Ensemble configuration

Every figure on this page is read from the system that produced it — physics from the namelist templates, runtimes from the driver's own timing log, the schedule from the prefetch scripts and crontab. Nothing here is typed by hand, so it cannot drift from what is actually running.

Members

Four physics configurations, each run from both NOAA GFS and ECMWF IFS boundary conditions, plus two AI-driven runs that use the control physics with initial and boundary conditions from ECMWF AIFS-single and NOAA AIGFS — ten members per full cycle. ARG is the control and is also the operational deterministic forecast — running the ensemble never costs a separate control run. The AI members take their atmosphere from the AI model and their land surface (soil temperature and moisture, skin temperature) from the operational IFS (for AIFS) or GFS (for AIGFS) initial state of the same family, because neither AI model carries the four soil layers WRF needs. The two families are never mixed, so AIFS starts after the operational IFS initial state is published, about 7.6 h after the cycle.

Domain

One-way nested; convection is parameterized on the 15 km parent and explicit on the 3 km nest. Mercator, because Lambert is badly conditioned at 2°S. All 189 INAMHI verification gauges, including all 44 in the Guayas basin, fall inside the nest.

Measured runtimes

Operational schedule

Compute

Members run concurrently inside ONE allocation, each pinned to its own pair of nodes with mpirun -hosts. The partition allows this account four concurrent jobs and production already uses three of them (driver, viewer postprocessing, hydro plots), so a packed single job is not an optimization — a second job would push production postprocessing into the queue.

ensemble viewer · deterministic viewer · verification