earth2studio-deterministic-forecast
Build deterministic forecast scripts with Earth2Studio (model, data source, IO, inference). Do NOT use for ensemble, diagnostics, data-only fetch, or install.
Security Assessment
About earth2studio-deterministic-forecast
Earth2Studio Deterministic Forecast guides users through building single-member weather forecast inference scripts using earth2studio.run.deterministic. It solves the assembly problem of wiring together the four required components, a prognostic model, a compatible data source, an IO backend, and the inference call, into a correct, runnable script tailored to the user's time horizon, variables, region, and available GPU memory.
The skill emphasizes fetching live documentation before recommending components, since APIs change between releases. Its workflow gathers requirements, selects a prognostic model filtered by horizon/region/VRAM, chooses a data source that provides all model input variables (verified against the source's lexicon VOCAB), picks an IO backend (ZarrBackend by default, with NetCDF4 or Xarray alternatives), and computes nsteps via integer division of forecast hours by the model step. It generates a complete script template and, when requested, a manual-loop alternative with an ordered checklist covering fetch_data, coordinate setup, add_array, create_iterator, the step loop, map_coords, split_coords, and io.write. It clearly delineates ownership (model/data/IO selection, nsteps, script generation) versus what it does not cover (ensembles, diagnostics, data-only fetch, installation, training), and bundles a troubleshooting reference.
It targets weather and climate ML practitioners with a CUDA-capable GPU who want to run deterministic forecasts. Typical use cases include producing a five-day global forecast for selected variables, pairing a regional HRRR-based model with the right data source, and scripting multiple initialization runs.
FAQ
What kind of forecast does this build?
Deterministic, single-member forecasts via earth2studio.run.deterministic, assembling a prognostic model, data source, IO backend, and inference call into a runnable script.
What are the hardware and software prerequisites?
Earth2Studio installed with a CUDA-capable GPU, Python 3.10+, and network access for model weights and data. All prognostic models require CUDA; CPU inference is not supported.
How is the number of forecast steps determined?
nsteps is integer division of the total forecast hours by the model's step size, for example a 5-day forecast with a 6-hour step gives nsteps = 120 / 6 = 20.
How does it ensure model and data source compatibility?
The chosen data source must provide all of the model's input variables, which is verified against the source's lexicon VOCAB file; common pairings are global models with GFS/ARCO/IFS and regional models with HRRR.
What is out of scope?
Ensemble workflows, diagnostics/post-processing, data-only fetch, installation, and model training are explicitly not owned by this skill; a separate troubleshooting reference covers common errors.
All Files
12 filesInstall earth2studio-deterministic-forecast
Quick Setup:
- Copy the skill folder to
.claude/skills/ - Claude will automatically detect and use the skill
Repository
nvidia/skills