Branch / tenant: dev/gfs.v17 (gw_v17) Path: ecf/scripts/gfs/product/atmos/gempak/goesupp/jgfs_atmos_goesupp_master.ecf Analysis date: 2026-06-11 Method: AWS Bedrock AgentCore agentcore-mcp-rag MCP (graph traversal, GraphRAG, code-context) cross-checked against the on-disk v17 worktree.

jgfs_atmos_goesupp_master.ecf is the NCO production ecFlow task template that runs the Unified Post Processor (UPP) in "GOES supplemental" mode to generate simulated geostationary satellite imagery products from GFS forecast output. It is a _master template (deployed per forecast hour). Unlike the other GEMPAK .ecf files in the same family, this one calls the shared JGLOBAL_ATMOS_UPP J-Job with UPP_RUN="goes", driving the Python-based exglobal_atmos_upp.py workflow via pygfs.task.upp.UPP.

#PBS -S /bin/bash
#PBS -N gfs_atmos_goesupp_%CYC%_f%FHR%
#PBS -j oe
#PBS -q %QUEUE%
#PBS -A %PROJ%-%PROJENVIR%
#PBS -l walltime=00:15:00
#PBS -l select=1:mpiprocs=1:ompthreads=1:ncpus=120:mem=500GB
#PBS -l place=vscatter:exclhost
#PBS -l debug=true

model=gfs
export cyc="%CYC%"
%include <head.h>
%include <envir-p1.h>

############################################################
# Load modules
############################################################
#TODO: load only the modules needed for each job
source ${HOMEgfs}/dev/ush/load_modules.sh upp

module list
export WGF="%WGF%"
export FHR3="%FHR%"
export FHR_LIST="%FHR_LIST%"
export FORECAST_HOUR="${FHR_LIST}"
export COMPONENT=${WGF}
export UPP_RUN="goes"
############################################################
# CALL executable job script here
############################################################
export CDATE=${PDY}${cyc}
export HOMEglobal=${HOMEgfs}

#### ${HOMEgfs}/dev/job_cards/rocoto/upp.sh
${HOMEgfs}/jobs/JGLOBAL_ATMOS_UPP
if [ $? -ne 0 ]; then
   ecflow_client --msg="***JOB ${ECF_NAME} ERROR RUNNING J-SCRIPT ***"
   ecflow_client --abort
   exit
fi

%include <tail.h>
%manual

%end

A single .ecf file backs per-forecast-hour ecFlow tasks. The ecFlow generator deploys it as jgfs_atmos_goesupp_f000.ecf, jgfs_atmos_goesupp_f003.ecf, etc. Each instance overrides FHR and FHR_LIST.

ecf/defs/gfs_prod.def : task jgfs_atmos_goesupp_fNNN
  +-- jgfs_atmos_goesupp_master.ecf            (PBS submit; deployed as *_fNNN.ecf)
       +-- ${HOMEgfs}/jobs/JGLOBAL_ATMOS_UPP   (J-Job, shared)
            +-- ${SCRglobal}/exglobal_atmos_upp.py   (Python driver)
                 +-- pygfs.task.upp.UPP   (wxflow-based UPP task class)
                      +-- UPP.initialize()  — copy static data to DATA/
                      +-- UPP.configure()   — prepare namelist, control file
                      +-- UPP.execute()     — run UPP binary via APRUN_UPP
                      +-- UPP.finalize()    — copy/index output GRIB2 master files

Confirmed via MCP graph (trace_data_flow, tenant gw_v17):

• JGLOBAL_ATMOS_UPP SOURCES jjob_header.sh, jjob_standard_vars.sh, jjob_shell_setup.sh, set_strict.sh.
• The ex-script is Python (exglobal_atmos_upp.py), invoking the pygfs.task.upp.UPP class from the wxflow/pygfs Python package.

When UPP_RUN="goes", the UPP generates simulated GOES satellite imagery fields in GRIB2 master format:

Input is read from ${COMIN_ATMOS_HISTORY} (model history files) and ${COMIN_ATMOS_ANALYSIS} (for surface fields). The upp_run config key controls which UPP YAML configuration and control files are selected.

Uses the EE2-correct pattern: "${SCRglobal}/exglobal_atmos_upp.py" && true followed by err_exit on non-zero status.

The J-Job sources jjob_header.sh -e "upp" -c "base upp", loading:

• config.base — fundamental paths and settings.
• config.upp — UPP-specific settings (resources, APRUN, control file paths).

• Python-based UPP driver: v17 replaces the traditional shell ex-script with exglobal_atmos_upp.py using the pygfs.task.upp.UPP class from wxflow. This is a major architectural shift from older develop-branch GEMPAK/UPP patterns.
• Shared J-Job: JGLOBAL_ATMOS_UPP is reused across multiple ECF task families (jgfs_atmos_product_master, jgfs_atmos_goesupp_master, potentially others) — the UPP_RUN variable selects the processing mode.
• Heavy resource requirements: 120 cpus, 500 GB, exclusive host placement — reflects the computational cost of satellite simulation fields at high resolution.
• load_modules.sh upp: loads UPP-specific modules (different from the run module set used by GEMPAK conversion tasks).
• Centralized module loading via load_modules.sh.
• HOMEglobal shim bridges v17 naming to NCO.

Note: although the graph shows 0 direct dependents for JGLOBAL_ATMOS_UPP, this J-Job is shared across multiple ECF task templates (jgfs_atmos_product_master, jgfs_atmos_goesupp_master, etc.). Changes to its behavior affect all UPP_RUN modes.

• On-disk v17 worktree: ecf/scripts/gfs/product/atmos/gempak/goesupp/jgfs_atmos_goesupp_master.ecf, dev/jobs/JGLOBAL_ATMOS_UPP, dev/scripts/exglobal_atmos_upp.py.
• agentcore-mcp-rag MCP (tenant gw_v17): trace_data_flow, get_change_impact, get_code_context.