onescience_run_manifest.yaml

onescience_manifest_version: "0.1"
resource_type: dataset
resource:
  id: OneScience/ERA5
  name: ERA5
  domain: earth
  domain_tags: [earth, atmosphere, reanalysis]
  task: weather_forecast
  task_tags: [weather_forecast, dataset, benchmark]
  modalities: [gridded_field]
  input_formats: [hdf5]
  output_formats: []
  summary: 1979-2025 年 ERA5 年度 HDF5 数据集，包含多变量格点场和全局归一化统计量。

platform_resource:
  primary:
    platform: modelscope
    repo_id: OneScience/ERA5
    repo_type: dataset
    url: https://modelscope.cn/datasets/OneScience/ERA5
    revision: main
    readme_path: README.md
    manifest_path: onescience_run_manifest.yaml
  mirrors: []
  access:
    visibility: public
    license: null

website_integration:
  enabled: true
  click_target:
    platform: modelscope
    resource_url: https://modelscope.cn/datasets/OneScience/ERA5
  llm_handoff:
    readme_required: true
    manifest_required: true
    download_readme_first: true
    resolve_related_models: true
    default_run_goal: minimal_validation

runtime:
  enabled: true
  onescience_domain: earth
  min_onescience_version: null
  supported_execution: [local, local_slurm, remote_slurm, scnet_mcp]

onescience:
  repo: https://gitee.com/onescience-ai/onescience
  official_links:
    gitee:
      doc: https://gitee.com/onescience-ai/onescience-doc
      onescience: https://gitee.com/onescience-ai/onescience
      skills: https://gitee.com/onescience-ai/oneskills
    github:
      doc: https://github.com/onescience-ai/OneScience-doc
      onescience: https://github.com/onescience-ai/OneScience
      skills: https://github.com/onescience-ai/oneskills
  install:
    required_by_default: false
    command: bash install.sh earth
  source_paths: []
  compatibility:
    examples_path: null
    status: standard_first

runtime_package:
  kind: standard_runtime_package
  package_root: .
  standard_layout:
    workdir: .
    config_dir: metadata
    checkpoint_dir: null
    data_dir: data
    output_dir: null
  apply_policy:
    mode: direct_use
    target: session_workdir
    overwrite: true
    protect_installed_onescience: true
  entry_files:
    - metadata/schema.yaml
    - scripts/validate_era5_dataset.py
  entrypoints:
    preflight: scripts/validate_era5_dataset.py
    inference: null
    train: null
    finetune: null
    evaluate: null
    visualize: null
    deploy: null

files:
  model_files: []
  dataset_files:
    - id: era5_yearly_hdf5
      path: data/*.h5
      role: train_data
      description_zh: 1979-2025 年 ERA5 年度 HDF5 文件，包含 fields、global_means 和 global_stds。
      required: true
      required_for: [preflight, train, inference, evaluate, visualize]
      format: hdf5
      source:
        platform: modelscope
        repo_id: OneScience/ERA5
        repo_type: dataset
        path: data/*.h5
        revision: main
        download_method: command_ref
        command_ref: commands.download.download_full_dataset
      mirrors: []
      local_path: data/*.h5
      size: "约 221G，47 个年度文件"
      checksum: null
      large_file: true
    - id: era5_schema
      path: metadata/schema.yaml
      role: schema
      description_zh: HDF5 数据结构、shape、dtype、attrs 和样本定义。
      required: true
      required_for: [preflight]
      format: yaml
      source:
        platform: modelscope
        repo_id: OneScience/ERA5
        repo_type: dataset
        path: metadata/schema.yaml
        revision: main
        download_method: command_ref
        command_ref: commands.download.download_full_dataset
      mirrors: []
      local_path: metadata/schema.yaml
      size: null
      checksum: null
      large_file: false
  config_files:
    - id: dataset_schema
      path: metadata/schema.yaml
      role: dataset_schema
      description_zh: 数据 schema 文件。
      required: true
      required_for: [preflight]
      format: yaml
      source:
        platform: modelscope
        repo_id: OneScience/ERA5
        repo_type: dataset
        path: metadata/schema.yaml
        revision: main
        download_method: command_ref
        command_ref: commands.download.download_full_dataset
      local_path: metadata/schema.yaml
      size: null
      checksum: null
      large_file: false
  sample_files: []

