A5 — Geospatial big data for Rwanda

Chapter: 10 · Datasets: CHIRPS daily rainfall, Google Open Buildings (Rwanda), WorldPop population, geoBoundaries districts/sectors · Effort: ~15 hours

Goal

Combine raster and vector big data to produce decision-relevant indicators for Rwanda at district and H3 levels.

Tasks

1. Rainfall climatology (25%). With xarray + Dask, compute Season A (Sep–Dec) and Season B (Mar–May) rainfall totals per year (at least 20 years) for each district. Map the long-term mean and the coefficient of variation.
2. Drought years (15%). Compute standardised anomalies per district and season. Identify the three worst seasons in the record and check whether they match known events (cite sources).
3. Built environment (25%). Aggregate Open Buildings (confidence ≥ 0.75) to H3 resolution 7 with DuckDB or Spark: building count and built area per cell. Join with WorldPop aggregated to the same cells.
4. Integrated indicator (20%). Construct an indicator combining population exposure and rainfall variability (justify its form). Map it at H3 resolution 7 and summarise per district.
5. Performance & limitations (15%). Report run times and memory; explain chunking choices. Discuss data quality: building detection errors, population model assumptions, CHIRPS accuracy in mountainous terrain.

Tips

• Clip rasters to Rwanda's bounding box early to reduce data.
• Use EPSG:32735 for area calculations.
• If bandwidth is limited, use CHIRPS monthly files instead of daily for task 1.