Presented by Vincent Seaman, Interim Deputy Director for the Strategy, Data and Analytics, Bill & Melinda Gates Foundation, at the 2017 GIS Working Group Annual Meeting.

1. Geospatial Reference
Information Database
(GRID)
How the Polio Eradication Effort in Nigeria
led to a Quest for Global Geospatial
Reference Data
Vince Seaman
Deputy Director (Interim)
Data & Analytics, Global Development
Bill & Melinda Gates Foundation
1

2. © 2013 Bill & Melinda Gates Foundation |
*** Confidential – for internal use only *** 2
WHY NIGERIA?
AFTER RECORDING ONLY 21 POLIO CASES IN 2010, NIGERIA EXPERIENCED LARGE
INCREASES IN 2011 & 2012, MAKING IT A TOP PRIORITY FOR THE GPEI PARTNERS
2
• Many cases from settlements
not visited by vaccination teams
• Microplans were incomplete
• Target populations were
inaccurate and grossly inflated
in many areas, leading to data
falsification and vaccine waste
• Quality of monitoring and
coverage data was poor
Reasons for Outbreak….
798
388
21
62
122
53
6 0
0
100
200
300
400
500
600
700
800
900
2008 2009 2010 2011 2012 2013 2014 2015
Polio Cases in Nigeria, 2009-15

3. © 2013 Bill & Melinda Gates Foundation |
*** Confidential – for internal use only ***
3
Ward Maps from
Nigeria
- Incomplete
- Inaccurate
- Out of Date
Many missed
settlements
were not on
existing
microplans

4. © 2013 Bill & Melinda Gates Foundation |
*** Confidential – for internal use only *** 4
Existing Public Databases Limited to Urban Centers
Automated Feature Extraction (FE) Settlements (ORNL)Adamawa State, Nigeria (OpenStreet Maps)

5. © 2013 Bill & Melinda Gates Foundation |
*** Confidential – for internal use only *** 5
Manual & Automated Feature Extraction of Satellite Imagery Field Data Collection Settlement Attributes used to
create Ward Boundaries
Points of Interest
2013-16: GIS Base Layers Collected for 11 Northern States

6. Creating Ward
Vaccination
Boundaries
Settlement metadata (ward
attribute) used with ESRI
Thiessen Polygon tool to
create Ward “operational”
boundary.
• Wards aggregated to form
LGA Boundaries
• LGAs aggregated to form
State Boundaries
• State Boundaries “fit” to
existing National Boundary
- No existing formal
Ward boundary maps
- Polio vaccination
campaigns occur at
Ward level

7. Ward/LGA Boundaries
later validated by polio
H2H team tracks

8. Gangara Ward
Jibia LGA, Katsina State
VTS Map, Feb. 2016
Hand-drawn Ward map
2011 Initial GIS trace
OSM Map – Feb. 2016
OSM – Gangara A
Settlement Mapping – The Reality
VTS data
uploaded
to OSM by
eHealth,
2015
> 10 settlements
missing from
local map
Accuracy of
settlement geo-
location is poor

9. © 2013 Bill & Melinda Gates Foundation |
*** Confidential – for internal use only ***
GIS TRACKING OF VACCINATION TEAMS
9
Given to Ward Focal Person (WFP)
at LGA HQ each morning
WFP Returns to Ward take-off point
and gives phones to vaccinators
WFP returns to LGA-HQ where
GPS tracks are downloaded
Vaccinators return phone to
WFP at the end of their day
Feedback for daily coverage provided to
WFPs and LGA team at daily meeting
5a-6:30a 7a-8a
11a-5p
2p–8p
Tracks uploaded to
EOCs/Dashboard via MiFi
Missed Settlement Report
generated at end of days 4 & 5
GPS – enabled
Android phone
Collects time-stamped
GIS coordinates every 2
minutes

10. © 2013 Bill & Melinda Gates Foundation |
*** Confidential – for internal use only ***
BUA Polygon divided into 50 meter grid squares
Geographic Coverage - BUAs
50 meter buffer around each hamlet
Geographic Coverage –
Hamlet Areas
75 meter Buffer around SS Point Feature
Geographic Coverage –
Small Settlements (SS)
GIS Tracking:
Calculation of
Geographic Coverage

11. © 2013 Bill & Melinda Gates Foundation |
*** Confidential – for internal use only *** 11
Post Campaign
Reports:
LGA/Ward Overview
• GeoCoverage
• Settlements Visited & % of total
• Target Population visited &
% of total
• Heat Map showing
visited/missed settlements
http://vts.eocng.org/

12. Improved Coverage of Border Settlements, Tudun Wada LGA, Kano State Jan 2015 tracks
Sept 2012 tracks
LGA Border

13. 1
3
DECLINING NUMBER OF CHRONICALLY-MISSED*
SETTLEMENTS IN KANO LGAS
17 of the 70 chronically missed
settlements are nomadic
* Chronically Missed = not visited the last 3 campaigns; same LGAs tracked since November 2013; all
LGAs to be tracked from April 2014
70
155
233
4.1% of all
settlements
tracked
2.7% of all
settlements
tracked
≈ 3000
Children
Missed
4
≈ 1860
Children
Missed
≈ 580
Children
Missed
1.2% of all
settlements
tracked
< 0.05% of all
settlements
tracked

14. 1
414
Aggregated tracks show road network

15. Settlement Total U5 Houses Total U5
UNG MAIGADI 5055 1011 45 369 75
State Master List GIS Est
Kaduna – Birnin Gwari LGA
Local Administrative
Population Data
Unreliable
• Census data unavailable below
Admin 1 level
• Inflated population counts
result in vaccine, bed net, and
other critical supply shortages
• 2014 H2H enumeration in Kano
state found polio target
population inflated by a factor
of 2, more than 3 million
children.
• Validation tool needed

16. December 14, 2017
Demographics &
Mobility mapping
Andy Tatem, University of Southampton
Total Population: 170,123,740
(July 2012 estimate)
Administrative units - 774
High Resolution
Population Distribution
In Northern Nigeria
BudhendraBhaduri
EddieBright,AnilCheriyadat,AmyRose,JakeMcKee,JeanetteWeaver,
MaryUrban,RajuVatsavai

17. 486 features
2 BUAs, 3 SSAs, 9 HA (56 hamlets)
14 settlement features
Aggregated Settlement Layer Serves as the Basis for Mapping
Raw FE layer Aggregated Settlement Layer
Imagery Courtesy of Digital Globe)
ORNL Semi-Automated Feature Extraction Captures 95+% Structures

18. • Power spectrum contours represent 20, 40,60 and 80% energy levels.
• Shape of the power spectrum characterizes the semantic category.
• Dominant orientations of Downtown, Suburban, Commercial Complex
structures captured in power spectrum.
Feature Extraction:
Different Objects Have Unique Spectral “Signatures”

19. Structure Edges Give Different LINE PATTERNS
Local line patterns a good descriptor of the spatial arrangements. Line statistics can
representative of structural dimensions

20. All Urban Areas Have 4-6
Neighborhood Types
- Based on size, shape, and
orientation of structures
- Neighborhood type is related to
building use: residential,
commercial, mixed-use, etc.
Local geospatial
neighborhoods are
represented using rich
feature descriptors
composed of edge,
texture, lines and
spectral attributes

21. Managed by UT-Battelle
for the Department of Energy
Neighborhood Classification Scheme – N. Nigeria
 Neighborhood Type Layer for Nigeria (based on Kano metro area)
– established 7 residential settlement types (6 Urban, 1 rural) + non-residential
 Population density of each neighborhood type determined
from microcensus data (>100 clusters for each type)
M: rural
Z: non-residential
Slums
Slums

22. December 14, 2017
© 2013 Bill & Melinda Gates
Foundation |
22
Rural residential land use and population density:
- Rural areas have less diversity and can be characterized by a single type (M)
M M M

23. GIS Population Model Microcensus Methods
Northern 10 States = 900 clusters in 9 states
Middle/South Total = 1600 clusters in 8 statesMicrocensus Methods:
• All buildings assigned to one of 8 neighborhood
types (Z= non-residential)
• 200 polygons selected randomly for each state
representing all neighborhood types in that state
equally (approx. 10,000 HH/state)
• Each polygon contains approx. 50 residential
structures
• Microcensus team obtains total population and the
U5 count for each HH in polygon
• Decision for which microcensus data is used for
which state based on proximity and demographics
Densities* (Kebbi, Zamfara)
M = 147
Densities* (Kaduna, Kano)
A B C D E F M
810 350 257 141 438 61 246
Densities* (Bauchi, Yobe)
M = 161
*Population density (per hectare) for each neighborhood type

24. Microcensus Data Collected in All Neighborhood Types
- 150 polygons (492 structures) in the 6 identified “neighborhood” areas
- Building features, use, occupancy, and photos collected
December 14, 2017 © 2013 Bill & Melinda Gates Foundation | 24

25. At least 5 Validation Data Sets per State
- entire population counted inside of defined area (min. 50 households)
- not used to inform model
December 14, 2017 © 2013 Bill & Melinda Gates Foundation | 25

26. Validation Sets
Average Variance = 8.7% (Range = 1 – 18%)
December 14, 2017 © 2013 Bill & Melinda Gates Foundation | 26
KN0902_5 Microcensus = 334 Model = 366
Variance = +9.3% Density = 5.3
Microcensus = 300 Model = 304
Variance = + 1% Density = 6.25
KN0902_6

27. OUTPUT: 90-meter population grid with total counts, or selected demographic

28. December 14, 2017 © 2013 Bill & Melinda Gates Foundation | 28
GIS Model and 2006 Census Projections Nearly Identical at
National Level – Wide Variations at Sub-National Levels
Current Model
POP2006 Pop2015 PopUnder5yrs State estimate diff diff (1,000s) diff/projected
4,676,465 6,318,333 1,263,667 Bauchi 5,842,730 -475,603 -475.60 -0.08
4,151,193 5,608,643 1,121,729 Borno 5,173,450 -435,193 -435.19 -0.08
4,348,649 5,624,614 1,124,923 Jigawa 5,288,378 -336,236 -336.24 -0.06
6,066,562 7,915,487 1,583,097 Kaduna 8,521,381 605,893 605.89 0.08
9,383,682 12,568,289 2,513,658 Kano 13,718,523 1,150,234 1,150.23 0.09
5,792,578 7,558,000 1,511,600 Katsina 8,039,212 481,212 481.21 0.06
3,238,628 4,262,742 852,548 Kebbi 4,215,941 -46,801 -46.80 -0.01
3,696,999 4,823,745 964,749 Sokoto 5,537,133 713,388 713.39 0.15
2,321,591 3,164,090 632,818 Yobe 3,566,837 402,747 402.75 0.13
3,259,846 4,328,270 865,654 Zamfara 3,762,484 -565,786 -565.79 -0.13
2006 Census Projections Model vs. Census Projections
Std. Dev.:
Northern States = 9%
All States = 33%
Current Model Ratio
POP2006 Pop2015 PopUnder5yrs estimate Model/Census
139,983,289 185,847,096 37,169,419 188,451,476 1.01
2006Census Projections
National Estimates

29. © 2013 Bill & Melinda Gates Foundation |
*** Confidential – for internal use only *** 29
Lack of Accurate Geospatial Reference Data Affects
Population Estimates and Spatial Analyses
I. Yearly growth projections applied to census data do not accurately reflect
urban/rural growth differences below the State Level
II. Demographic groups/target populations are not a fixed percentage of total
population across country
III. Sub-National administrative boundary layers are imprecise
IV. Standard Cluster survey methods may not result in a representative
sample

30. December 14, 2017
© 2013 Bill & Melinda Gates
Foundation |
30
Detail from
Ungogo, 2006
Detail from
Ungogo, 2015
I. Yearly growth projections are not reliable below the State level
due to uneven growth rates in rural & urban areas.

31. December 14, 2017 © 2013 Bill & Melinda Gates Foundation | 31
Urban LGAs grow at different rates based on space, location, etc.
Census projections result from a growth rate (2.7-3.4%/year) applied at the state level,
but highly urban LGAs grow much faster, rural LGAs much slower.
SOLUTION: GIS estimates, based on 2015-16 imagery building footprint, reflect actual urban and
rural population distribution down to the settlement level
2006 and 2014 Kano
Settlement Extents
Percent Change in Settled
Area by LGA, 2006-2014
Nigeria

32. Managed by UT-Battelle
for the Department of Energy
Modeled Population Change – Kano Metro Area

33. Managed by UT-Battelle
for the Department of Energy
Calculated Rates of Annual Population Change for Both Methods (2006-2014)
0
1
2
3
4
5
6
7
8
9
Prorating
Modeling
(Census Projections)
(GIS Estimates)

34. % Children under 5 Varies from North to South, East to West
Alegana, et al. 2015 http://rsif.royalsocietypublishing.org/
II. MODELED DEMOGRAPHICS BASED ON NATIONAL
HH SURVEYS INDICATE FIXED FRACTIONS ARE
INAPPROPRIATE
State %U1 %U5 %U15
Abia 2.5% 13.2% 37.9%
Adamawa 2.9% 18.5% 52.2%
Akwa Ibom 3.2% 13.1% 36.1%
Anambra 2.3% 14.0% 39.7%
Bauchi 3.6% 20.7% 54.2%
Bayelsa 3.5% 14.7% 38.9%
Benue 2.5% 15.3% 44.8%
Borno 3.2% 21.6% 56.5%
Cross River 2.9% 13.3% 37.2%
Delta 2.8% 14.2% 38.8%
Ebonyi 2.4% 14.4% 42.4%
Edo 2.1% 13.0% 36.6%
Ekiti 2.0% 12.2% 35.4%
Enugu 2.2% 14.3% 41.2%
Fct, Abuja 2.2% 15.0% 41.2%
Gombe 3.1% 19.8% 53.1%
Imo 2.5% 13.9% 39.5%
Jigawa 3.6% 22.5% 58.2%
Kaduna 2.9% 18.3% 48.9%
Kano 3.1% 21.1% 54.3%
Katsina 3.3% 21.1% 54.8%
Kebbi 3.4% 19.6% 52.2%
Kogi 2.0% 14.3% 41.9%
Kwara 2.2% 13.3% 37.9%
Lagos 2.0% 13.0% 35.2%
Nasarawa 2.5% 15.7% 44.4%
Niger 3.1% 17.2% 47.1%
Ogun 1.9% 13.3% 38.4%
Ondo 2.1% 13.1% 38.2%
Osun 1.7% 12.6% 37.2%
Oyo 1.8% 12.5% 36.3%
Plateau 2.4% 16.5% 46.0%
Rivers 3.0% 13.4% 36.0%
Sokoto 3.6% 20.7% 52.8%
Taraba 2.7% 16.2% 47.0%
Yobe 3.7% 22.2% 57.2%
Zamfara 3.2% 20.7% 55.4%
National Average 2.7% 16.9% 46.0%
Nigeria Official % 4.0% 20.0% 47.6%
GIS Modeled % U1, U5 and U15
For the various demographic groups targeted by the GoN, a flat % is used
across the entire country: U1 = 4%, U5 = 20%, U15 = 47.6%

35. III. Sub-National Administrative
Boundaries are Imprecise
VTS*
Boundary
Published
Census
Boundary
Kano Metro LGAs
Nigeria
*VTS Boundary from polio GIS settlement mapping

36. December 14, 2017 © 2013 Bill & Melinda Gates Foundation | 36
LGA (Admin 2) Boundaries Referenced in Census Based on Area
No authoritative shapefiles available, however land area matches UN Admin 2 boundaries

37. December 14, 2017 © 2013 Bill & Melinda Gates Foundation | 37
Census Enumeration Areas do not
align with published boundaries
VTS* Boundary Census/UN Boundary (white)
*VTS Boundary from polio GIS settlement mapping
Geo-referenced census enumeration area map

38. • Polio “Operational”
Boundaries (VTS*)
GADM** and UN-WHO
(Census) all Differ
Gwale LGA, Kano State
Jan 2015
VTS
Boundary
GADM
Boundary
UN/Census
Boundary
**GADM = internationally-recognized global boundary
resource developed by Robert Hijmans & colleagues at
the University of California, Berkeley and the University
of California, Davis (Alex Mandel)
http://www.gadm.org/
Commonly used
“Official” boundaries
do not align with field
data & settlement
attributes
*VTS Boundary from polio GIS settlement mapping

