Improving the accuracy of extant gridded population maps using multisource map fusion

Abstract

Population distribution is the most direct indicator used to describe human activities. Grid-based population distribution maps overcome the drawbacks of statistical data and are thus more suitable for integrated analysis with environmental data. However, current modeling methods seeking to improve accuracy ignore the role of many existing products, resulting in the ineffective use of advantageous information from different gridded population maps. In this study, the multisource map fusion method is developed and combined with population mapping to simply and efficiently achieve improved results accuracy by understanding the uncertainty of different gridded products. Three areas in China with significant environmental differences were used as case studies to validate the results. The case studies use representative Granger-Ramanathan (GR), variance weighted (VW), and random forest (RF) algorithms to implement multisource map fusion for three existing population maps–GPW4, LandScan, and WorldPop. The results of the experiments indicate that fusing multisource population maps can produce a more accurate product than input maps. Compared with the highest accuracy input data, the maximum reduction percentages for the RMSE and MAE of fused maps at the grid scale are 13.66% and 20.39% in the Beijing, Tianjin, and Hebei Region (BTHR); 15.47% and 18.29% in Guangdong Province; and 5.05% and 6.15% in Guizhou Province. This study provides new strategies for producing high-accuracy population distribution maps, and its inexpensive features make it especially suitable for developing countries to produce wide-range gridded population maps that are more accurate than existing products using a few surveys.