In this work, we propose a variable selection approach integrated within the well-known Fixed Rank Kriging (FRK) model. Variable selection plays a crucial role in statistical applications involving large datasets, where covariates can be particularly numerous. A correct identification of relevant regressors promotes model parsimony, reducing complexity and improving the interpretability of the results. This aspect becomes even more important in spatial contexts, where data heterogeneity and the spatial autocorrelation between observations require careful selection of covariates that significantly explain the studied phenomena. The idea is to leverage penalised techniques within the FRK estimation to shrink to zero the non significant coefficients of less relevant covariates. The result provided would be used in “Growing Resilient INclusive and Sustainability” (GRINS) project, which provides large-scale environmental datasets for all of Italy.