Percentage of test points outside the predicted 3σ (left), 2σ (middle), and 1σ (right) prediction interval, shown for Gaussian process models using Matérn (magenta) and RBF (black) kernels. The horizontal dotted lines represent the theoretical expected percentage of points that lie outside the “n” sigma prediction interval. The RBF kernel overshoots the percentage of points outside the prediction interval compared to Matérn in most flare classes for the 3σ performance evaluation, rendering the Matérn kernel the better choice for uncertainty estimation.