WRF-Chem and HYSPLIT forecast workflow support

The problem

Teams that run WRF-Chem or HYSPLIT operationally spend much of their effort on things other than atmospheric science: chained preprocessing steps, boundary-condition downloads that fail, jobs killed by wall-time or memory limits, disks filling with output, and post-processing scripts only one person understands. A missed cycle means a missing forecast on the public dashboard. The modelling scientists need an engineer who understands the model workflow well enough to make it robust.

What you get

• Review of the existing forecast chain (preprocessing, model run, post-processing, publication), with a list of failure points
• Scripted and scheduled workflow with batch-job submission, dependency handling and logging
• Run-health monitoring and alerting for failed or late cycles
• Post-processing to CF-compliant NetCDF and map/time-series products ready for dashboards
• Basic model-versus-observation comparison outputs for the scientific team
• Version-controlled repository and documentation

How it works

1. 1

   Audit

   Walk through the current run end to end on the client's HPC or server with the modelling team, and record each manual step and each recent failure.

2. 2

   Automate

   Script the chain under Git, handle external data dependencies with retries, and submit and track jobs through the cluster's scheduler.

3. 3

   Monitor

   Add checks for job status, output completeness and plausibility, and storage, with notifications to the operators.

4. 4

   Publish

   Convert outputs with xarray, CDO and NCO into the formats and services the dashboard and analysts need, including observation-matched extracts for evaluation.

Tools

• WRF-Chem
• HYSPLIT
• Linux HPC and batch schedulers
• Python (xarray, NumPy)
• CDO
• NCO
• THREDDS Data Server
• Git