# C7 - Statistical postprocessing and stochastic physics for ensemble predictions

*Principal investigators: Prof. Dr. Tilmann Gneiting, Dr. Kirstin Kober*

*Other researcher: Dr. Sebastian Lerch (Postdoc)*

Ensemble forecasts seek to quantify uncertainty in numerical predictions of atmospheric processes. A critical requirement for the resulting probabilistic forecasts is that they are calibrated (i.e. the predicted probabilities are reliable) and sharp (i.e. more concentrated than climatological forecasts). Both the design of the ensemble prediction system itself and any statistical postprocessing of the resulting model output aim to maximize sharpness, subject to calibration. An important consideration is that forecast uncertainty varies between weather regimes.

Central tasks of this project are to develop regime-dependent statistical postprocessing methods, and to assess the effects of stochastic physics parameterizations, with the overarching goal of a comparative quantification of the value and relative merits of the two approaches for quantitative precipitation.

In the statistical postprocessing effort, we will develop flow dependent techniques for precipitation forecasts, using ramifications of ensemble model output statistics equipped with cutting edge Bayesian computing, along with empirical copula based methods, such as ensemble copula coupling.

Concerning stochastic physics, the advanced process-oriented methods developed in Project A6 will be applied and tested against more general perturbations. For a comparative assessment of predictive performance, addressing both calibration and sharpness, we will use proper scoring rules, and the development and mathematical study of novel proper scoring rules that are tailored to probabilistic quantitative precipitation forecasts forms an integral part of this project.