Research

Speaker: Henry Santer, University of Maryland

Probabilistic forecasting systems seek to recognize the uncertainty in numerical weather prediction forecasts, and frequently use ensembles to sample from a distribution of possible forecast outcomes. Accordingly, care must be taken to ensure that such ensembles are well-calibrated and straightforward to interpret. Common methods for interpreting and verifying ensemble-based forecasts emphasize pointwise statistics or marginal distributions. These methods provide an easily digestible overview of the center and spread of the forecast distribution. However, they also tend to “smooth out” information about significant spatial features in individual ensemble members. This project explores contour band depth (cBD), a method of analyzing uncertainty in terms of contours of scalar fields. cBD is fully nonparametric and induces an ordering on ensemble members that leads to box-and-whisker-plot-type visualizations of uncertainty for multivariate data. Experiments with cBD applied to synthetic ensembles demonstrate that it provides enhanced information about the spatial structure of ensemble uncertainty, although the usefulness of the cBD analysis depends on the presence of multiple modes and multiple scales within the ensemble. We demonstrate the potential value of cBD for analyzing ensemble data with a convective-scale application resembling the NOAA Warn-on-Forecast system, and with a hurricane application within the UFS. Finally, we note that the cBD-median leverages information from the entire ensemble to describe the ensemble center, and, unlike the ensemble mean, has the attractive property that it still is a true model solution. The talk will finish with a discussion of my ongoing research as a visiting graduate student at NCAR.