Forecasting with Geospacial Data

Geo-statistics is a subfield of statistics focused on spatial or spatiotemporal datasets, AKA data with location or longitudinal data with location. 

 In the following case, we have a large set of longitudinal data with location and we want to make guesses about the future in those locations. Not to be confused with kriging  which is a gaussian process of estimating a variable at an  unobserved location, based on the estimates of nearby or similar locations.  

 Fun fact : Many of the techniques we use today originated from mining engineers who desired to know what minerals might lay beyond a surface given a sample. 

 The following notes aim to provide the fundamentals of creating a network of connected spaces, and forecasting techniques for spatial network data. 

 Networks 

 A  network is a discrete set of items (referred to as  nodes ) with some connection between them (referred to as  edges ). Typically, a graph network is represented mathematically as G = (N, E); with a set of N nodes and E edges. 

 Time Series 

 A parametric model can be used to describe a relationship between independent and dependent variables. It's not unusual for a time series to exhibit a trend, such as seasonality. 

 Think of a time series as a series of components:  

 

 A trend-cycle component 

 A seasonal component 

 A remainder component