National flood risk assessment: what it involves and how to commission one
A national flood risk assessment shows where floods would do most harm across a whole country, and to whom, so a government knows where to look first. In Ghana I screened all 260 districts from open global data, tested the result nine ways and found 13 districts that stay at the top in all nine.
- 13 Act
- 13 Investigate
- 234 Monitor
How to read this: each shape is one of Ghana’s 260 districts. Orange districts are in the national top 20 for modelled river flood damage in all nine tests, amber in some, grey in none. Grey does not mean safe. Not an official priority list.
- districts of Ghana screened the same way, from open global data
- 260
- district-level damage calculations in a Monte Carlo (Latin hypercube) analysis
- 11.7million
- districts in the national top 20 for modelled damage under all nine combinations of assumptions
- 13
- estimated value of the Ghana river assessment if contracted; US$155k with the coastal screen
- est. US$90k
About the author. Hisham Nuhu is a UK-based flood and disaster risk specialist. He worked at England’s Environment Agency from 2019 to 2025 and holds an MSc in Water Science, Policy and Management from the University of Oxford. He works remotely or in-country with governments, development banks, UN agencies and consulting firms, and in England with councils, communities, landowners and developers.
What is a national flood risk assessment, and what decisions does it support?
It is a screen of flood hazard, exposure and vulnerability for a whole country, done the same way everywhere so places can be compared. It supports three decisions: where to spend on detailed studies first, which communities to put first for warning and preparedness, and how big the national problem is. It decides where to look, not what to build.
In Ghana the answer was strikingly concentrated. Twenty of the 260 districts hold 86.8% of the modelled annual flood damage to homes, and the top five hold 49%. Most sit on the lower Volta and the edge of Greater Accra.
That concentration is what makes a national screen worth paying for. It turns 260 local authorities into a short list where a local study, a conversation with the communities affected or a check on preparedness is money well spent.
The detailed tier comes next. At Hamstall Ridware in England the hydrology was entirely my work, and I built a 1D–2D model of one village jointly with an Environment Agency modelling colleague. A national screen tells you where that kind of study should go.
Five steps from a national flood map to a short list, with what each gave in Ghana
Hazard: where the water goes
Flood depths for several flood sizes on one national grid. Ghana: open global flood maps for five flood sizes, 1 in 10 to 1 in 500 years, at 90 m.
Exposure: who is in the way
Homes and people inside each flood. Ghana: about 13 million building footprints; 1.18 million people inside the 1-in-100-year flood.
Vulnerability: how much harm
The share of a home’s value lost at each depth. Ghana: three damage curves tested side by side, because none comes from Ghanaian records.
Risk: harm and how likely it is
Average damage a year, with its range. Ghana: about GHS 5.19 billion a year to homes and contents, from 11.7 million district-level damage calculations.
Decision: where to act first
A short list that holds under every assumption. Ghana: 13 Act, 13 Investigate and 234 Monitor districts.
How to read this: the steps light up in turn; tap, or tab to, any step to read it.
What data does it need, and can it be done with open global data?
Five layers: flood depths for several flood sizes, buildings, people, what the buildings are worth, and how much damage each depth does. In Ghana all five came from open data: JRC flood maps, Google Open Buildings, WorldPop, night-time lights and the 2021 census. Local data makes the result better. It is not a reason to wait.
The value layer needs the most care. I built a map of what homes would cost to rebuild from night-time lights, calibrated at ten Ghanaian neighbourhoods. At district level it ranks closely with census welfare (Spearman ρ = 0.83). It gets the order right; it does not give a precise bill.
The damage layer is the weakest everywhere I have looked. No depth–damage curve has yet been derived from Ghanaian flood records, so I ran three from elsewhere as alternatives rather than averaging them. At 1 m depth they disagree widely: 10%, 38% and 59% of a home’s value.
| Layer | Open data used in Ghana | Local data that improves it |
|---|---|---|
| Flood hazard | JRC global river flood hazard maps (version 2), five flood sizes on a 90 m grid | Surveyed channels and defences; satellite maps of past floods; local hydraulic models |
| Buildings | Google Open Buildings (version 3), about 13 million footprints | Building use, storeys and floor levels in flood-prone streets |
| People | WorldPop 2026; 2021 census shares of children, older people and people with disabilities | Current settlement data, including informal settlements |
| Value | Rebuilding cost from VIIRS night-time lights, calibrated at ten neighbourhoods | Construction cost surveys; insurance or recovery records |
| Damage | Three depth–damage curves: JRC Africa, a Sri Lanka curve selected for Accra, and Ouagadougou | Damage records from the country’s own past floods |
How do you show the uncertainty?
