Post on 09-Jun-2020
New estimates of flood exposure in developing
countries using high-resolution population data
Jeff Neal1,2,
Andrew Smith2, Chris Sampson2, Niall Quinn2, Paul Bates1,2
1 School of Geographical Sciences, University of Bristol, UK
2 Fathom Global, Engine Shed, Bristol, UK
Overview of findings
• Current estimates of global flood
exposure are made using data sets that
distribute population counts across large
areas of lowland floodplain.
• When intersected with simulated water
depths, this results in a significant mis-
estimation of flood risk.
• Here, we use new highly resolved
population information from the
Facebook/Columbia University High
Resolution Settlement Layer (HRSL)
and hazard data from Fathom.
• We find that humans make more
rational decisions about flood risk than
current demographic data suggest.
Overview of presentation
• Flood hazard data
• Fathom global flood model
• 1 in 100 year return period
• Intersect with population
• Worldpop
• LandScan
• HRSL
• Breakdown results into
urban, semi-urban or rural
• Global Human Settlement
Layer
Sampson, C. C., A. M. Smith, P. D. Bates, J. C. Neal, L. Alfieri, and J. E.
Freer (2015), A high-resolution global flood hazard model, Water Resour.
Res., 51, 7358–7381, doi:10.1002/2015WR016954.
How the global flood model works
Smith, A., C. Sampson, and P. Bates (2015), Regional flood
frequency analysis at the global scale, Water Resour. Res., 51,
539–553, doi:10.1002/2014WR015814.
How the global flood model works
Neal, J., G. Schumann, and P. Bates (2012), A subgrid
channel model for simulating river hydraulics and floodplain
inundation over large and data sparse areas, Water
Resour. Res., 48, W11506, doi:10.1029/2012WR012514.
St Louis Stage Validation
Illinois River at Hardin(1 in 1000) 43.3 ft(1) 42.4 ft on 08/03/1993(1 in 100) 39.3 ft(2) 38.2 ft on 04/29/1973(3) 36.7 ft on 05/29/1995(4) 36.5 ft on 04/14/1979(5) 36.3 ft on 04/26/2013(6) 35.8 ft on 06/28/2008(7) 34.8 ft on 05/16/2002(8) 34.6 ft on 04/27/1993(9) 34.5 ft on 10/09/1986(10) 34.0 ft on 03/09/1985(1 in 10) 33.8 ft
Mississippi River at St Louis(1 in 1000) 51.6 ft (1) 49.6 ft on 08/01/1993(1 in 100) 48.1 ft(2) 43.2 ft on 04/28/1973(1 in 10) 42.8 ft (3) 42.0 ft on 04/01/1785(4) 41.9 ft on 05/22/1995(5) 41.3 ft on 06/27/1844(6) 40.5 ft on 06/04/2013(7) 40.3 ft on 07/02/1947(8) 40.2 ft on 07/22/1951(9) 39.3 ft on 12/07/1982(10) 39.1 ft on 10/09/1986
Recent large-scale flood exposure analysesModel/ study name Hydrology component Flow routing
component
Inundation data resolution
after downscalingPopulation data set Population data resolution
GLOFRIS
Ward et al.3
PCR-GLOBWB (0.5 degree) driven by
EU-WATCH reanalysis 1960–1999
Kinematic wave
0.5 deg
30 arc sec
~900 m
LandScan™
Bhaduri et al.14
30 arc sec
~900 m
CaMa-UT
Hirabayashi et al. 37
MATSIRO-GW (1 degree)-driven by
JRA-25 Reanalysis 1979-2010
+GPCP rain gauge correction
Inertial wave
0.25 deg
18 arc sec
~540 m
Gridded Population of the
World (GPW) version 3
CIESIN and CIAT38
2.5 arc minutes
~4500 m
CIMA-UNEP
(GAR, 39)
Regional FFA from global gauge data
+ ECEarth bias corrected
Manning’s equation at
multiple points
3 arc sec
~90 m
LandScan™
Bhaduri et al. 14
Data aggregated to 1x1km
within 10km of a coastline,
5x5km elsewhere
GLOFRIS, CaMA-UT,
CIMA-UNEP, Fathom-
Global90 (formerly known
as SSBN), JRC, ECMWF
Trigg et al. 10
Various Various 3-30 arc sec
~90-900 m
WorldPop
Stevens et al. 15
Data aggregated to 30 arc
sec resolution to match
coarsest model output
~900 m
CaMa-UT
Kinoshita et al. 40
MATSIRO-GW (1 degree)-driven by
JRA-25 Reanalysis 1979-2010
+GPCP rain gauge correction
Inertial wave
0.25 deg
18 arc sec
~540 m
History Database of the
Global Environment (HYDE)
version 3.1 for the year 2005
Goldewijk et al. (2010)18
5 arc minutes
~10 km
Fathom-US
Wing et al 21
Regional FFA from global gauge data Inertial wave
1 arc second ~30 m
No downscaling
1 arc sec
~30 m
US Environme-ntal
Protection Agency (EPA)
EnviroAtlas
Pickard et al. 41
1 arc sec
~30m
A non-exhaustive summary of recent large-scale flood risk analyses and the
population data sets used (after Trigg et al.).
