What does this skill do, and when should you use it?
This skill guides users in building single-member weather forecast inference scripts with Earth2Studio’s deterministic workflow. It covers model and data-source selection, IO backend selection, forecast-step calculation, and output-variable filtering. It requires Python 3.10+, an installed Earth2Studio environment, a CUDA-capable GPU, and network access for model weights and data. It does not cover ensemble workflows, diagnostics, data-only fetching, installation, or model training.
Collects forecast horizon, requested variables, region, and available GPU/VRAM; selects prognostic models using live documentation; checks model input variables against a data source’s vocabulary; chooses Zarr, NetCDF4, or Xarray IO backends; calculates nsteps; generates Python scripts using earth2studio.run.deterministic; provides a manual inference-loop checklist when explicitly requested; and explains how to read Zarr output and continue with diagnostics.
- A weather researcher needs a single-member forecast script for selected variables and a defined region.
- An Earth2Studio user needs to select a model based on forecast horizon, model timestep, region, and available VRAM.
- A developer needs to pair GFS, ARCO, IFS, or HRRR data with compatible model input variables.
- A user needs to save forecast output through Zarr, NetCDF4, or in-memory Xarray IO backends.
How do you install this skill?
- The core workflow depends on a CUDA-capable GPU and network downloads for weights and data; confirm environment and mainland-China reachability first.
- Live documentation and model/data-source compatibility require manual verification; the examples should not be treated as validated code.
- The supplied evaluation targets visibly conflict with their ground truth; correct them and rerun tests covering key paths before publication.
- Do not use this skill for installation, ensemble, diagnostics, or data-only fetching tasks.
- Network access
- Local filesystem
Python 3.10+Earth2StudioCUDA-capable GPUPyTorchNumPy
Install the skill from the NVIDIA/skills catalog with the skills CLI:
npx skills add nvidia/skills --skill earth2studio-deterministic-forecast --yes
To target Codex explicitly:
npx skills add nvidia/skills --skill earth2studio-deterministic-forecast --agent codex --yes
The source does not provide an Earth2Studio installation command; it only states that Earth2Studio must already be installed.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/NVIDIA/skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/earth2studio-deterministic-forecast" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- Build an Earth2Studio deterministic forecast script for 2025-01-01T00:00:00, with a 5-day horizon, 6-hour model steps, and t2m plus wind.
After installation, ask a compatible agent for a request such as: “Build an Earth2Studio deterministic forecast script for 2025-01-01T00:00:00, with a 5-day horizon, 6-hour model steps, and t2m plus wind.” The skill uses the forecast horizon, variables, region, and GPU/VRAM information to consult current model and data-source documentation, verify compatibility, and generate a CUDA-based Python script using earth2studio.run.deterministic. If variables are not specified, it preserves the full model output; for wind speed, it includes both u10m and v10m.
What are this skill's strengths and limitations?
- Covers the workflow from model selection through data compatibility, IO selection, and nsteps calculation.
- Explicitly checks model input variables against a data source VOCAB.
- Uses ZarrBackend as the default while supporting output filtering and alternative backends when justified.
- Provides both a run.deterministic path and a manual-loop checklist.
- Requires a CUDA-capable GPU and does not support CPU inference.
- Requires network access to model weights and data, plus live documentation checks before recommending models or data sources.
- Does not cover ensemble forecasting, diagnostics, data-only fetches, installation, or model training.
- The provided source does not describe a test suite or a broader set of validated platforms.
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Earth2Studio Deterministic Forecast this page ✓ NVIDIA · Official | 53 · Use with care | ★ 3.8k | today | Apache-2.0 |
| Earth2Studio Forecast Wrapper Builder ✓ NVIDIA · Official | 50 · Use with care | ★ 3.8k | today | Apache-2.0 |
| Earth2Studio Diagnostic Builder ✓ NVIDIA · Official | 56 · Use with care | ★ 3.8k | today | Apache-2.0 |
| Earth2Studio Discover ✓ NVIDIA · Official | 56 · Use with care | ★ 3.8k | today | Apache-2.0 |
| Earth2Studio Data Source Builder ✓ NVIDIA · Official | 43 · Not recommended | ★ 3.8k | today | Apache-2.0 |
How did FollowSkills review this skill?
The skill is scoped to deterministic weather inference, with no evidence of malware, credential theft, or covert exfiltration; however, it instructs users to download model weights and data over the network and write forecast outputs, without user-confirmation guidance, permission boundaries, sensitive-data handling, or rollback instructions, so points are deducted.
The workflow, manual-loop checklist, and troubleshooting table cover the main path; however, model/API compatibility depends on live documentation, examples are not execution-verified, and the evaluation targets conflict with their stated ground truth on models, data sources, and filenames. Static calibration also caps this dimension at 10, so the score remains conservative.
The description clearly separates deterministic forecasting from ensembles, diagnostics, installation, and other non-fit scenarios, and the evaluation includes positive and negative trigger cases; however, CUDA GPU, network access, and potentially overseas documentation/data services are hard prerequisites, with no mainland-China reachability assessment, CPU alternative, or finer input boundaries, so points are deducted.
The skill provides a name, version, license, author, live documentation links, ownership, limitations, and troubleshooting; however, it lacks the recommended Instructions and Examples sections and provides no clear changelog, maintenance commitment, or compatibility matrix. Progressive disclosure and installation notes are incomplete, so points are deducted.
It can guide generation of deterministic forecast scripts and covers step calculation, variable filtering, data-source selection, and IO selection, giving it clear core-task value; however, the examples depend on unverified current APIs, and the supplied evaluation materials contain target-versus-ground-truth inconsistencies, so outputs still require user validation and adjustment.
Committed evaluation configuration, positive and negative tasks, a static-validation report, and target scripts provide limited auditable evidence; however, Tier 3 results are unavailable, there are no execution results, and the report lacks detailed signals and coverage proof, so the score is limited.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →