Conformal Prediction for Geographically Weighted Functional Regression models: an application for environmental impact assessment

In this work we introduce a general framework for distribution-free predictive inference in Geographically Weighted Functional Regression (GWFR) using conformal inference. A prediction band for the functional response variable using a functional estimator of the regression function and a new non conformity mea- sure is thus introduced. We also investigate our procedure for producing prediction bands with locally varying length, in order to adapt the approach in presence of het- eroskedasticity in the data. An application for environmental impact assessment is proposed.