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.
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-10
- Stars
- ★ 1
- Forks
- 0
01 前言 Front Matter
- name
- earth2studio-deterministic-forecast
- license
- Apache-2.0
- version
- 0.16.0
- metadata
- { "tags": [ "earth2studio", "earth2", "python", "inference", "forecast", "deterministic" ], "author": "NVIDIA Earth-2 Team" }
- description
- Build deterministic forecast scripts with Earth2Studio (model, data source, IO, inference). Do NOT use for ensemble, diagnostics, data-only fetch, or install.
02 资源文件 11
- 参考文档BENCHMARK.md3.0 KB
- 参考文档evals/config.yml1.2 KB
- 其他evals/environment/Dockerfile702 B
- 脚本evals/environment/setup/bootstrap.sh959 B
- 参考文档evals/evals.json5.5 KB
- 脚本evals/targets/eval_1_target.py1.6 KB
- 脚本evals/targets/eval_2_target.py1.4 KB
- 脚本evals/targets/eval_3_target.py3.2 KB
- 参考文档references/troubleshooting.md1.5 KB
- 参考文档skill-card.md2.7 KB
- 其他skill.oms.sig6.2 KB
03 SKILL.md 正文
Earth2Studio Deterministic Forecast Skill
Guide users through building deterministic (single-member) weather forecast
inference scripts using earth2studio.run.deterministic.
Prerequisites
- Earth2Studio installed with CUDA-capable GPU
- Python 3.10+, network access for model weights and data
Live Doc References
Fetch relevant docs to verify current APIs before recommending components:
| Component | URL |
|---|---|
| Prognostic models | https://nvidia.github.io/earth2studio/modules/models_px.html |
| Data sources (analysis) | https://nvidia.github.io/earth2studio/modules/datasources_analysis.html |
| Data sources (forecast) | https://nvidia.github.io/earth2studio/modules/datasources_forecast.html |
| IO backends | https://nvidia.github.io/earth2studio/modules/io.html |
run.deterministic |
https://github.com/NVIDIA/earth2studio/blob/main/earth2studio/run.py |
Workflow
1. Gather Requirements (skip what's already provided)
- Time horizon (hours/days/weeks)
- Variables of interest (t2m, wind, geopotential, etc.)
- Region (global or specific like CONUS)
- GPU/VRAM available
2. Select Model
Fetch prognostic models page. Filter by time horizon, region, VRAM. Note model's:
- Input variables (
input_coords["variable"]) - Time step size (
output_coords["lead_time"])
3. Select Data Source
Data source must provide all model input variables. Verify via lexicon at
earth2studio/lexicon/<source>.py. Common pairings: Global models → GFS/ARCO/IFS;
Regional → HRRR.
4. Select IO Backend
Default: ZarrBackend. Use NetCDF4Backend for legacy tools, XarrayBackend
for in-memory/small runs.
5. Calculate nsteps
nsteps = forecast_hours / model_step_hours
Example: 5-day forecast with 6h step → nsteps = 120 / 6 = 20
6. Decide: output_coords Filtering
- Filter variables (
output_coords) when user requests specific variables (e.g., "t2m and wind") - reduces output size - Save all variables (omit
output_coords) when user says "all variables" or doesn't specify - preserves full model output
7. Generate Script
from collections import OrderedDict
import numpy as np
import torch
from earth2studio.models.px import <ModelClass>
from earth2studio.data import <DataSourceClass>
from earth2studio.io import <IOBackendClass>
from earth2studio.run import deterministic
model = <ModelClass>.load_model(<ModelClass>.load_default_package())
data = <DataSourceClass>()
io = <IOBackendClass>("<output_path>")
# Include output_coords ONLY if user requested specific variables
output_coords = OrderedDict({"variable": np.array(["t2m", "u10m"])})
io = deterministic(
time=["YYYY-MM-DDTHH:MM:SS"],
nsteps=<N>,
prognostic=model,
data=data,
io=io,
output_coords=output_coords, # omit if saving all variables
device=torch.device("cuda"),
)
8. Manual Loop Alternative
When user explicitly requests manual implementation (NOT using earth2studio.run.deterministic), follow this checklist in order:
- fetch_data - Get initial conditions:
x, coords = fetch_data(data, time, model.input_coords, device) - Setup total_coords - Build coordinate arrays for time and lead_time dimensions
- io.add_array - Initialize IO backend with total_coords before loop
- create_iterator - Create prognostic iterator:
model_iter = model.create_iterator(x, coords) - Loop through nsteps -
for step, (x, coords) in enumerate(model_iter): if step >= nsteps: break - map_coords - Filter output variables if needed:
x_out, coords_out = map_coords(x, coords, output_coords) - split_coords - Prepare for IO write:
x_out, coords_out = split_coords(x_out, coords_out) - io.write - Write each step to backend
9. Explain Next Steps
- How to change forecast time or run multiple initializations
- How to read output (
xr.open_zarr(...)) - Point to diagnostic workflow for post-processing
Ownership
Owns: Model selection, data source compatibility, IO backend selection,
nsteps calculation, generating earth2studio.run.deterministic scripts.