39. GIS Population Estimates: VTS1, GADM2, UN-WHOBoundaries
2GADM Version 2.8, March 2016. http://www.gadm.org/
VTS Boundaries
Pop. Est. = 678,198
GADM Boundaries
Pop. Est. = 372,703
UN-WHO (Census) Boundaries
Pop. Est. = 484,934
Gwale LGA, Kano State, Nigeria
Use of incorrect
boundaries impacts
population estimates
1VTS Boundary from polio GIS settlement mapping

40. Z = Non-Residential
Neighborhood Types - Kano Metro Area
National HH Survey 2016
Cluster locations – Kano Metro LGAs
2016 Cluster Survey –
HH Points
> 90% of Household cluster
points from Types B & E,
none from Types A & FIV. Cluster Surveys - Are
They Truly Representative?

41. © 2013 Bill & Melinda Gates Foundation |
*** Confidential – for internal use only ***
http://geopode.world/

42. December 14, 2017
© 2013 Bill & Melinda Gates
Foundation |
42
Select the Layers tab to
see the drop-down
Map layers and Total
Population or < 5
population can be
selected here
Type in coordinates to go
to a specific place
Custom Demographics
slider: 0-12 mos, 5 year
intervals
Polygon and point
buffer options
Print Screen
Change
Basemap
Scale Bar
GIS POPULATION MODEL – USER INTERFACE OPTIONS
http://geopode.world/

43. SIMPLE CATCHMENT AREA - SELECT USER-DEFINED BUFFER AROUND A POINT
43
Retrieving settlement names and estimated
population/target population using a 2km buffer
around a Health Facility
Other Potential Output Columns:
• H2R/Outreach Settlement? Y/N
• Target Pop: <12mos, <15 years
• Vaccine/Supply requirements

44. December 14, 2017 © 2013 Bill & Melinda Gates Foundation | 44
GIS SUPPORT TO DEVELOPING COUNTRIES
Geospatial
Reference
Information
Database
Settlements and Points of Interest
Administrative Boundaries
Population / Demographics
Transportation Network
plus Capacity-Building (NSO, National Mapping Agency, etc.)
=

45. ecember 14, 2017Melinda Gates Foundation |
DfID-BMGF Partnership to co-fund GRID and
other Key Geospatial and Data-related Projects
DFID
Priority BMGF-DfID GRID Geographies - 2017
• Collect basic geospatial reference data (access geo-referenced national census data where available)
• Build capacity within Census/Population Commission, Bureau of Statistics (UNFPA, Flominder)
• Develop Population/Demographics & Population dynamics modeling
• Build data management/use capacity across all sectors
PROJECT 1 (census-based)
Support National Statistics Office/Population Council to
conduct georeferenced census & manage data
Year 1 Countries: Ethiopia, Tanzania, Zambia
PROJECT 2 (no census)
Support National Statistics Office/Population
Council to collect/model geospatial reference data
Year 1 Countries: Nigeria, DRC

46. December 14, 2017 © 2013 Bill & Melinda Gates Foundation | 46
GRID PROJECT DELIVERABLES
1. Geo-referenced layer of all settlements and key POIs
(from feature extraction layer)
2. Validated sub-national boundary layers (from settlement attributes)
3. Population & demographic estimates at 90 meters
(from neighborhood classification and microcensus data)
4. Capacity-building for NSO, NGA, and other government agencies
5. Country and Global Data Platforms

47. Intensive Capacity-Building (minimum 24 months)
December 14, 2017 © 2013 Bill & Melinda Gates Foundation | 47
NATIONAL STATISTICS OFFICE/POPULATION COMMISSION
• Training, software & hardware provision, technical support
• Manage, use and curate census data and other national statistics
NATIONAL GEOSPATIAL AGENCY
• Training, software & hardware provision, technical support
• Manage, use and curate national geodatabase
• Regular updates of boundaries, settlements, & POIs
OTHER GOVERNMENT MINISTRIES/AGENCIES (FINANCE, ELECTORAL, EDUCATION, UTILITIES, ETC.)
• Identify priority use-cases & applications
• Assist NSO and NGA in supporting other agencies
REGIONAL WORKSHOPS & TRAINING
• Additional opportunities to enhance GIS skills
• Network and share best practices with other AFRO country teams

