Context & Background

2025 analysis

Approach

The 2025 analysis combined multiple forecasting and observational data sources to develop a flood trigger for the Wuroboki station (Adamawa state):

Source	Role
GloFAS reanalysis (1998–2024)	Primary discharge signal (`dis24`)
Google GRRR reanalysis & reforecast	Complementary streamflow signal
Floodscan SFED	Observational ground truth (satellite flood extent)
NiHSA water levels	Observational records at Wuroboki

Trigger definition

A dual-source OR trigger was found to give the best balance of detection accuracy and lead time:

Source	Threshold	Approx. RP
GloFAS	3,132 m³/s	~5.4 yr
Google GRRR	1,195 m³/s	~5.4 yr

Logic: trigger activates when either GloFAS or Google GRRR exceeds its threshold. The combined activation frequency gives an overall ~4.5-year RP, targeting 5-year RP flooding events (as required by CERF).

Floodscan wet-season (Jul–Nov) annual maxima were used as observational ground truth for return period calibration.

Key findings

GloFAS peaks typically 10–20+ days ahead of observed flooding; Google GRRR provides 15–25 days lead time
The OR-logic trigger maximised F1 score while maintaining 20+ days average lead time
Mean pixel aggregation (rather than admin-level aggregation) was selected for Floodscan
Analysis was based on reanalysis throughout; operational reforecasts showed lower but acceptable performance

Stations

Station	River	State	GloFAS key	Google HYBAS
Wuroboki	Benue	Adamawa	`wuroboki`	`hybas_1120842550`
Makurdi	Benue	Benue	`makurdi`	`hybas_1120911340`

Data sources

Dataset	Path / location	Notes
GloFAS reanalysis	Azure blob	Via `src.datasources.glofas`
Google GRRR reanalysis	`gs://flood-forecasting/…/reanalysis/streamflow.zarr/`	Via `src.datasources.grrr`
Google GRRR reforecast	`gs://flood-forecasting/…/reforecast/streamflow.zarr/`	Via `src.datasources.grrr`
Google HYBAS locations	`gs://flood-forecasting/…/hybas_outlet_locations_UNOFFICIAL.zarr/`	Gauge lat/lon only
Floodscan SFED	Azure blob — `floodscan/daily/v5/processed/`	300s COGs, 1998–2025
Processed Floodscan (Benue)	`ds-aa-nga-flooding/processed/floodscan/fs_benue_state_pixels_1998_2025.parquet`	Pixel-level, ~10km river buffer
Processed Floodscan (Adamawa)	`ds-aa-nga-flooding/processed/floodscan/fs_adamawa_pixels_1998_2025.parquet`	Pixel-level, ~10km river buffer
NiHSA risk communities	`ds-aa-nga-flooding/raw/AA-nigeria_data/NiHSA/AFO_communities_atrisk_2026.csv`	`lat`, `lon`, `depth_zone`
Google inundation history	`ds-aa-nga-flooding/processed/google_inundation_history/combined_nga.parquet`	Geoparquet
WorldPop 2020 (1km)	WorldPop STAC — `nga_pop_2020_…_1km_…`	Via `worldpop.load_worldpop_from_stac()`