View metadata

dc.titlePutting Numbers on a Continent: Spatial Measurement of Economic Activity in Amazonia
dc.contributor.authorGarcía, Pablo M.
dc.contributor.authorDuran-Fernandez, Roberto
dc.contributor.authorFigueroa Guajardo, David Alejandro
dc.contributor.orgunitProductivity, Trade and Innovation Sector
dc.coverageAmazon Region
dc.date.available2026-03-27T00:03:00
dc.date.issue2026-03-27T00:03:00
dc.description.abstractDespite its continental scale and global relevance, Amazonia lacks spatially disaggregated, consistent GDP measures, leaving its economy largely unmeasured at the scale where development and environmental decisions are made. This paper introduces a scalable framework to estimate subnational economic activity by combining machine learning, nighttime lights (NTL), and spatial data within a unified 5×5 km grid. We develop Random Forest models integrating NTL, electricity consumption, population density, geographic features, spatial spillovers, and temporal dynamics. To assess the contribution of different information sources, we estimate six model specifications that progressively incorporate additional covariates, including a comparison between raw luminosity and satellite-derived GDP measures. To ensure realistic performance, we apply rolling-origin temporal cross-validation to prevent temporal leakage. Our findings show that combining NTL with spatial structure and temporal dynamics significantly improves subnational GDP estimation relative to luminosity-only methods. Crucially, the paper distinguishes forecasting over time from generalizing across space: while single-country models outperform in temporal prediction, multi-country training markedly improves spatial generalization in data-scarce regions. These findings highlight the importance of distinguishing temporal forecasting from spatial generalization when applying machine-learning models to economic measurement in large and heterogeneous territories.
dc.format.extent51
dc.identifier.doihttp://dx.doi.org/10.18235/0013995
dc.identifier.urlhttps://publications.iadb.org/publications/english/document/Putting-Numbers-on-a-Continent-Spatial-Measurement-of-Economic-Activity-in-Amazonia.pdf
dc.language.isoen
dc.publisherInter-American Development Bank
dc.subjectGross Domestic Product
dc.subjectInfrastructure Development
dc.subjectSustainable Development
dc.subjectElectricity Consumption
dc.subjectLand Use
dc.subjectEconomic Development
dc.subjectMachine Learning
dc.subjectForest Resource
dc.subject.jelcodeC53 - Forecasting and Prediction Methods • Simulation Methods
dc.subject.jelcodeC55 - Large Data Sets: Modeling and Analysis
dc.subject.jelcodeQ56 - Environment and Development • Environment and Trade • Sustainability • Environmental Accounts and Accounting • Environmental Equity • Population Growth
dc.subject.jelcodeE01 - Measurement and Data on National Income and Product Accounts and Wealth • Environmental Accounts
dc.subject.keywordsAmazonia;Nightime Lights (NTL);sustainable development
dc.typeTechnical Notes
idb.identifier.pubnumberIDB-TN-03319
idb.operationRG-T4395
Return to Publication