This invention relates to the field of
population spatiotemporal distribution modeling technology, and proposes a
population model generation method based on crowdsourced data, comprising the following steps: Step 1, collecting socioeconomic and
natural environment data to obtain standardized multi-source
raster data; Step 2, based on the standardized data, constructing a
daytime population weight layer by training a heterogeneous basic regression model with five-fold cross-validation and building an adaptive spatial weighted support vector regression meta-learner; Step 3, extracting building volume and nighttime
light intensity raster
layers from the standardized data to construct a nighttime population weight layer; Step 4, constructing a non-working day population weight layer by weighted fusion based on the day and night weight
layers; Step 5, constructing basic
spatial mapping units to complete high-scale
downscaling reconstruction of population space, obtaining initial high-resolution population distribution results; Step 6, introducing a time
smoothing model to correct the temporal inconsistency problem of the initial results, generating the final population
distribution model. This invention effectively solves the problems of insufficient
spatiotemporal resolution, poor time-segment adaptability, and unreasonable assignment of
missing data areas in traditional population distribution models, providing accurate, comprehensive, and highly
usable population distribution data support for scenarios such as
urban planning,
resource allocation, and emergency management.