Estimating Socio-Economic Parameters via Machine Learning Methods Using Luojia1-01 Nighttime Light Remotely Sensed Images at Multiple Scales of China in 2018

Mapping socio-economic indicators with a raster format is still a great challenge. The nighttime light (NTL) datasets have been widely utilized to estimate the socio-economic parameters. However, the precision of the published datasets was too coarse to meet related issues such as flood losses assessment, urban planning, and epidemiological studies. The present study calibrated gross domestic product (GDP), population (POP), electric consumption (EC), and urban build-up area (B-A) at 100 m resolution for 45 cities of China in 2018 using Luojia1-01 NTL datasets via random forest (RF) as well as geographically weighted regression (GWR) model. The linear regression (LR), back propagation neural network (BPNN), and support vector machine (SVM) methods were selected for comparison with GWR and RF models. Besides, the Suomi National Polar-Orbiting Partnership-Visible Infrared Imaging Radiometer Suite (NPP-VIIRS) was chosen for comparison with Luojia1-01. The ten-folded cross-validation (CV) has been used for evaluating accuracy at county and city scales. Finally, the distribution maps of socio-economic parameters were illustrated and some findings were obtained. First, the validation results revealed that the calibration at the city-scale outperformed the county or district scale. Second, the precision of the Luojia1-01 NTL dataset surpassed the NPP-VIIRS NTL dataset on the same administrative scale except for some specific situations. Third, the precision of the simulation for the gross domestic product (GDP) is the highest than the others, followed by electric consumption (EC), build-up area (B-A), and population (POP). Fourth, the optimum model varied according to the socio-economic parameters. Fifth, the distribution of socio-economic parameters exhibited obvious spatial heterogeneity. This paper can supply scientific support for calibrating socio-economic parameters in other regions.

Abstract licence: CC BY-NC-ND

A person counts once a day per paper. We keep no cookie and no address: a reader is a hash that changes every day. Any month under five reads shows no number.