Poverty-mapping-in-Sri-Lanka

Mapping the spatial distribution of poverty using call detail records and remote sensing data

Project Overview

This project aims to map the spatial distribution of poverty using a combination of remote sensing data and call detail records. The analysis includes Principal Component Analysis (PCA) to calculate a ground truth variable, "pc_1", which serves as a proxy for poverty. Three spatial models were tested: Spatial Error Model, Spatial Lag Error Model, and Spatial Durbin Error Model. After careful evaluation, the Spatial Durbin Error Model was selected as the best-performing model.

Data Sources

Call Detail Records

"call_count" "avg_call_duration" "nighttime_call_count" "avg_nighttime_call_duration" "incoming_call_count" "avg_incoming_call_duration" "radius_of_gyration" "unique_tower_count" "spatial_entropy" "avg_call_count_per_contact" "avg_call_duration_per_contact" "contact_count" "social_entropy"

Remote Sensing Data

"travel_time_major_cities" "population_count_worldpop" "population_count_ciesin" "population_density" "aridity_index" "evapotranspiration" "nighttime_lights" "elevation" "vegetation" "distance_roadways_motorway" "distance_roadways_trunk" "distance_roadways_primary" "distance_roadways_secondary" "distance_roadways_tertiary" "distance_waterways" "urban_rural_fb" "urban_rural_ciesin" "global_human_settlement" "protected_areas" "land_cover_woodland" "land_cover_grassland" "land_cover_cropland" "land_cover_wetland" "land_cover_bareland" "land_cover_urban" "land_cover_water" "pregnancies" "births" "precipitation" "temperature"

Analysis Workflow

1. Data Preprocessing: Combine call detail records and remote sensing data. Remove Grama Niladhari divisions(Gnds) with zero population. Log transformation of skewed variables. Calculate "pc_1" using PCA.

2. Model Selection: Test three spatial models - Spatial Error Model, Spatial Lag Error Model, and Spatial Durbin Error Model.

3. Model Evaluation: Compare model performance using appropriate evaluation metrics, considering the spatial autocorrelation in the data.

4. Model Selection: The Spatial Durbin Error Model was chosen as the best-performing model based on the Akike information criterion and Bayesian information criterion and its ability to capture spatial relationships in the data and provide accurate poverty estimates.

5. Mapping Poverty: Utilize the selected model to map the spatial distribution of poverty across Sri Lanka.

For any questions or assistance, please contact Chanuka at chanuka@lirneasia.net or Merl at merl@lirneasia.net

Acknowledgments

We acknowledge the sources of the data used in this analysis and express gratitude to all contributors who made this research possible.