Abstract:

High-accuracy population grid datasets are of great value in the fields of risk assessment, disaster emergency response, ecological environment protection, and regional development. The characteristics and advantages of the datasets vary because of the different input data accuracy and model selection. Therefore, it is of great significance to evaluate the accuracy of datasets and analyze the applicable conditions of datasets. To this end, this study evaluated the accuracy of the WorldPop and GPWv4 datasets in the arid areas of the Gansu Province, Ningxia Hui Autonomous Region, and the Qinghai Province of northwest China. The accuracy of each dataset was quantitatively evaluated by calculating the spatial distribution of statistical and relative errors. Taking the best available unit of census data of China (township administrative division) as the research unit, the correlation analysis was conducted between the WorldPop and GPWv4 datasets and the seventh census data in 2020. The spatial distribution of statistical and relative errors is obtained through correlation analysis to quantitatively evaluate the accuracy of each dataset. Furthermore, the mapping performance of the dataset was qualitatively analyzed by visual estimation. Finally, the error sources of the dataset are discussed. The statistical error results show that WorldPop has higher accuracy than GPWv4. The correlation coefficient (r), root mean square error, average absolute error, and average absolute percentage error of WorldPop are 0.76, 23016, 0.73, and 0.60, respectively, while those of GPWv4 are 0.70, 22297, 0.75, and 0.58, respectively. Concurrently, according to the spatial distribution of the relative error, WorldPop accurately estimates the population of more areas. The visual estimation results show that the mapping performance of the two population grid datasets is similar, with the characteristics of a dense and sparse population in the east and west of the study area, respectively. This study on the accuracy of population grid datasets in arid areas is conducive to analyzing the error sources of datasets and guiding the rational use of datasets. In future research, it would be a beneficial direction to use the auxiliary data of human life to generate a unique population distribution pattern in an arid area to improve the accuracy of population datasets in the northwest arid area of China.

Key words: population grid dataset, GPWv4 data set, WorldPop data set, accuracy evaluation, arid area in northwest China