Perturbing Topography in a Convection-Allowing Ensemble Prediction System for Heavy Rain Forecasts

Four methods to perturb model terrain are demonstrated in a convection-allowing scale (3 km) ensemble prediction system (EPS) for heavy rain forecasts. The impact of methods was examined in terms of ensemble mean, spread-skill relationship and probabilistic forecast. The study was carried out for four major precipitation events. It is found that: (a) using different combinations of topography smoothing and interpolation is a promising way to perturb model topography. Even perturbing the topography alone can greatly improve heavy rain forecasts. The benefit of ensemble forecasts increases with the increase in rainfall amount. (b) Adding random perturbations to model topography is not ideal and could hurt ensemble performance due to a too noisy structure although it increases ensemble spread. Perturbation structure plays a more important role than perturbation size. (c) Using the terrain difference between a higher and lower resolution model to estimate possible errors in model terrain is explored. By adding such terrain differences to the terrain-generating-scheme based perturbation (the control experiment) can not only enlarge perturbation size but also keep a good spatial structure. Such a modified ensemble performed similarly in ensemble mean forecasts but improved the ensemble spread and probabilistic forecast. It is recommended that perturbed model terrain should be included in operational EPSs to improve heavy precipitation forecasts. The simple terrain-generating-scheme based perturbation method, as described in this study, is effective and can be implemented in an existing operational EPS without extra computing cost.