Does not own: Ensemble workflows, diagnostics, data-only fetch, installation, model training.
Troubleshooting
See references/troubleshooting.md for common errors and solutions.
Reminders
- Always fetch live docs before recommending models or data sources - APIs change between releases
- Verify lexicon compatibility - Model input variables must exist in data source's VOCAB
- Use
load_default_package()- This is the standard pattern for loading model weights - Time format is ISO 8601 - Use
"YYYY-MM-DDTHH:MM:SS"format for thetimeargument - Wind speed needs both components - If user asks for "wind speed", include both
u10mandv10m - nsteps is integer division -
nsteps = total_hours // model_step_hours - ZarrBackend is the default - Only suggest alternatives if user has specific requirements
- GPU is required - All prognostic models require CUDA; CPU inference is not supported
04 同仓库更多 Skills
vercel-sandbox
Run agent-browser + Chrome inside Vercel Sandbox microVMs for browser automation from any Vercel-deployed app. Use when the user needs browser automation in a Vercel app (Next.js, SvelteKit, Nuxt, Remix, Astro, etc.), wants to run headless Chrome without binary size limits, needs persistent browser sessions across commands, or wants ephemeral isolated browser environments. Triggers include "Vercel Sandbox browser", "microVM Chrome", "agent-browser in sandbox", "browser automation on Vercel", or any task requiring Chrome in a Vercel Sandbox.
slack
Interact with Slack workspaces using browser automation. Use when the user needs to check unread channels, navigate Slack, send messages, extract data, find information, search conversations, or automate any Slack task. Triggers include "check my Slack", "what channels have unreads", "send a message to", "search Slack for", "extract from Slack", "find who said", or any task requiring programmatic Slack interaction.
electron
Automate Electron desktop apps (VS Code, Slack, Discord, Figma, Notion, Spotify, etc.) using agent-browser via Chrome DevTools Protocol. Use when the user needs to interact with an Electron app, automate a desktop app, connect to a running app, control a native app, or test an Electron application. Triggers include "automate Slack app", "control VS Code", "interact with Discord app", "test this Electron app", "connect to desktop app", or any task requiring automation of a native Electron application.
dogfood
Systematically explore and test a web application to find bugs, UX issues, and other problems. Use when asked to "dogfood", "QA", "exploratory test", "find issues", "bug hunt", "test this app/site/platform", or review the quality of a web application. Produces a structured report with full reproduction evidence -- step-by-step screenshots, repro videos, and detailed repro steps for every issue -- so findings can be handed directly to the responsible teams.
agent-browser
Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a website", "fill out a form", "click a button", "take a screenshot", "scrape data from a page", "test this web app", "login to a site", "automate browser actions", or any task requiring programmatic web interaction.
shadcn
Manages shadcn components and projects — adding, searching, fixing, debugging, styling, and composing UI, including chat interfaces. Provides project context, component docs, and usage examples. Applies when working with shadcn/ui, component registries, presets, --preset codes, or any project with a components.json file. Also triggers for "shadcn init", "create an app with --preset", "switch to --preset", and for building a design system or applying a DESIGN.md or brand spec to shadcn/ui.