The CAST package for training and assessment of spatial prediction models in R

• DOI: 10.48550/arxiv.2404.06978
• Publication Date: 2024-04-10
• Authors: Hanna Meyer; Marvin Ludwig; Carles Milà; Jan Linnenbrink; Fabian Schumacher
• Keywords: R package

One key task in environmental science is to map variables continuously space or even and time. Machine learning algorithms are frequently used learn from local field observations make spatial predictions by estimating the value of variable interest places where it has not been measured. However, application machine strategies for mapping involves additional challenges compared "non-spatial" prediction tasks that often originate autocorrelation training data independent identically distributed. In past few years, we developed a number methods support which development suitable cross-validation performance assessment model selection, feature assess area applicability trained models. The intention CAST package predictive implementing such making them available easy integration into modelling workflows. Here introduce its core functionalities. At case study plant species richness, will go through different steps workflow show how can be more reliable predictions.