Population exposure datasets
• Landscan
• Dasymetric downscaling approach using spatial data and imagery to
disaggregate census data to a regular grid, 30 arc sec, ~900 m
• Worldpop
• Dasymetric downscaling based on land use and other data to
disaggregate census data to a regular grid, 100 m
• HRSL
• High resolution (1 arc second, ~30 m) population density data derived
using Convolutional Neural Network techniques and 0.5 m resolution
satellite imagery capable of resolving individual buildings.
• Census data redistributed to buildings.
Population exposed to the 1 in 100 year
flood (millions).
Country WorldPop HRSL Change % LandScan™ HRSL Change %
Burkina Faso 2.31 1.74 -25 2.95 1.74 -41
Cambodia 6.00 4.69 -22 6.76 4.69 -31
Ghana 3.16 2.64 -16 3.65 2.64 -28
Haiti 3.15 3.11 -1 2.82 3.11 10
Madagascar 4.29 3.50 -18 4.45 3.50 -21
Malawi 2.57 1.61 -38 2.68 1.61 -40
Mexico 27.03 24.37 -10 28.69 24.37 -15
Mozambique 5.10 3.76 -26 5.80 3.76 -35
Philippines 44.00 42.65 -3 47.61 42.65 -10
Puerto Rico 0.79 0.68 -15 0.82 0.68 -18
Rwanda 0.79 0.59 -25 1.14 0.59 -48
South Africa 2.88 2.02 -30 4.66 2.02 -57
Sri Lanka 3.78 2.84 -25 4.67 2.84 -39
Tanzania 7.24 5.29 -27 7.59 5.29 -30
Uganda 4.17 1.66 -60 3.99 1.66 -58
Population exposure examples
Analysis snapshot in Cambodia: A) WorldPop data B) Facebook HRSL C)
1in100 year flood hazard intersecting with WorldPop, D) 1in100 year flood
hazard intersecting with Facebook HRSL
Population exposure examples
Example of population data and generated exposure maps for Lilongwe city, in
Malawi. A, B & C displays the HRSL, WorldPop and LandScan™ population
datasets respectively. D, E &F shows the exposure to the 100 year flood hazard
footprint for each demographic dataset respectively.
Cumulative distribution of exposure
Cumulative distribution of exposed population across all of the modelled ‘wet’ cells. Red indicates WorldPop exposure, Green the LandScan™ exposure and Blue the exposure calculated using the HRSL data.
Exposure to flooding across land use type
Proportion of area returning an exposure
value, across each land use type
Population exposed to the 1 in 100 year
flood (millions).
