If you're in domain of, say, weather forecasting. You need a model correlating a number of variables both spatially and temporary. Spatial data is kind of hard

If you're in domain of, say, weather forecasting. You need a model correlating a number of variables both spatially and temporary. Spatial data is kind of hard to work with, and apparently common approaches hasn't incorporated time very well, focusing mostly on problems related to problems of space, so they improved it. Example of a problem, you have a set of sensors at various geographical locations. At some points between sensors you need to forecast a measure, say temperature. Even just interpolating a current value can be challenging, forecasting in time doesn't make it easier.