dataset:
  format: [hdf5]
  sample_unit: 一个由 ERA5Datapipe 从年度 HDF5 相邻时间步构造的输入/目标格点场样本。
  schema:
    path: metadata/schema.yaml
  splits:
    train:
      years: [1979, 1980, 1981, 1982, 1983, 1984, 1985, 1986, 1987, 1988, 1989, 1990, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016, 2017, 2018, 2019, 2020, 2021, 2022, 2023]
    validation:
      years: [2024]
    test:
      years: [2025]
  hdf5:
    file_pattern: data/<year>.h5
    file_count: 47
    fields_shape: [5, 243, 721, 1440]
    dtype: float32
    variable_count: 243
    time_step_hours: 6
    statistics:
      - global_means
      - global_stds

relations:
  required_datasets: []
  optional_datasets: []
  compatible_models:
    - id: OneScience/FengWu/
      role: train_data
      required_for: [preflight, train, inference, evaluate, visualize]
      resource_ref:
        platform: modelscope
        repo_id: OneScience/FengWu/
        repo_type: model
        url: https://modelscope.cn/models/OneScience/FengWu/
        revision: main
        readme_path: README.md
        manifest_path: onescience_run_manifest.yaml
      expected_local_path: data/ERA5
      adapter: null

run_matrix:
  scenarios:
    - name: minimal_validation
      default: true
      capability: preflight
      required_datasets:
        - id: OneScience/ERA5
          role: train_data
          local_path: data
      required_model_files: []
      command_refs:
        - commands.preflight.validate_dataset
      expected_outputs:
        - ERA5 dataset validation passed
      preconditions:
        - data/1979.h5 和 data/2025.h5 存在且可读。
    - name: full_structure_validation
      default: false
      capability: preflight
      required_datasets:
        - id: OneScience/ERA5
          role: train_data
          local_path: data
      required_model_files: []
      command_refs:
        - commands.preflight.validate_dataset_all_files
      expected_outputs:
        - ERA5 dataset validation passed
      preconditions:
        - 全量数据已下载或链接到 data/。

capabilities:
  preflight: true
  inference: false
  train: false
  finetune: false
  evaluate: false
  visualize: false
  deploy: false

commands:
  download:
    - name: download_full_dataset
      description: 下载 ERA5 标准数据集。
      target: dataset
      repo_id: OneScience/ERA5
      source: .
      local_path: .
      required: true
      required_for: [preflight, train, inference, evaluate, visualize]
      command: modelscope download --dataset OneScience/ERA5 --local_dir .
    - name: download_for_fengwu
      description: 在 FengWu 模型包根目录下载 ERA5 到模型期望路径。
      target: dataset
      repo_id: OneScience/ERA5
      source: .
      local_path: data/ERA5
      required: true
      required_for: [preflight, train, inference, evaluate, visualize]
      command: modelscope download --dataset OneScience/ERA5 --local_dir data/ERA5
  preflight:
    - name: validate_dataset
      description: 快速验证首尾年度 HDF5 文件结构和可读性。
      cwd: .
      command: python scripts/validate_era5_dataset.py --data-dir data
      required_files:
        - data/1979.h5
        - data/2025.h5
        - metadata/schema.yaml
      required_env: [python, h5py]
      expected_outputs:
        - ERA5 dataset validation passed
      success_criteria: [exit_code == 0]
    - name: validate_dataset_all_files
      description: 逐文件验证 1979-2025 全部 HDF5 文件结构和可读性。
      cwd: .
      command: python scripts/validate_era5_dataset.py --data-dir data --check-all
      required_files:
        - data/*.h5
        - metadata/schema.yaml
      required_env: [python, h5py]
      expected_outputs:
        - ERA5 dataset validation passed
      success_criteria: [exit_code == 0]
  prepare: []
  inference: []
  train: []
  finetune: []
  evaluate: []
  visualize: []
  deploy: []

expected_outputs:
  - path: stdout
    type: text
    produced_by: commands.preflight.validate_dataset
    contains: ERA5 dataset validation passed

diagnostics:
  - id: missing_hdf5
    match: no HDF5 files found
    severity: error
    message_zh: data 目录下没有年度 HDF5 文件。
    fix_zh: 运行 modelscope download --dataset OneScience/ERA5 --local_dir .，或检查标准化链接是否存在。
  - id: missing_year
    match: missing expected yearly files
    severity: error
    message_zh: 1979-2025 年度文件不完整。
    fix_zh: 对照 checksums 文件重新整理或下载缺失年份。
  - id: schema_mismatch
    match: expected [T, 243, 721, 1440]
    severity: error
    message_zh: HDF5 fields 数据结构不符合当前 FengWu 适配 schema。
    fix_zh: 使用 OneScience/ERA5 标准数据，或为新 schema 更新模型配置、README 和 Manifest。

domain_extension:
  earth:
    dataset_family: ERA5
    year_range: [1979, 2025]
    file_count: 47
    grid: 0.25_degree
    time_step_hours: 6
    variable_count: 243
    fengwu_selected_channels: 189