We currently have a Euclidean distance model, which is wrong when working with data on the globe.
This is probably less of a limitation than the fact that geographic boundaries aren't respected, but an improvement nonetheless.
A VonMises3D distribution, would be a better prior on internal locations than the current flat prior, constraining locations to the sphere. We'd need to transform this back into latitude and longitude.
We currently have a Euclidean distance model, which is wrong when working with data on the globe.
This is probably less of a limitation than the fact that geographic boundaries aren't respected, but an improvement nonetheless.
A VonMises3D distribution, would be a better prior on internal locations than the current flat prior, constraining locations to the sphere. We'd need to transform this back into latitude and longitude.