Skip to content
CGIAR Climate Action CGIAR Climate Data Hub — Wikis

CGIAR Climate Action CMIP7 hazards pipeline

What this page covers

The methodology and pipeline that turns the CGIAR-led CMIP7 downscaled and bias-corrected dataset into hazard indicators — the analysis-ready variables that feed the AAA Adaptation Atlas, the Build a Climate Rationale notebook, and partner-facing climate-rationale tools across CGIAR Climate Action.

For the partner-facing transition overview (what CMIP7 is, when it lands, what proposal authors should do today), see What’s next — CMIP7 and CORDEX-Africa.

Scope

To be drafted by Pete / Harold / Brayden.

This pipeline replaces the existing CMIP6-era hazards processing in the hazards_prototype and atlas_notebooks codebases. Its outputs are the variables a climate-rationale author actually cites — degree days, dry-spell length, extreme-precipitation thresholds, heat indices, drought indicators — at the spatial and temporal scales the AAA Adaptation Atlas surfaces.

Hazard variables to compute

To be drafted. The starting list inherits the CMIP6-era hazards catalogue:

  • Thermal: degree days (growing, heating, cooling); heat-stress indices (WBGT, UTCI); consecutive days >35 °C; tropical-nights count.
  • Precipitation: total rainfall (annual, seasonal); R10mm, R20mm, R95p; consecutive dry days (CDD); consecutive wet days (CWD); RX1day, RX5day.
  • Drought: SPI / SPEI at multiple timescales; soil-moisture deficit.
  • Composite: heat × drought concurrence; growing-season length; rainfall-onset / cessation; agroclimatic-zone shifts.

Each hazard variable carries:

  • A defined calculation method (ETCCDI / WMO / sector-specific)
  • A reference period and baseline
  • A spatial / temporal aggregation (gridded, admin-1, country-level)
  • An ensemble-aggregation convention (mean, percentile spread)

Methodology — per hazard

To be drafted by Pete / Harold / Brayden.

Suggested points for each hazard:

  • Calculation method + reference
  • Daily-vs-monthly input handling
  • Threshold conventions (fixed vs percentile, climatology-relative vs absolute)
  • Bias-correction implications — which hazard variables are robust to which BC methods
  • Known issues for African regions (e.g. SPI for the Sahel; degree-day formulations for tropical crops)

Compute infrastructure — shared with the downscaling pipeline

To be drafted by Pete / Harold / Brayden in coordination with Carlos Navarros.

The hazards pipeline runs on the same compute and storage infrastructure as the downscaling pipeline. Decisions taken upstream (cloud vs HPC, workflow orchestrator, storage format, region) propagate downstream.

Suggested design considerations:

  • Read pattern — hazards processing reads downscaled daily data sequentially over time and spatially in chunks. Zarr chunked appropriately is much faster than netCDF for this access pattern.
  • Compute envelope — typically 1–2 orders of magnitude smaller than the downscaling step (hazards are derived statistics, not regridded fields).
  • Parallelism — embarrassingly parallel by model × scenario × hazard; suits Dask / Snakemake.
  • Caching — intermediate hazard outputs cached at multiple aggregation levels (gridded, admin-1, country).

Data products produced

To be drafted by Pete / Harold / Brayden.

Tentative product list:

  1. Gridded hazard layers — netCDF/Zarr, full African continent, per-model and ensemble-aggregated.
  2. Admin-1 / country-level tabular indices — parquet, per-country time series, suitable for direct ingest into the Build a Climate Rationale notebook.
  3. Threshold-exceedance maps — when does each grid cell first cross a hazard threshold under each scenario.
  4. Climate-rationale ready summary indices — pre-computed change signals (e.g. “+X% rainfall in MAM by 2050”) with uncertainty bounds, designed for direct quotation in GCF / Adaptation Fund concept notes.

Integration with the AAA Adaptation Atlas

To be drafted.

  • Read path: AAA Adaptation Atlas’s parquet layer consumes the admin-1 / country tabular indices.
  • Versioning: hazards-pipeline output versioned per CMIP7 ensemble release; AAA Atlas pins specific hazard-pipeline versions for citation stability.
  • Backward compatibility: existing CMIP6 hazards layers retained until partners citing CMIP6 work close out their cycles.

Validation and QA/QC

To be drafted by Pete / Harold / Brayden.

  • Reference data — observed hazard indices from CHIRTS / CHIRPS / ERA5-Land for the historical period.
  • Regression vs CMIP6 hazards pipeline — hazards under CMIP7 inputs vs hazards under CMIP6 inputs for the historical overlap period.
  • Internal consistency — composite hazards (e.g. heat × drought) should be self-consistent across spatial and temporal aggregations.
  • Sector spot-checks — agronomic colleagues at Alliance Bioversity-CIAT / ILRI / IFPRI on whether the hazard fields look right for their crops / livestock systems.

Timeline and dependencies

To be drafted. Key dependency: the downscaling pipeline must publish before the hazards pipeline can ingest.

QuarterMilestoneDependency
2026 Q4Hazards catalogue review; spec consolidationNone blocking
2027 Q1Pipeline scaffold built; runs end-to-end on CMIP6 pilot inputsDownscaling pipeline pilot
2027 H2Pipeline running against first CGIAR-AAA-CMIP7 outputsDownscaling v0.1 release
2028 H1First CMIP7 hazards-layer release; integration into AAA Atlas testDownscaling production release
2028 H2 / 2029Production CMIP7 hazards layer live in the AAA Adaptation AtlasAll upstream + validation complete

Open design questions

To be answered during pipeline scoping:

  • Which hazards to compute first — the highest-cited subset, or full parity with CMIP6 hazards day one?
  • How to handle the East African paradox in projected hazards — flag-and-disclose, weight-down, or exclude affected models from East African long-rains-derived hazards?
  • Use the AFR-13 sub-ensemble equivalent or run all CMIP7 models then sub-set per hazard?
  • Output format priority — Zarr (cloud-native) or netCDF (community legacy)?

Further reading