48. Other BMGF-Supported
GIS Projects

49. Share the
VISION!
Contact Info:
Vince Seaman
Senior Program Officer
Country Support, Polio
Global Development
V +1.206.770.2351
C +1.206.669-7259
E Vincent.Seaman@gatesfoundation.org

50. © 2013 Bill & Melinda Gates Foundation |
*** Confidential – for internal use only *** 50
EXTRA SLIDES
GRID Data: Applications and Use Cases
1. Health Facility/service area catchment mapping
2. Cluster maps for fixed post vaccination campaigns
3. DEM maps show water flow/accumulation and catchment areas/populations
from sampling point – WASH, polio, malaria, Cholera, typhoid?
4. Imagery change analysis to determine settlement status in Borno state

51. December 14, 2017 © 2013 Bill & Melinda Gates Foundation | 51
1. Define HF Catchment Areas and Target Populations
1. Obtain Ward map from VTS – verify all HFs are accurately represented
2. Obtain Ward Settlement populations from VTS (Total and selected
demographics). Adjust populations if recent enumeration is available.
3. Add local names to settlements, or new settlements where needed.
4. Assign settlements to HF based on location, access, and services
5. Add populations (total and demographics) of selected settlements to
determine estimated catchment area population.
Note: A “Microplan” can be printed from the VTS that includes the Ward map, a list of primary and sub-place names, and

52. December 14, 2017 © 2013 Bill & Melinda Gates Foundation | 52
HF Catchment
area must be
determined
by local HF
and LGA staff

53. December 14, 2017 © 2013 Bill & Melinda Gates Foundation | 53
2. Cluster maps for fixed post vaccination campaigns.
1. Fixed post (FP) locations identified by 1 km settlement clusters
2. Target population in each cluster determines # of FPs and days
3. Ward target population can be used to estimate vaccine/personnel
requirements
4. Daily vaccination totals at each FP collected with smartphone to track
progress

54. 5
4
2. Cluster maps for fixed post vaccination campaigns.
Settlements
within 1km
clustered
# of Health
Camp days
calculated
Problem:
IPV Health Camps (HCs)
had to be located no
further than 1km from
any resident.
Solution:
An automated tool was
created that clustered
settlements within 1km
of one another.
Target populations were
then used to determine
the number of days the
HC would work in a
cluster.
Result:
>95% coverage overall,
no missed settlements

55. 55
Measles Campaign
Northern States, Oct. 2015
Microplanning Map
(5 day campaign)
- Rural Fixed Post = 125/day
- Urban Fixed Post = 175/day
- Settlements grouped in 1km
clusters
- Target Populations (< 1 year)
used to calculate # of fixed post
days

56. © 2013 Bill & Melinda Gates Foundation |
*** Confidential – for internal use only ***
Digital Elevation
Map (DEM) layers
can detect
changes in
elevation based
on the resolution.
For 30 meter
resolution, the
contour lines are
spaced 30 meters
apart.
3. DEM maps to Assess Environmental Surveillance Sites
- catchment area size, location and population estimates

57. Jakarta Police StnKuma Masallachi-Fagge Gogau Fagge
Kano Environmental Surveillance Sites
DEM Layers Used to Assess Polio Environmental Surveillance Sites

58. Junction Point
Drainage Line
Watershed-Catchment
Est. Population
Collected at Junction
Tablet-based
maps used in
the field to
locate optimum
locations for
ES sites

59. 59
Assessing Environmental
Surveillance Catchment
Areas
http://maps.novel-t.ch/#/catalog/all
17 Countries
300 sites

60. 2013 2016
4. Imagery change analysis to
determine settlement status in
Borno state.
CDC/GRASP Change Analysis of 2013 vs 2016 Imagery
Identified Damaged/Destroyed Settlements by Boko
Harum.

61. December 14, 2017 © 2013 Bill & Melinda Gates Foundation | 61
GIS Settlement maps and population used as base-
layer for analysis - output used to direct vaccination
teams and also by humanitarian response partners
(UN-OCHA, WFP)

62. 5. Population U1 living > 1km from a Health Facility
Requested by NPHCDA ED for Public Health Strengthening Assessment