Whenever I talk to people that are doing geospatial analysis, they think that geospatial data is something very special. Well let us have a brief look at this.
From a mathematical perspective it is either vector data or raster data. The latter means that it is a 2+ dimensional array (tensor of rang 2+). Hence many algorithms we know and use in other fields of artificial intelligence are directly applicable. We simply have to remember that geospatial data is (mainly) at least 2 dimensional and linked to a reference frame called projection.
The fundamental assumption of geospatial data is that there is a spatial relation between points not in a classical topological. The underlying assumption is that data points are partial observations of a (mostly) continuous space meaning that points that are closer together are more similar than distant points. Another important assumption is that we assume (mostly) no overlapping of features in one layer and closed topology - there is no empty place in space.
Hence, geospatial is kind of special but not that much ;).