1. Climate Data Sources

The climate risk analysis is grounded in a comprehensive analysis of historical climate data (1981–2024) obtained from observation (satellite) and model-based datasets. Specifically, the Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS) and the fifth-generation ECMWF atmospheric reanalysis dataset (AgERA5) were utilized to provide high-resolution historical data. The choice of including these datasets is based on their well-documented strengths in climate risk analysis, particularly for agricultural applications in Africa.

• CHIRPS provides high-resolution precipitation estimates by blending satellite imagery with station data, ensuring better coverage in data-sparse regions like Mozambique.
• AgERA5, derived from ECMWF’s ERA5, is optimized for agricultural modelling, offering consistent, high-quality surface climate data.

These datasets were chosen over alternatives (e.g., GPCC, CRU, ERA5-Land) due to their superior validation records, high spatial resolution, and extensive usage in climate risk studies. Future climate projections were derived from five Global Climate Models (GCMs) participating in Phase 6 of the Coupled Model Intercomparison Project (CMIP6). The following five GCMs from CMIP6 were selected:

• ACCESS-ESM1-5 (Australia’s CSIRO and BOM) – Strong representation of large-scale atmospheric circulation.
• MPI-ESM1-2-HR (Max Planck Institute, Germany) – High-resolution model known for accurate precipitation simulations.
• EC-Earth3 (EC-Earth Consortium, Europe) – Advanced coupled system for ocean-atmosphere interactions.
• INM-CM5-0 (Institute of Numerical Mathematics, Russia) – Effective in representing land-surface processes.
• MRI-ESM2-0 (Meteorological Research Institute, Japan) – Robust performance in extreme climate event projections.

These models were chosen to ensure a balance between different climate system representations and climate sensitivity, ensuring comprehensive climate risk projections. Projections were developed for two future timeframes: the near future (2021–2040) and mid-century (2041–2060). The analysis considers two emission scenarios defined by the Shared Socioeconomic Pathways (SSPs): SSP245 and SSP585. SSP245 represents a moderate emissions scenario aligned with efforts to limit global warming to below 2°C above pre-industrial levels, while SSP585 represents a high-emissions scenario with significantly less mitigation action.

We applied statistical downscaling and bias correction using a monthly climate anomaly approach, ensuring that GCM outputs align with historical observations. The bias correction process involves:

• Computing Monthly Climate Anomalies: Differences between simulated and observed climate conditions are calculated for each month.
• Applying These Anomalies to Historical Data: The computed anomalies are then superimposed on historical CHIRPS (precipitation) and AgERA5 (temperature) datasets.
• Correcting Systematic Biases: The approach accounts for model-specific deviations, improving the reliability of future projections.

This method ensures that climate projections maintain consistency with historical climate trends while correcting biases in raw GCM outputs. After the downscaling, all historical and projection datasets were standardized at 0.05° spatial resolution.

In alignment with the World Meteorological Organization (WMO) guidelines, baseline reference periods typically span 30 years (e.g., 1971–2000 or 1981–2010). However, for this study, the WMO’s proposed 20-year baseline interval (1995–2014) was adopted. This interval encompasses the final two decades of historical simulations using observed radiative forcings, providing a robust baseline for comparison with selected future time windows. This approach ensures the analysis remains relevant and consistent with contemporary climate assessment standards.

Several indicators were calculated to assess historical and projected climate trends as listed in table below. These metrics were selected to capture the key climatic factors influencing the region's environmental and socio-economic systems, providing a robust foundation for climate impact assessments and adaptation planning.

Climate variables and indices

2. Z-Scores

A Z-score is a way of measuring how unusual or extreme a value is compared to the average (mean) of a dataset. It tells us how far a value is from the average, scaled by how spread out the data is (standard deviation). A positive Z-score means the value is above average, while a negative Z-score means it’s below average.

In this report, Z-scores are used to identify years with particularly high or low rainfall in Mozambique’s regions:

• Z-scores greater than +1.5 indicate years with much more rainfall than usual (“extreme wet years”).
• Z-scores less than -1.5 indicate years with much less rainfall than usual (“extreme dry years”).

For example, a Z-score of +1.5 means that the rainfall for that year was one standard deviation above the average rainfall, which is moderately higher than usual. Similarly, a Z-score of -1.5 means the rainfall was one standard deviation below the average, which is moderately lower than usual. Values beyond these thresholds are considered extreme, helping to highlight significant climate variability and patterns like consecutive dry or wet years, as well as sudden shifts between extremes.

Formula: Z = (X – μ) / σ

• Z : The Z-score
• X : The rainfall value for a specific year
• μ : The average rainfall over the entire dataset
• σ : The standard deviation of rainfall across the dataset

By standardizing the rainfall data in this way, Z-scores make it easier to compare and visualize periods of extreme weather and their impact across different regions.

3. Exposure to Risk (Value of Production)

Crop value of production data comes from MapSPAM 2017 V2r3 (Spatial Production Allocation Model) where production values are multiplied by country specific FAOstat international crop prices. Livestock value of production in 2005 International Dollars were provided by the authors of Herrero et al. (2013).

The process for calculating exposure to risk comprises several critical steps:

• Annual or Seasonal Summarization: Raw hazard data is aligned with the GGCMI Phase 3 crop calendar for maize, then climate hazards are summarized at either annual or seasonal timescales. Future updates of the Atlas will incorporate crop-specific calendars.
• Occurrence: Hazards classified into severity levels (moderate, severe, extreme). This study applies severe thresholds.
• Risk: We average the classified hazards over the years or seasons encompassed by the time-series to obtain an estimation of occurrence risk.
• Compound Risks: To assess the risk of compound hazards, we multiply the risks associated with dry, heat, and wet conditions. This multiplication highlights areas where multiple hazards may coincide, posing greater risk to agricultural outputs.
• Integration with Economic Data: Finally, we multiply the compound crop hazard risk data with crop value of production data. For example, if a crop is exposed to a hazard in 50% of growing seasons and the value of production is $2M/year then, on average, $1M of production is exposed to the hazard. The exposed value of production is then extracted by administrative areas to indicate potential economic impacts on crop production.