Sparse Group Lasso for Regression on Land Climate Variables

Abstract

The large amount of reliable climate data available today has promoted the development of statistical predictive models for climate variables. In this paper we have applied Sparse Group Lasso to build a predictive model for land climate variables using ocean climate variables as covariates. We demonstrate that the sparse model provides better predictive performance than the state-of-the-art, is climatologically interpretable and robust in variable selection.