5  Adaptation Atlas Hazards

5.1 Key Hazard Datasets

5.1.1 Hazard Monthly Summaries (Hazard mean month)

Monthly hazard tables are generated by R/2.1_create_monthly_haz_tables.R and saved in the hazard_timeseries_mean_month/ folder in Parquet format. These datasets are created by aggregating gridded hazard data (e.g., rainfall, temperature) to admin1 units using zonal statistics. Values are then organized into monthly and seasonal groupings, anomalies are calculated against historical baselines, and results are ensembled across multiple GCMs to provide central estimates and inter-model spread. Long-term averages and Sen’s slope trend estimates are also derived to capture climatological means and temporal changes.

5.1.1.1 Output File Types

Table Name Suffix	Description	Example Filename
_seasons.parquet	Monthly values (or seasonal sums) for each model, year, and scenario. Includes anomalies vs baseline.	haz_3months_adm_mean_2061-2080_anomaly-historic_seasons.parquet
_ensemble_seasons.parquet	Same as above but ensembled across GCMs. Includes inter-model stats (mean, min, max, SD).	haz_3months_adm_mean_2061-2080_anomaly-historic_ensemble_seasons.parquet
_ensemble.parquet	Seasonal or annual values averaged over the entire time period. Represents long-term averages per GCM.	haz_3months_adm_mean_2061-2080_anomaly-historic_ensemble.parquet
_trends.parquet	Sen’s slope trend results per GCM per location. Includes slope, intercept, p-value, confidence interval.	haz_3months_adm_mean_2061-2080_anomaly-historic_trends.parquet
_trends_ensemble.parquet	Trend results averaged across GCMs.	haz_3months_adm_mean_2061-2080_anomaly-historic_trends_ensemble.parquet
_trends_ensemble_minimal.parquet	Filtered ensemble trend outputs for specific hazards and stats of interest.	haz_3months_adm_mean_r2061-2080_anomaly-historic_trends_ensemble_minimal.parquet

5.1.1.2 Core Fields in All Tables

Field Name	Description
admin0_name	Country name
admin1_name	First-level administrative unit
scenario	Emissions scenario (e.g., ssp245, ssp126)
timeframe	Future period label (e.g., 2021-2040, 2041-2060)
model	Climate model name (e.g., MPI-ESM1-2-HR); omitted in ensemble tables
hazard	Hazard variable (e.g., PTOT, TAVG, HSH-max)
season	3-month window or annual label (e.g., MAM,JFM, Annual)
baseline_name	Name of the baseline used for computing anomalies (e.g., 1995-2014, 1981-2024)
value	Monthly or seasonal hazard statistic (mean/sum depending on hazard)
anomaly	Difference between value` and historical baseline average

Additional fields in ensemble and trend tables include:

• mean, max, min, sd: Statistical summaries across GCMs
• value_slope, value_decade, value_diff, etc.: Trend metrics using Sen’s slope

5.1.1.3 Note

• Ensemble summaries are computed after anomalies to preserve variance structure across models.

5.1.2 Hazard X Exposure

Here’s a structured markdown draft for your Hazard × Exposure dataset, reflecting the info you provided:

---

5.2 Hazard × Exposure

This dataset quantifies risk-weighted exposure by combining climate hazard frequency data with crop and livestock exposure indicators (e.g., harvested area, production value, livestock numbers). It is a core product of the Africa Agriculture Adaptation Atlas (AAAA) risk modeling pipeline and builds directly on the exposure datasets described in earlier chapters.

5.2.1 Overview

Hazard × Exposure outputs are generated by intersecting hazard rasters with exposure layers. Two main types of outputs are produced:

• solo — Single hazard variables intersected with exposure layers.
• int — Multiple hazard interactions combined before intersection with exposure.

Each intersection is calculated for three severity levels: moderate, severe, and extreme. Outputs include both historic periods and future ensemble projections based on climate models, with separate calculations for annual totals and seasonal windows (e.g., Jagermeyer seasons).

Hazard × Exposure layers are primarily stored as Cloud-Optimized GeoTIFFs (COGs), and supporting .parquet tables are provided for analysis at subnational administrative levels (admin0–2).

5.2.2 File Structure

<temporal_resolution>/<crop_or_livestock>_<period>_<severity>_<hazard_code>_<type>_<exposure_metric>_<units>.tif

5.2.3 Key Notes

• Exposure data include crop VoP (nominal USD & international dollars), harvested area (ha), and livestock numbers or VoP.
• Hazard layers include annual and seasonal frequencies from climate projections or historical observations.

5.2.4 Output Format

• Raster COGs for each hazard × exposure × severity × period combination.
• Parquet tables summarizing exposure by admin unit for integration into downstream analyses.

Note: int outputs should be used when analyzing combined hazard interactions, whereas solo outputs are appropriate for single-hazard assessments. Severity level selection (moderate, severe, extreme) should match the risk appetite of the analysis.