Modelling & prediction
Predicting values where nobody measured them, and deciding where things should go — with the limits of each model attached.
Three topics that answer questions the data does not contain: what is the value where nobody measured, and where should the next thing go?
Both are answerable, and both are easy to overstate. An interpolated surface honours every sample and is a guess everywhere else — and which guess depends on the method, so the same five measurements produce different surfaces under IDW, kriging and a spline. A machine-learning model scores well on a random split because neighbouring rows are nearly the same row, and it will predict cheerfully over ground it has never seen. A location model optimises exactly the objective you wrote down, which is rarely exactly the objective you meant.
What ties them together is the third thing every prediction needs: not just a value, but the area and the conditions it is allowed to speak for.
Topics in Modelling & prediction
Each topic page brings that subject's concepts, how-to guides and fixes onto one page.
- Spatial machine learning Features, spatial cross-validation, prediction rasters and the area a model can honestly speak for.
- Interpolation & surfaces IDW, kriging, TIN and variograms — turning scattered samples into a surface, with its support attached.
- Location & accessibility Catchments, access to services, gravity models, site suitability and choosing where the next facility should go.