High-resolution Wealth Index Map, Senegal, 2020
Senegal, 2020
Reference ID
SEN_2020_HRWIM_GEO_v01_M
Metadata
Identification
Title
High-resolution Wealth Index Map, Senegal, 2020
Identifier
SEN_2020_HRWIM_GEO_v01_M
Edition
v.1
Edition date
2021-06-10
Status
pending
Language
ENG
FRA
characterSet
utf-8-sig
Date
| Date | Type |
|---|---|
| 2020-08-31 | creation |
| 2020-09-01 | released |
| 2021-09-10 | distribution |
Graphic overview
Download
Responsible party
Kamwoo Lee
(author)
World Bank, Development Data Group (DECDG), Data Analytics and Tools unit (DECAT)
klee16@worldbank.org
Presentation form
mapDigital
tableDigital
imageDigital
Series name
High-resolution Wealth Index Maps in Sub-Saharan Africa
Citation
Lee, K., & Braithwaite, J. (2020). High-Resolution Poverty Maps in Sub-Saharan Africa. arXiv preprint https://arxiv.org/abs/2009.00544.
Abstract
High-resolution Wealth Index Maps in Sub-Saharan Africa have been created using machine learning techniques and several publicly available geospatial data including satellite images, OpenStreetMap (OSM), UN OCHA settlements dataset (OCHA), High Resolution Population Density Maps (HRSL), WorldPop Gridded population dataset, NOAA Visible Infrared Imaging Radiometer Suite (VIIRS), and Database of Global Administrative Areas (GADM). This dataset contains information on buildings, roads, points of interest (POIs), night-time luminosity, population density, and estimated wealth index for 1-square mile inhabited places identified by the underlying datasets. The wealth level is an estimated value of the International Wealth Index which is a comparable asset based wealth index covering the complete developing world.
purpose
The purpose of this dataset is to provide village-level wealth estimates for places where an up-to-date poverty map is needed.
Point of contact
Kamwoo Lee
(author)
World Bank, Development Data Group (DECDG), Data Analytics and Tools unit (DECAT)
klee16@worldbank.org
Resource maintenance
| Update frequency |
|---|
| annually |
Descriptive keywords
| Type | Thesaurus | Keyword |
|---|---|---|
| theme | SEDAC Theme | dataset |
| theme | Data Granularity | 1-square-mile |
| theme | ISO Topic | Economy, Society |
| theme | World Bank Theme Taxonomy and Definitions | Inclusive Growth (131), Spatial Growth (134) |
| theme | JEL Classification System | Large Data Sets: Modeling and Analysis (C55), Measurement and Analysis of Poverty (I32), Technological Innovation (Q55) |
Spatial representation type
vector
Spatial extent
| Place | East | West | North | South |
|---|---|---|---|---|
| Senegal | -11.368064532 | -17.518811091 | 16.670517443 | 12.342567805 |
Constraints
Access constraints
| unrestricted |
Use constraints
| unrestricted |
Use limitations
| The information contained in this dataset is for general information purpose only. This dataset is not a result of a World Bank project. The World Bank does not represent, endorse, or guarantee its accuracy and accepts no liability for any loss or damage arising from inaccuracies or omissions. |
Distribution
Distribution format
| Name | Specification |
|---|---|
| Shapefile | Esri vector data storage format |
| text/csv | RFC4180 - Common Format and MIME Type for Comma-Separated Values (CSV) Files |
Distributor
World Bank, Development Data Group (DECDG), Data Analytics and Tools unit (DECAT)
Online resources
Data quality
Lineage statement
The High-resolution Wealth Index Maps in Sub-Saharan Africa have been created by utilizing publicly available geospatial data and machine learning methods. Method-wise, a new learning mechanism was devised to combine an XGBoost model and a convolutional neural network (CNN) model to estimate wealth index from both geospatial features and satellite images. Data-wise, training data was collected from 6 data sources:
- OpenStreetMap (OSM)
- Demographic and Health Survey (DHS)
- UN OCHA settlements dataset (OCHA)
- Visible Infrared Imaging Radiometer Suite Nighttime Lights (VNL)
- Daytime satellite images through Google Static Maps API (Google Maps)
- High-Resolution Settlement Layer (HRSL) and the WorldPop gridded population dataset (WorldPop).
- OpenStreetMap (OSM)
- Demographic and Health Survey (DHS)
- UN OCHA settlements dataset (OCHA)
- Visible Infrared Imaging Radiometer Suite Nighttime Lights (VNL)
- Daytime satellite images through Google Static Maps API (Google Maps)
- High-Resolution Settlement Layer (HRSL) and the WorldPop gridded population dataset (WorldPop).
Lineage process step
<b>1. Identifying inhabited places:</b> <br>
This step combines OCHA, OSM, HRSL and WorldPop to create a harmonized 1-sqmi-level settlement dataset.
