal to estimate changes outside of the near-field

    Background signals are often derived from measurements made just upstream from the model domain or a coarser model product such as Carbon Tracker.

  12. Analysis and visualizations

Advanced

After setting model parameters and executing run_stilt.r, meteorological data is symbolically linked to the user’s home directory to avoid fortran issues with paths longer than 80 characters. This will appear with the default format paste0('m', project) but can be adjusted or disabled.

User parameters are then passed to stilt_apply, which manages dispatching simulations across parallel threads or executing serially. If using SLURM for job submission, stilt_apply will use the rslurm package to submit jobs across n_nodes using n_cores per node. If running in parallel on a single node without SLURM, stilt_apply will use the parallel package to run simulations on the current node across n_cores. Otherwise, stilt_apply will run the simulations serially using lapply().