ABSTRACT
The predictive performance of a spatial interpolation method can be assessed in several ways. For example, people can compare the observations and the predictions made at a set of locations using the Mean Absolute Error (MAE), the Root Mean Squared Error, and the 95% Coverage Probability. If the spatial interpolation method provides prediction distributions, the Continuous Ranked Probability Score (CRPS) can be used to compare observations and predictions accounting for the uncertainty. The CRPS for a set of observations can be calculated by aggregating the CRPS of the individual observations using an average or weighted average. A perfect CRPS score is equal to 0. Note that the CRPS reduces to the MAE if the predicted distribution is just a point estimate and not a distribution. The performance indices can be computed using a new dataset or by splitting an existing dataset into a training dataset to fit the model and a testing dataset for validation.