It first gets information on inhabited places from the OCHA dataset if it's available. Then it adds OSM places where the place tag is 'city', 'town', 'village', 'hamlet', or 'isolated_dwelling'. If the distance from an OSM place to an OCHA place is less than 300 m (around 1000 feet), it assumes that they are the same place and adds supplemental info to the OCHA place. If the minimum distance from an OSM place to any OCHA places is more than 300 m (around 1000 feet), it adds the OSM place as a separate inhabited place.
In order not to overlook numerous unnamed places and many smaller communities within big cities, this step has an additional process. It divides each of the smallest admin regions into a grid of 10 km^2 (around 4-sq-mi) squares and finds a 1-sq-mi area that has the largest population within a grid. If the population is larger than 100 and there is no listed place in UN OCHA and OSM datasets within 2 km around the HRSL place, it is added as a separate inhabited place.
This step combines OCHA, OSM, HRSL and WorldPop to create a harmonized 1-sqmi-level settlement dataset.
It first gets information on inhabited places from the OCHA dataset if it's available. Then it adds OSM places where the place tag is 'city', 'town', 'village', 'hamlet', or 'isolated_dwelling'. If the distance from an OSM place to an OCHA place is less than 300 m (around 1000 feet), it assumes that they are the same place and adds supplemental info to the OCHA place. If the minimum distance from an OSM place to any OCHA places is more than 300 m (around 1000 feet), it adds the OSM place as a separate inhabited place.
In order not to overlook numerous unnamed places and many smaller communities within big cities, this step has an additional process. It divides each of the smallest admin regions into a grid of 10 km^2 (around 4-sq-mi) squares and finds a 1-sq-mi area that has the largest population within a grid. If the population is larger than 100 and there is no listed place in UN OCHA and OSM datasets within 2 km around the HRSL place, it is added as a separate inhabited place.
Date: 2020-08-31
Processor
Kamwoo Lee
(author)
Sources
| Description | Citation source | Organization |
|---|---|---|
| OCHA | SenegalSettlements |
United Nations Office for the Coordination of Humanitarian Affairs
https://data.humdata.org/
|
| OSM | GEOFABRIK dump |
Geofabrik GmbH
https://www.geofabrik.de/data/download.html
|
| HRSL | High Resolution Population Density Maps |
Facebook Connectivity Lab and Center for International Earth Science Information Network – CIESIN – Columbia University
https://data.humdata.org/dataset/highresolutionpopulationdensitymaps
|
| WorldPop | WorldPop gridded population estimate |
WorldPop
https://www.worldpop.org/methods/populations
|
<b>2. Extracting buildings, roads, and points of interest (POIs) information for each inhabited place:</b> <br>
This step extracts information on man-made structures and services from OSM.
Building info: Total building area, and the number of buildings within 1 km radius surrounding the GPS coordinates of each inhabited place.
Road info: Total length of roads and number of junctions within 1 km radius surrounding the GPS coordinates of each inhabited place. Distance to the closest road and junctions from the GPS coordinates of each inhabited place.
POI info: The number of and distance to 110 types of service locations. POIs: bar, bbq, biergarten, cafe, drinking water, fast food, food court, ice cream, pub, restaurant, college, driving school, kindergarten, language school, library, toy library, music school, school, university, bicycle parking, bicycle repair station, bicycle rental, boat rental, boat sharing, bus station, car rental, car sharing, car wash, vehicle inspection, charging station, ferry terminal, fuel, grit bin, motorcycle parking, parking, parking entrance, parking space, taxi, atm, bank, bureau de change, baby hatch, clinic, dentist, doctors, hospital, nursing home, pharmacy, social facility, veterinary, arts centre, brothel, casino, cinema, community centre, fountain, gambling, nightclub, planetarium, public bookcase, social centre, stripclub, studio, swingerclub, theatre, animal boarding, animal shelter, baking oven, bench, childcare, clock, conference centre, courthouse, crematorium, dive centre, embassy, fire station, give box, grave yard, gym, hunting stand, internet cafe, kitchen, kneipp water cure, marketplace, monastery, photo booth, place of worship, police, post box, post depot, post office, prison, public bath, ranger station, recycling, refugee site, sanitary dump station, sauna, shelter, shower, telephone, toilets, townhall, vending machine, waste basket, waste disposal, waste transfer station, watering place, water point.
This step extracts information on man-made structures and services from OSM.
Building info: Total building area, and the number of buildings within 1 km radius surrounding the GPS coordinates of each inhabited place.
Road info: Total length of roads and number of junctions within 1 km radius surrounding the GPS coordinates of each inhabited place. Distance to the closest road and junctions from the GPS coordinates of each inhabited place.