By running the whole chain many times and watching what changes. In Ghana I varied seven uncertain inputs 1,000 times (Monte Carlo, Latin hypercube) under three flood edges and three damage curves: 11.7 million district-level damage calculations. Where the flood edge sits mattered most. It moved national annual damage from GHS 2.93 billion to 7.33 billion.
The three flood edges came from checking the map against a real flood. Against the 2020 White Volta flood the model usually sat within a few hundred metres of the satellite edge, but in one test area it missed water up to 2.8 km away. So I ran the map as published, with its edge pulled in two cells and pushed out one.
The nine combinations carry no probabilities, and I did not average them. I asked a simpler question: which districts stay a priority in all of them?
Bigger, rarer floods do more damage, and where the flood edge sits matters most
How to read this: left to right, floods get rarer and bigger. The line is the median damage and the band its P5–P95 range; the dashed lines are the other two flood edges. Switch the flood edge, or tap a flood size, to read it.
Modelled damage to homes and contents from one river flood across all 260 districts, JRC Africa damage curve. Model estimates, not reported losses.
Show the data as a table
| Flood | Narrower flood edge, GHS bn (P5–P95) | Flood edge as mapped, GHS bn (P5–P95) | Wider flood edge, GHS bn (P5–P95) |
|---|---|---|---|
| 1 in 10 years (10% a year) | 19.5 (16.1–23.4) | 38.5 (31.8–46.1) | 58.5 (48.3–70.1) |
| 1 in 50 years (2% a year) | 35.5 (29.3–42.6) | 60.6 (50.1–72.6) | 83.5 (69.0–100.1) |
| 1 in 100 years (1% a year) | 40.5 (33.4–48.5) | 67.2 (55.5–80.5) | 90.7 (74.9–108.7) |
| 1 in 200 years (0.5% a year) | 44.4 (36.7–53.2) | 73.1 (60.4–87.6) | 97.5 (80.5–116.8) |
| 1 in 500 years (0.2% a year) | 49.5 (40.9–59.4) | 80.8 (66.7–96.8) | 106.2 (87.7–127.3) |
| Annual average (1 in 10 to 1 in 500) | 2.93 a year (2.42–3.52) | 5.19 a year (4.28–6.21) | 7.33 a year (6.05–8.78) |
How do you decide where to act first?
Rank every district in every combination of assumptions and keep the ones that stay near the top every time. In Ghana, 13 districts were in the national top 20 for modelled damage in all nine combinations (Act), 13 in some (Investigate) and 234 in none (Monitor). Act means appraise first. Monitor does not mean safe.
Ningo/Prampram, Keta and South Tongu lead the Act list. The full list, with damage and people for each district, is on flood-prone districts in Ghana.
Weija Gbawe, on the Densu west of Accra, shows why the test matters. Its rank moves from 1st to 11th across the nine combinations, yet it never leaves the top 20. A single model run would have hidden that.
The cut-off is a policy choice, so agree it with the people who will use the list. A top-10 rule gives 7 Act districts; a top-30 rule gives 20.
The 13 Act districts cluster on the lower Volta and around Accra
How to read this: switch between priority class, damage and people; tap a district, or choose one from the list, to read its figures. “Accra to Keta” zooms in on the cluster.
Modelled river flooding for all 260 districts. Damage is to homes and their contents; people are those living inside the modelled 1-in-100-year flood.
Sources
Independent consultancy (2025–26), also my Oxford MSc dissertation. Hazard: JRC global river flood hazard maps v2 (CC BY 4.0). Buildings: Google Open Buildings v3. People: WorldPop (CC BY 4.0). Boundaries: OCHA/HDX COD-AB (Ghana Statistical Service).
