Exécution de notre outil de synthèse sur un nouvel article

Vérifiez l’article ci-dessous, puis lancez sa synthèse.


Developing a Subdistrict-Level Economic Progress Index Using Geospatial Data

Meenu Mohan, Siddhartha Chattopadhyay

52 pages (46 pages main text, 6 pages appendices), 5 figures, 13 tables (6 main text, 7 appendix). An earlier version was presented at the XKDR Alternate Data Workshop, Mumbai. JEL: R11, R12, O18, E01, C43, C38
Licence : CC BY 4.0

Résumé : This paper develops an Economic Progress Index (EPI), a composite measure built from Earth Observation (EO) and open geospatial data with unsupervised Principal Component Analysis (PCA) weighting. The index captures the observable productive and enabling capacity of local economies and the intensity with which economic space is used. Sub-national monitoring in developing economies is held back by sparse data: district GDP estimates arrive with multi-year lags, rely on price deflators and top-down allocation, and do not exist at the subdistrict (taluka/tehsil) scale where place-based policy is targeted. The EPI fuses multi-sensor satellite data, gridded population and OpenStreetMap road and amenity networks across 357 subdistricts of Maharashtra, India, over 46 quarters (2015 Q1 - 2026 Q2). Seasonal-Trend decomposition using LOESS (STL) separates long-run trends from monsoon and crop-cycle seasonality. A non-negative PCA gives sign-consistent, interpretable weights and rankings nearly identical to standard PCA ($ρ$ = 0.998). Across districts, the index explains 74% of the variation in the NFHS-5 wealth index, 52% in household consumption (HCES 2023-24) and about 78% in real district GDP levels (65% in logs), but does not track short-run output growth. Because it measures structural capacity rather than annual output, the index kept rising through the FY2020-21 COVID-19 contraction. Across Maharashtra's six revenue divisions, relative inequality fell by 44% even as the absolute gap between leading and lagging divisions widened by 32%; district-level $β$-convergence implies a half-life of about 14 years. The framework offers a scalable, deflator-free basis for structural monitoring in data-sparse economies.

Soumis à arXiv le 18 Nov. 2026