Assessing accuracy improvement of integrating digital footprints into gridded population mapping: spatiotemporal variations and data bias

Population mapping has wide applications in various fields such as resources, environment, economy, and health. With the increasing availability of digital footprint data, integrating them into population mapping is expected to improve spatial resolution and mapping accuracy. However, there is a lack of systematic evaluation of the accuracy improvement effects of integrating digital footprint data, the spatiotemporal variations in mapping accuracy, and the impact of digital footprint data bias. To address this, this study implemented widely used univariate and multivariate population estimation methods within a unified population spatialization framework, utilizing mobile phone location request data (Tencent location request data, TLR) and geotagged microblog data to generate population distribution data at a 1-km grid scale. The results showed that integrating digital footprint data significantly improved population mapping accuracy, with the univariate model using TLR data achieving the best performance, reducing the RMSE by 8–92% compared to models using conventional population covariates. Further analysis revealed significant spatial heterogeneity in both the accuracy improvement effects of integrating digital footprint data and the mapping accuracy, with simple univariate models exhibiting higher accuracy in areas with lower population densities (<43 people/km2) and complex multivariate models performing better in areas with higher population densities (>2074 people/km2). Additionally, the TLR-based model showed evident diurnal variations in accuracy, with higher accuracy at night than during the day. Finally, this study quantitatively assessed the impact of digital footprint data bias on mapping accuracy, finding that TLR data had higher spatial consistency with census data compared to microblog data, and the data bias was positively correlated with mapping accuracy. The implementation of population constraint effectively reduced the impact of digital footprint data bias on mapping accuracy. The spatiotemporal variation patterns and influencing factors of mapping accuracy revealed in this study can provide important references for optimizing population mapping methods.