Multi-scale estimation of poverty rate using night-time light imagery

Explicit poverty data are critical for policymaking and targeting humanitarian aid. Poverty rate is the most widely accepted definition of a poverty status. However, poverty rate data are commonly available at countrylevel. Here, we proposed an approach that estimates the poverty rate at different spatial scales with a consistent standard within a country. We first trained the model based on household survey data and publicly available remote sensing data to derive a wealth index map, and then we developed the relationship between the wealth index and the poverty rate at country-level. The relationship finally was applied to estimate the multi-spatial scale poverty rates in the study area. We made validation between the estimated and statistical province-level poverty rates using relative error and R-Square. A case study was carried out in Mozambique. The results showed that the proposed model has a good ability to estimate poverty rate with an overall accuracy of 85.21%, as well as an R-Square of 0.94. There was a huge gap within Mozambique, with the Northern Provinces holding high poverty rates and the Southern Provinces holding low poverty rates. The district-level poverty rate map might reflect the negative impact of climate disasters, as well as the positive influence of economic and trade exchange. Given that the data we use are publicly available, the proposed methodology can be applied to other countries to estimate poverty rates at various spatial scales.