Show the 13 Act districts as a table
| District | Region | Damage to homes a year, GHS m (median) | People in the 1-in-100-year flood extent | National rank across 9 combinations | Class |
|---|---|---|---|---|---|
| Ningo/Prampram | Greater Accra | 631.0 | 47,692 | 1st–3rd | Act |
| Keta Municipal | Volta | 598.6 | 112,294 | 1st–3rd | Act |
| South Tongu | Volta | 476.4 | 89,113 | 2nd–4th | Act |
| Ketu South | Volta | 421.3 | 101,768 | 4th–6th | Act |
| Weija Gbawe Municipal | Greater Accra | 414.4 | 51,164 | 1st–11th | Act |
| Ada East | Greater Accra | 371.6 | 63,149 | 5th–6th | Act |
| Anloga | Volta | 369.6 | 63,481 | 4th–7th | Act |
| North Tongu | Volta | 307.4 | 66,672 | 7th–8th | Act |
| Ada West | Greater Accra | 130.6 | 42,492 | 8th–12th | Act |
| Central Tongu | Volta | 107.8 | 27,705 | 9th–17th | Act |
| Shai Osudoku | Greater Accra | 101.3 | 17,392 | 10th–15th | Act |
| Shama | Western | 97.9 | 23,030 | 9th–20th | Act |
| Asuogyaman | Eastern | 70.9 | 15,552 | 13th–19th | Act |
How do you count people as well as assets?
Count them separately, then compare the two rankings. In Ghana about 1.18 million people live inside the modelled 1-in-100-year flood, about 421,700 of them children. The two rankings share the same top ten, then part: Mamprugu Moagduri, in North East Region, is 32nd for people exposed and 94th for damage.
Damage follows money. The model values homes from building size and night-time light, so the same depth of water costs less to repair where incomes are lower. A ranking by damage alone drifts away from where people live.
So I recommend a people-based criterion beside the economic one, with weights agreed with national partners. Switch Fig. 3 to “People exposed” and the northern districts appear.
How do you know the model matches reality?
Test it against a flood that really happened. I compared Ghana’s modelled flood with Sentinel-1 satellite maps of the September 2020 White Volta flood in five 20 km squares. Agreement averaged 0.58 on the critical success index. The model also put 6,371 people inside the flooded area; NADMO, the national disaster agency, reported 7,561 affected.
A score of 1 would mean the model and the satellite flood match exactly. In a published comparison, global flood models scored a mean of 0.34 (JRC) and 0.30 (Fathom, CaMa-Flood) for 1-in-100-year maps. The tests differ, so treat that as context, not a ranking.
NADMO’s count is for the Upper East Region, while the satellite test areas lie further south, in North East and Northern regions. The two do not cover the same ground, so I read them only as a sign the people layer is in a plausible range, not as a check.
One event in one region checks the method. It does not prove every district right. More on that flood: the White Volta flood of 2020.
How long does it take, and what does it cost?
It depends mostly on data access and on how many review rounds the client wants. In Ghana I did the river screen alone, part-time over 2025–26; it was also my Oxford MSc dissertation. Benchmarked against comparable Environment Agency products, the river assessment is worth an estimated US$90k, or US$155k with the coastal screen.
A commissioned assessment adds steps my independent work did not need: data-sharing agreements, an inception workshop, a validation workshop with national agencies, and sign-off by the agencies that will own the results.
For Ghana I proposed a 90-day scoping phase before any detailed study: agree a government owner and the decision, review existing models and candidate sites, then cost the local appraisal.
What should the terms of reference include?
Eight things: the decision the results must inform, the flood types in scope, the unit of analysis, the flood sizes, how uncertainty will be tested, a check against a past flood, results for people as well as assets, and who will own and update the work. Leave one out and the ranking is hard to defend later.
| Item | What to specify | How Ghana handled it |
|---|---|---|
| 1. The decision | Which decision the ranking informs, and who takes it | Where to send local appraisal first |
| 2. Flood types | River, coastal, surface water: say which are in and which are out | River flooding; the coast screened separately; drainage flooding out of scope |
| 3. Unit of analysis | Districts, sub-basins or grid cells | 260 districts, summed from a 90 m grid |
| 4. Flood sizes | Enough return periods to draw a loss curve | Five, from 1 in 10 to 1 in 500 years |
| 5. Uncertainty | The alternatives to test, and the rule for a priority that holds | Three flood edges and three damage curves, 1,000 samples each; top 20 in all nine |
| 6. Validation | A past flood with satellite or field records, and the score to report | White Volta, September 2020; critical success index |
| 7. People | Who is counted, beyond totals | Children, people aged 65 and over, people with disabilities |
| 8. Ownership | Who hosts, updates and explains the results; open data and documented code | Proposed as the first step of a 90-day scoping phase |
How does it fit the Sendai Framework and Early Warnings for All?
