space-time GAM (simulated data, Chaco NDVI)

Data to support a publication: Abstract: This paper demonstrates an approach to application of Generalized Additive Models (GAMs) with space-time smooths to model processes that vary over space and time. The approach is to create and evaluate multiple GAMs, each with the predictor variables specified in different ways. It emphasises the need to determine the nature of the space-time dependencies present in the data relationships rather than to assume them, for example based on the perceived data generating process. The approach is explored using simulated data with known space-time dependencies. The GAMs are compared with Multiscale Geographically and Temporally Weighted Regression (MGTWR) models and shown to have marginally weaker predictive performance and to be marginally better at coefficient recovery. The inferential costs of mis-specifying variables in models are quantified both for individual variable main effects and interacting mis-specifications. The approach is then applied to an empirical case study of NDVI in the Chaco dry rainforest. The best model is determined and used to construct space-time varying coefficient estimates. The methods and results are discussed and number of areas of further work and enhancements to the `stgam` R package used to undertake this analysis, are identified.