Quantifying the spatial distribution of population is a basis and hot issue in population geography researches. At present, there are large differences between different scales of spatialized population data in the world, because of various production methods, data sources, etc. This leads to the inconsistency of population spatialization, especially the 1 km-scale data which is widely needed. This paper takes Beijing-Tianjin- Hebei region as study area to build a population spatialized model at 1 km spatial resolution, based on multi-source data such as the township scale census data in 2000 and available mobile location data. The statistic population distribution weight (p) is calculated using the light projection method. Preliminary population spatialization is calculated using the area-weighted method, and the preliminary data is further modified by the exponential smoothing algorithm. Finally, the population spatialization dataset (PJ2000) with 1 km resolution in Beijing-Tianjin-Hebei region is obtained. This dataset integrates the small-scale characteristics of the township street demographic data and the advantages of mobile phone location data. The PJ2000 dataset reflects the actual location and the detailed characteristics of the population distribution in Beijing-Tianjin-Hebei region. Combined with the population density dataset (i.e., WorldPop) and China's kilometer gridded population spatial distribution dataset, the accuracy assessment of PJ2000 is carried out from three aspects: method difference, quantitative error, and regional comparison. The PJ2000 dataset solves the problem of the different distribution of population density over the same land cover type but different towns, and addresses the large difference in the gridded data of population spatialization. The overall accuracy of PJ2000 dataset is 90%, with 87% townships (streets) showing relative error less than 0.5. The correlation coefficient (r) between PJ2000 and the pop2000 township demographic data in the year of 2000 is 0.95. In addition, the population density distribution of this dataset is relatively uniform at the local to large scale. Our results prove that the accuracy of the population density dataset with 1km scale is significantly improved. The population spatialization model is constructed by integrating multi-source data such as township-level demographic data and mobile location data. In the future, it is expected that this method could be applied to obtain the population spatialization distribution for other city agglomerations. Our model could provide high-quality population density dataset for collaborative development of urban agglomeration and risk assessment of natural and man-made disasters in cities, such as earthquake, flood, fire, and public infectious diseases.