Block-level urban socioeconomic characterization using satellite imagery and points of interest: the case of Lima, Peru

Identifying which observable urban features are associated with socioeconomic status matters for policy design, yet remains understudied at fine spatial scales. We develop an Urban Deprivation Index (UDI) for Metropolitan Lima, Peru, using the 102,368 census block records from the Instituto Nacional de Estadística e Informática (INEI). Our approach integrates Sentinel-2 multispectral imagery, Sentinel-1 synthetic aperture radar (SAR), VIIRS nighttime lights, and 40,826 Points of Interest (POIs) from OpenStreetMap. Using a Random Forest with 20 parsimonious variables, we obtain a Pearson correlation of r = 0.786 ± 0.050 and a mean absolute error of 0.24 strata under spatial cross-validation (500 partitions, 50 districts). 88.8% of blocks fall within ±0.5 strata of the official classifications. SHapley Additive exPlanations (SHAP) analysis reveals that only two variables, Green Accessibility and Urban Remoteness, dominate 79% of individual predictions. Removing them reduces performance by only 2.05%: the remaining 18 variables carry redundant information. Socioeconomic segregation in Lima is structured by environmental amenities and urban centrality.