Mapspace

Summary

The explosive growth of socially-sensed geolocated data (i.e., geosocial data) has offered us an unprecedented opportunity to model global land use dynamics at varying spatial scales, semantic granularities and  temporal resolutions.  Geosocial data (e.g., geotagged social media, user-generated ground images, POI data), with its universal availability and important role in daily human activities, has been widely used for land use characterization. MapSpace aims develop a time-variant global land use model with multiple spatial scales and semantic granularities. The first phase of MapSpace has developed a POI-based global land use modeling framework based on global POI data from the PlanetSense project.  The combination of spatial distribution, semantic characteristics, and sometimes temporal dynamics of POIs inside an AOI can capture its unique land use characteristics. Heterogeneous POIs data sources were fused using a unified semantic representation framework and combined with a neural network language model to generate spatially explicit AOI embedding at different spatial scales. Road network hierarchy was incorporated into the neural network language model to integrate spatial dimension and semantic dimension of POIs. The scalable land use modeling framework can be easily extended to incorporate other geosocial data sources and  efficiently produce global land use map at different spatial scales and semantic granularities.