Performance Evaluation of Prediction Methods of Social Statistics for Nighttime Light Data

The observed brightness of satellite nighttime lights (NTL) is often considered as a proxy measure for large-scale socio-economic indicators, such as population, gross domestic product (GDP), and energy consumption. However, few studies have explored and compared the correlation between NTL data and socioeconomic indicators from a grid scale and further analyzed their correlations. Aiming at the uncertainty problem in the method of social statistic prediction of NTL, based on DMSP-OLS, NPP-VIIRS NTL data and social statistics (GDP, population and energy consumption) of Beijing- Tianjin-Hebei Counties in 2013, five regression models including linear, quadratic, power, least-squares support vector machine (LSSVM) and BP neural network (BPNN) were used to predict social statistics at the scales of 500 m, 1 km and 2 km respectively and evaluate the predicted results. The results indicate that the accuracy of NPP-VIIRS data in predicting social-economic indicators was significantly higher than that of DMSP-OLS data; the BPNN model predicted the best social statistics while the power model predicted the worst; The accuracy of social statistics predicted by two NTL data on three scales is basically the same. The research results provide new insights into the use of data to estimate and forecast population, GDP and energy consumption.