docs-metmar

Forecast Mode

!!! warning The hands-on in this section is intended for internal BMKG users. Non-BMKG users may still be able to follow the tutorial without access to the computational resources.

!!! info Please contact the team for further information at produksi.maritim@bmkg.go.id.

The operational InaCAWO forecast system runs automatically on the BMKG HPC under account cawofcst_prod, managed by ROME. This page describes how the forecast mode works; for hands-on hindcast exercises see Hindcast Mode.

Forecast schedule

InaCAWO produces four forecast cycles per day:

Cycle (UTC) Forecast length Atmospheric IC/BC Output products
00, 12 10 days (240 h) ECMWF-IFS (primary), GFS (backup) FULL + ESSENTIAL
06, 18 90 hours GFS only FULL only

End-to-end workflow

flowchart TD
    A[Download global model data] --> B[Pre-process IC/BC]
    B --> C[72-hour ROMS/SWAN spin-up]
    C --> D[6-hour coupled analysis cycle]
    D --> E[Launch forecast run]
    E --> F[Post-process and disseminate]

1. Data ingest and pre-processing

Global model data are downloaded and converted into model-ready files:

See Configurations for data source details.

2. Spin-up and analysis cycling

Before each forecast, the system runs:

  1. Standalone spin-up (72 h) — ROMS and SWAN are driven with GFS analysis forcing to produce balanced hot-start files
  2. Coupled analysis cycle (6 h) — the full 3-way coupled model runs from T−6 to T0, producing final ocean and wave initial conditions

3. Forecast execution

The coupled COAWST model runs in time chunks to minimize wall-clock delay:

10-day runs (00/12 UTC): 0–24, 24–48, 48–96, 96–144, 144–192, 192–240 h

90-hour runs (06/18 UTC): shorter chunked sequence

Each chunk waits for its IC/BC data, with ECMWF preferred and GFS as automatic fallback.

4. Post-processing and delivery

As forecast output timesteps become available:

See Dissemination for output variable lists and file naming.

Key differences from hindcast mode

Aspect Forecast (operational) Hindcast
Trigger Automatic via ROME crontab Manual execution
Input data Near-real-time global models Pre-acquired reanalysis (ERA5, GLORYS12V1)
ROMS OBCs Zero-gradient velocities Radiation + nudging with climatology
Time pressure Must complete within service window No real-time constraint
Workflow ROME job tree with backups Simplified independent components

Accounts and access

Account Node Use
cawofcst_prod mdclogin0 Production — do not modify without authorization
cawofcst_dev mdclogin1 R&D testing and script development

Software deployment repositories are available at https://mms.bmkg.go.id/gitlab/mms/inacawo-deployment/.

Monitoring a forecast cycle

To check the status of a running or recent cycle:

# Set cycle (example: 12 UTC on 2024-05-30)
export CYCLE=2024053012

# Check cycle log directory
ls $RT_LOGS/cycles/$CYCLE/

# Check for flagged errors
ls $RT_LOGS/cycles/$CYCLE/verify/

# Review cron launch logs
ls $RT_LOGSCRON/ | grep $(date -u +%Y%m%d)

For detailed job listings and error recovery procedures, see Production and Troubleshooting.

References