POI info: The number of and distance to 110 types of service locations. POIs: bar, bbq, biergarten, cafe, drinking water, fast food, food court, ice cream, pub, restaurant, college, driving school, kindergarten, language school, library, toy library, music school, school, university, bicycle parking, bicycle repair station, bicycle rental, boat rental, boat sharing, bus station, car rental, car sharing, car wash, vehicle inspection, charging station, ferry terminal, fuel, grit bin, motorcycle parking, parking, parking entrance, parking space, taxi, atm, bank, bureau de change, baby hatch, clinic, dentist, doctors, hospital, nursing home, pharmacy, social facility, veterinary, arts centre, brothel, casino, cinema, community centre, fountain, gambling, nightclub, planetarium, public bookcase, social centre, stripclub, studio, swingerclub, theatre, animal boarding, animal shelter, baking oven, bench, childcare, clock, conference centre, courthouse, crematorium, dive centre, embassy, fire station, give box, grave yard, gym, hunting stand, internet cafe, kitchen, kneipp water cure, marketplace, monastery, photo booth, place of worship, police, post box, post depot, post office, prison, public bath, ranger station, recycling, refugee site, sanitary dump station, sauna, shelter, shower, telephone, toilets, townhall, vending machine, waste basket, waste disposal, waste transfer station, watering place, water point.
Date: 2020-08-31
Processor
Kamwoo Lee
(author)
Sources
| Description | Citation source | Organization |
|---|---|---|
| OSM | GEOFABRIK dump |
Geofabrik GmbH
https://www.geofabrik.de/data/download.html
|
<b>3. Extracting population density information for each inhabited place:</b> <br>
This step extracts estimated population within 1-square mile, 25km2, 100km2 areas around each inhabited place.
This step extracts estimated population within 1-square mile, 25km2, 100km2 areas around each inhabited place.
Date: 2020-08-31
Processor
Kamwoo Lee
(author)
Sources
| Description | Citation source | Organization |
|---|---|---|
| HRSL | High Resolution Population Density Maps |
Facebook Connectivity Lab and Center for International Earth Science Information Network – CIESIN – Columbia University
https://data.humdata.org/dataset/highresolutionpopulationdensitymaps
|
| WorldPop | WorldPop gridded population estimate |
WorldPop
https://www.worldpop.org/methods/populations
|
<b>4. Extracting nighttime luminosity information for each inhabited place:</b> <br>
This step extracts nighttime luminosity within 1-square mile, 25km2, 100km2 areas around each inhabited place.
This step extracts nighttime luminosity within 1-square mile, 25km2, 100km2 areas around each inhabited place.
Date: 2020-08-31
Processor
Kamwoo Lee
(author)
Sources
| Description | Citation source | Organization |
|---|---|---|
| VNL | Annual VIIRS Cloud Mask - Outlier Removed - Nighttime Lights |
National Oceanic and Atmospheric Administration (NOAA)
https://www.ngdc.noaa.gov/eog/viirs/download_dnb_composites.html
|
<b>5. Training XGBoost model and CNN model:</b> <br>
This step trains/estimates/validates wealth index for each inhabited place.
First, an XGBoost model is trained on the buildings, roads, POIs, night-time luminosity, and population density features. It estimates wealth levels for all inhabited places.
Second, a customized CNN model is trained on the estimated wealth index from the XGBoost model and day-time satellite images downloaded from Google Static Map. It estimates the 4 probabilities that each populated place is a rich, upper-middle, lower-middle, or poor area.
Third, the featurized information (4 probabilities) from the CNN model is added to the XGBoost model as an additional inputs resulting in better estimates.
Fourth, the CNN model is retrained on the better estimates from the XGBoost model.
The whole process creates a cycle where the two models improve each other by providing better training data with every iteration.
The wealth index dataset is the result of the 7th iteration.
This step trains/estimates/validates wealth index for each inhabited place.
First, an XGBoost model is trained on the buildings, roads, POIs, night-time luminosity, and population density features. It estimates wealth levels for all inhabited places.
Second, a customized CNN model is trained on the estimated wealth index from the XGBoost model and day-time satellite images downloaded from Google Static Map. It estimates the 4 probabilities that each populated place is a rich, upper-middle, lower-middle, or poor area.
Third, the featurized information (4 probabilities) from the CNN model is added to the XGBoost model as an additional inputs resulting in better estimates.
Fourth, the CNN model is retrained on the better estimates from the XGBoost model.
The whole process creates a cycle where the two models improve each other by providing better training data with every iteration.
The wealth index dataset is the result of the 7th iteration.
Date: 2020-08-31
Processor
Kamwoo Lee
(author)
Sources
| Description | Citation source | Organization |
|---|---|---|
| DHS | Demographic and Health Surveys |
United States Agency for International Development (USAID)
https://dhsprogram.com/
|
| Google Maps | Daytime satellite images through Google Static Maps API |
Google
https://maps.googleapis.com
|