Country WorldPop HRSL Change % LandScan™ HRSL Change %
Burkina Faso 2.31 1.74 -25 2.95 1.74 -41
Cambodia 6.00 4.69 -22 6.76 4.69 -31
Ghana 3.16 2.64 -16 3.65 2.64 -28
Haiti 3.15 3.11 -1 2.82 3.11 10
Madagascar 4.29 3.50 -18 4.45 3.50 -21
Malawi 2.57 1.61 -38 2.68 1.61 -40
Mexico 27.03 24.37 -10 28.69 24.37 -15
Mozambique 5.10 3.76 -26 5.80 3.76 -35
Philippines 44.00 42.65 -3 47.61 42.65 -10
Puerto Rico 0.79 0.68 -15 0.82 0.68 -18
Rwanda 0.79 0.59 -25 1.14 0.59 -48
South Africa 2.88 2.02 -30 4.66 2.02 -57
Sri Lanka 3.78 2.84 -25 4.67 2.84 -39
Tanzania 7.24 5.29 -27 7.59 5.29 -30
Uganda 4.17 1.66 -60 3.99 1.66 -58
Population exposed to the 1 in 100 year
flood (millions).
Country WorldPop HRSL Change % LandScan™ HRSL Change %
Burkina Faso 2.31 1.74 -25 2.95 1.74 -41
Cambodia 6.00 4.69 -22 6.76 4.69 -31
Ghana 3.16 2.64 -16 3.65 2.64 -28
Haiti 3.15 3.11 -1 2.82 3.11 10
Madagascar 4.29 3.50 -18 4.45 3.50 -21
Malawi 2.57 1.61 -38 2.68 1.61 -40
Mexico 27.03 24.37 -10 28.69 24.37 -15
Mozambique 5.10 3.76 -26 5.80 3.76 -35
Philippines 44.00 42.65 -3 47.61 42.65 -10
Puerto Rico 0.79 0.68 -15 0.82 0.68 -18
Rwanda 0.79 0.59 -25 1.14 0.59 -48
South Africa 2.88 2.02 -30 4.66 2.02 -57
Sri Lanka 3.78 2.84 -25 4.67 2.84 -39
Tanzania 7.24 5.29 -27 7.59 5.29 -30
Uganda 4.17 1.66 -60 3.99 1.66 -58
Total exposed population range between 3.12M and
3.01M for the WorldPop and HRSL
A small change of -4%.
In WorldPop data this exposure is spread over an area of
around 40,000 km2, compared with an area of around
3700 km2 when using HRSL data.
Population exposed to the 1 in 100 year
flood (millions).
Country WorldPop HRSL Change % LandScan™ HRSL Change %
Burkina Faso 2.31 1.74 -25 2.95 1.74 -41
Cambodia 6.00 4.69 -22 6.76 4.69 -31
Ghana 3.16 2.64 -16 3.65 2.64 -28
Haiti 3.15 3.11 -1 2.82 3.11 10
Madagascar 4.29 3.50 -18 4.45 3.50 -21
Malawi 2.57 1.61 -38 2.68 1.61 -40
Mexico 27.03 24.37 -10 28.69 24.37 -15
Mozambique 5.10 3.76 -26 5.80 3.76 -35
Philippines 44.00 42.65 -3 47.61 42.65 -10
Puerto Rico 0.79 0.68 -15 0.82 0.68 -18
Rwanda 0.79 0.59 -25 1.14 0.59 -48
South Africa 2.88 2.02 -30 4.66 2.02 -57
Sri Lanka 3.78 2.84 -25 4.67 2.84 -39
Tanzania 7.24 5.29 -27 7.59 5.29 -30
Uganda 4.17 1.66 -60 3.99 1.66 -58
In Malawi, around 80% of modelled wet cells
overlay inhabited areas according to the WorldPop
and LandScan™ population data, compared with
only around 2% when the HRSL population data
are used.
Conclusions
• In this analysis we use flood hazard data from a ~90m resolution
hydrodynamic inundation model to demonstrate the impact of different
population distributions on flood exposure calculations for 18
developing countries spread across Africa, Asia and Latin America.
• In the new data, populations are represented as more risk-averse and
largely avoiding obvious flood zones.
• The results also show that existing demographic datasets struggle to
represent concentrations of exposure, with the total exposed
population being spread over larger areas.
• A substantial shift in exposure from rural to urban communities is
observed with the HRSL data
• The results suggest that many large-scale flood risk estimates may
require significant revision.