Migrating From R-STILT

R-STILT organizes projects around R functions, the run_stilt.r script, and filesystem conventions. PYSTILT expresses the same broad ideas through Python classes, config.yaml, and a output simulation index.

R-STILT concept	R-STILT API	PYSTILT equivalent
Project initialization	stilt_init(project_path)	stilt init or stilt.Model(project=...)
Run controls	run_stilt.r variables	stilt.ModelConfig fields
Meteorology	met_directory, met_file_format, met_file_tres	stilt.MetConfig in ModelConfig.mets
Receptors	receptor data.frame	stilt.PointReceptor / stilt.ColumnReceptor / stilt.MultiPointReceptor or stilt.read_receptors()
Footprint grid	xmn, xmx, ymn, ymx, xres, yres	stilt.Grid inside stilt.FootprintConfig
Parallel execution	slurm=TRUE and scheduler options	execution.backend: slurm plus executor config
Before-footprint transform	callback hooks in R	FootprintConfig.transforms or runtime particle transforms
Trajectory output	R serialized trajectory objects	Parquet trajectory files
Footprint output	NetCDF footprint outputs	NetCDF footprint outputs with output per-name status
Simulation identity	timestamp plus location string	<met>_<YYYYMMDDHHMM>_<location_id>

Key differences

• configuration is validated before execution

• project, output, and compute roots can be separate

• output status and reruns are first-class rather than incidental

• local, Slurm, and Kubernetes execution share one runtime model

Practical advice

• Port one receptor family at a time.

• Start with one met stream and one footprint.

• Treat config.yaml as your migration boundary.

• Validate parity on a few representative receptors before scaling out.