Performance of Satellites to Predict GDP in South Asian Economies: A Comparative Analysis of DMSP and VIIRS Nightlights Data for District-level GDP Estimation in Pakistan

Abstract

Nightlights data have emerged as a valuable proxy for economic activity, enabling gross domestic product (GDP) prediction at sub-national levels. The availability of district-level GDP plays a crucial role in understanding regional economic disparities and formulating targeted policy interventions. However, the choice between the Defense Meteorological Satellite Program (DMSP) and the Visible Infrared Imaging Radiometer Suite (VIIRS) nightlights data remains a critical decision in the context of accurate GDP prediction. This research article aims to address the question: ‘DMSP or VIIRS: Which nightlights data to use for the prediction of GDP in South Asian economies?’ The renowned DMSP nightlights data, despite being widely utilized, have been subject to concerns regarding calibration, top coding and blurring, which may lead to biased GDP predictions and spatial inequality estimates. On the other hand, the VIIRS lights data, a more reliable alternative, have not been fully explored by economists for GDP prediction. To achieve the objectives of this study, the predictive performance of DMSP and VIIRS nightlights data for national GDP is compared, and the suitability of each dataset for district-level GDP estimation is assessed. Additionally, the research examines the impact of both datasets in countries with distinct economic characteristics, such as agricultural and industrial dependence, as well as varying population densities. The findings reveal that VIIRS lights data exhibit superior predictive capabilities for the national GDP of South Asian economies compared to DMSP lights data. This advantage becomes more apparent when using DMSP nightlights data in more spatially aggregated units, such as sub-national administrative units. Nevertheless, this performance difference between the two datasets persists across agriculture-dependent countries, industry-dependent countries, low-population-density countries and high-population-density countries. Notably, our results corroborate Gibson’s observations that the limitations of using DMSP lights data are less evident when applied to more aggregate spatial units, particularly at the national level. By shedding light on the comparative performance of DMSP and VIIRS data, this study contributes to advancing the understanding of the complexities involved in harnessing remote sensing data for economic analysis at the sub-national level.