用于即时预报的卫星数据：在机器学习框架中使用卫星数据实时估算柬埔寨的国内生产总值

Satellite Data forNowcasting: EstimatingCambodia’s GDP inReal Time Using Satellite Data Iyke Maduako, Dharana Rijal, Alberto Sanchez Rodelgo. SIP/2026/001 IMF Selected Issues Papers are prepared by IMF staff asbackground documentation for periodic consultationswith member countries.It is based on the information 2026JAN IMF Selected Issues Paper Asia Pacific Department Satellite Data for Nowcasting: Estimating Cambodia’s GDP in Real Time Using Satellite Data in aMachine Learning Framework Authorized for distribution by Kenichiro Kashiwase IMF Selected Issues Papersare prepared by IMF staff as background documentation for periodicconsultations with member countries.It is based on the information available at the time it was ABSTRACT:Cambodia is not alone in facing capacity limitations in the production and timely release of keyofficial statistics neededfor data-driven policy decisions. This paper demonstrates that combining satellite-derived indicators (e.g., nighttime lights, NO₂emissions, vegetation indices) with traditional high-frequency RECOMMENDED CITATION:Maduako, I., Rijal, D. & Sanchez Rodelgo, A. (2025). Satellite Data forNowcasting: Estimating Cambodia’s GDP in Real Time Using Satellite Data in a Machine Learning Framework. SELECTED ISSUES PAPERS Satellite Data for Nowcasting:Estimating Cambodia’s GDP in Real Time UsingSatellite Data in a Machine Learning Framework Cambodia Prepared by Iyke Maduako, Dharana Rijal, Alberto Sanchez Rodelgo. CAMBODIA SELECTED ISSUES November 5, 2025 ApprovedByAsia and Pacific Preparedby Iyke Maduako, Dharana Rijal, Alberto SanchezRodelgo. CONTENTS SATELLITE DATA FOR NOWCASTING ________________________________________________ 3 A. Motivation – Why Satellite Data _____________________________________________________ 3B. Data and Methodology_______________________________________________________________ 4C. Results and Interpretation____________________________________________________________ 8References _____________________________________________________________________________ 10 BOX 1. Satellite Indicators to Gain Timely and Granular Insights on MacroeconomicDevelopments__________________________________________________________________________5 SATELLITE DATA FOR NOWCASTING A.Motivation – Why Satellite Data 1.Cambodia faces limited institutional capacity in the production and timely release ofquality official statistics, limiting policymakers’ ability to make agile and effective policydecisions.While the country has made significant improvements on the availability and quality ofnational statistics, further strengthening of statistical capacity is needed. GDP data is available onlyat annual frequency and published with a significant lag, limiting timely analysis of comprehensive 2.Satellite indicators are available for nearly all countries in the world. They exist innear-real time and at granular levels capturing nuances that may otherwise go undetected. They can serve as proxies for economic activity in various sectors of the economy. In particular, dataon nighttime lights, nitrogen dioxide (NO2) emissions, and vegetation-related indices can helpuncover underlying patterns and trends in sectors like manufacturing and agriculture.2In Cambodia,quarterly GDP growth rate (interpolated, see section on Data and Methodology) is positively 3.Machine learning models can make the best use of satellite indicators, along withmacroeconomic data to analyze their complex interactions for nowcasting GDP.First, thedataset is split into 'train' and 'test' sets. The model learns patterns based on the train set, and itspredictions are then evaluated against observed values in the test set—data that was not used B.Data and Methodology 4.The machine learning method applies quarterly satellite indicators, along with thetraditional variables, for training the nowcasting model.The satellite (“non-traditional”)indicatorsused in this analysis include data on nighttime lights (NTL), NO2 emissions (NO2),precipitation (PCP), normalized difference vegetation index (NDVI), enhanced vegetation index (EVI), •Nighttime Lights (NTL)are satellite-based measurements of the intensity of light emitted •Nitrogen dioxide (NO₂)is a pollutant, primarily produced by the combustion of fossil fuelsin power plants, industrial facilities, and vehicles. Because NO₂ is emitted in large quantities when economic activity is high, satellite-based observations of NO₂ can approximate the •Normalized Difference Vegetation Index (NDVI)andEnhanced Vegetation Index (EVI) are computed using the red (R) and near-infrared (NIR) bands of satellite imagery. Theseindices measure vegetation health and can be used to proxy agricultural output, and land •Agricultural Stress Index (ASI)is a satellite-based indicator designed to detect areas ofcropland experiencing water stress—such as drought conditions—during the growing •Vegetation Health Index (VHI)is computed using NDVI and Land Surface T