About this study
This dashboard presents a research assessment of the October 2023 flood in part of Ghana’s Lower Volta region. The study examines how earlier radar images influence deep learning flood predictions.
What the maps show
The primary event image is Sentinel-1 radar from 23 October 2023. The primary model used 24 August and 5 September as reference dates. An alternative used 17 and 29 September. The valid analysis area covers 4,301.13 km² within the UNOSAT event assessment extent, rather than the entire basin or country.
The primary model predicted 228.29 km² of flooding. Later references produced 184.43 km². These are model estimates, not independently verified flood areas. The difference measures sensitivity to reference choice.
Evidence and limitations
- No independent local flood labels were available for validating the Ghana predictions.
- Training used sigma0 calibration and Lee Sigma filtering. Ghana inputs used gamma0 terrain correction. This processing difference remains.
- Flood probability and uncertainty layers are uncalibrated.
- Agreement with UNOSAT measures agreement between satellite products, not accuracy against field observations.
- Terrain shows elevation, not flood depth. Night lights describe August 2023 radiance, not population, damage or outages.
- Satellite basemap acquisition dates vary. The background is location context, not October 2023 event evidence.
Privacy and review notes
Review notes are stored in your browser. The dashboard does not upload them. Export a CSV copy before clearing browser data. View links contain map settings but no review notes. Hosting requests are processed by Cloudflare.