It is the evidence behind Sendai Priority 1, understanding disaster risk, and the first pillar of Early Warnings for All, disaster risk knowledge. A district list tells a warning service where people sit in the flood’s way. It also tells a ministry where investment in risk reduction, Sendai Priority 3, should be appraised first.
FrameworksUN Sendai Priority 1, understanding disaster risk; Early Warnings for All pillar 1ADB project identification; risk-based assessment for river basinsWorld Bank evidence for a Project Concept Note; disaster risk analytics
Alignment of language and method, not certification or endorsement. More on warning: risk information for early warning.
Who can carry out a national flood risk assessment?
Usually a consulting firm or a small team of experts, working with the national disaster agency, the hydrological and meteorological services and the statistics office. I do this work as an individual consultant or as a key expert in a firm’s team. Ghana’s 260-district screen is my worked example, presented for discussion with the World Bank.
The Ghana work was independent consultancy, and the river part was also my Oxford MSc dissertation. I am now sharing the results with the World Bank’s Accra-based team.
I work from the UK, remotely or in-country, and can provide a CV in World Bank, ADB or UN P11 format. My record is summarised on the CV page, and every service is on Services.
What does a national screen not show?
Stated plainly, so the results are used for what they can support.
One kind of flooding at a timeGhana’s river screen leaves out drainage flooding in towns and dam releases. The coast was screened separately.
Homes onlyThe money figures count homes and their contents, not roads, farms, businesses or lives.
Screening resolutionA 90 m grid compares districts. It cannot say whether one building floods.
A partial annual figureThe annual damage covers floods from 1 in 10 to 1 in 500 years only.
Questions people ask
Can the same method be used in another country?
Yes. Every input is open global data, so the chain can be rebuilt for another country or a river basin, including the nine-way test. The Ghana numbers do not transfer; the method does.
Is a national screen enough to design a flood scheme?
No. It tells you where to look. Designing a scheme needs local hydrology, survey and a hydraulic model, the kind of work described on river hydrology and 1D–2D modelling.
What is the difference between hazard, exposure and vulnerability?
Hazard is where the water goes, how deep and how often. Exposure is who and what is in the way. Vulnerability is how much harm a given depth does. Risk combines all three with how likely each flood is.
What is expected annual damage?
The damage to expect on average each year, with each flood weighted by its chance of happening. In Ghana’s model it is about GHS 5.19 billion a year for homes and contents, counting floods from 1 in 10 to 1 in 500 years.
Does a national assessment cover coastal flooding?
It can, but the coast needs its own method: extreme sea levels from tide, surge and waves. For Ghana I screened the coast separately. See coastal extreme water levels and inundation screening.
Sources
- Nuhu, H. (2026). National fluvial flood-risk assessment and priority stability under uncertainty in Ghana. MSc dissertation, University of Oxford. Independent consultancy, presented for discussion with the World Bank. Method and results.
- UNDRR (2015). Sendai Framework for Disaster Risk Reduction 2015–2030.
- UNDRR (2017). Words into Action guidelines: national disaster risk assessment.
- United Nations. Early Warnings for All.
- European Commission Joint Research Centre. Global river flood hazard maps (CC BY 4.0).
- Google Research. Open Buildings.
- WorldPop, University of Southampton. Gridded population (CC BY 4.0).
- Earth Observation Group. VIIRS night-time lights.
- Huizinga, J., de Moel, H. and Szewczyk, W. (2017). Global flood depth-damage functions. European Commission Joint Research Centre.
- Copernicus Emergency Management Service. Global Flood Monitoring, Sentinel-1 flood extent, September 2020.
- Ghana Statistical Service and OCHA. Ghana subnational administrative boundaries (COD-AB).
- FloodList (20 September 2020). Ghana: 5 dead after floods in Upper East Region, quoting NADMO: 7,561 people affected and 1,596 homes damaged in the Upper East Region.
- Risling, A., Lindersson, S. and Brandimarte, L. (2024). A comparison of global flood models using Sentinel-1 and a change detection approach. Natural Hazards 120, 11133–11152. doi:10.1007/s11069-024-06629-7.
Work with me
Need a national or basin flood risk picture you can defend?
Send the terms of reference, or the decision the work must inform. I will reply with how I would scope it, what data it needs and when I am available, remotely or in-country.