No abstract

Abstract. Flood forecasts are essential to issue reliable flood warnings and to initiate flood control measures on time. The accuracy and the lead time of the predictions for head waters primarily depend on the meteorological forecasts. Ensemble forecasts are a means of framing the uncertainty of the potential future development of the hydro-meteorological situation.This contribution presents a flood management strategy based on probabilistic hydrological forecasts driven by operational meteorological ensemble prediction systems. The meteorological ensemble forecasts are transformed into discharge ensemble forecasts by a rainfall-runoff model. Exceedance probabilities for critical discharge values and probabilistic maps of inundation areas can be computed and presented to decision makers. These results can support decision makers in issuing flood alerts. The flood management system integrates ensemble forecasts with different spatial resolution and different lead times. The hydrological models are controlled in an adaptive way, mainly depending on the lead time of the forecast, the expected magnitude of the flood event and the availability of measured data.The aforementioned flood forecast techniques have been applied to a case study. The Mulde River Basin (SouthEastern Germany, Czech Republic) has often been affected by severe flood events including local flash floods. Hindcasts for the large scale extreme flood in August 2002 have been computed using meteorological predictions from both the COSMO-LEPS ensemble prediction system and the deterministic COSMO-DE local model. The temporal evolution of a) the meteorological forecast uncertainty and b) the probability of exceeding flood alert levels is discussed. Results from the hindcast simulations demonstrate, that the sysCorrespondence to: J. Dietrich (joerg.dietrich@rub.de) tems would have predicted a high probability of an extreme flood event, if they would already have been operational in 2002. COSMO-LEPS showed a reasonably good performance within a lead time of 2 to 3 days. Some of the deterministic very short-range forecast initializations were able to predict the dynamics of the event, but others underpredicted rainfall. Thus a lagged average ensemble approach is suggested. The findings from the case study support the often proposed added value of ensemble forecasts and their probabilistic evaluation for flood management decisions.

Abstract. The development and use of nowcasting systems should inevitably be accompanied by the development and application of suitable verification methods. A thorough verification strategy is needed to adequately assess the quality of the system and consequently to lead to improvements. Different verification methods for thunderstorms and its attributes are discussed along with the importance of observational data sets. They are applied to two radar-based nowcasting algorithms for a convective season using various observation data sets. The results show, that the combination of the two algorithms outperforms a single algorithm.

The following results have been obtained from long-term observations on the ozone layer and UV at the Meteorological Observatory Hohenpeigenberg:The seasonally varying decline of the ozone layer determines the maximum exposure to UV. Since ozone decline shows the highest rates in the spring months the UV exposure has most strongly increased in this time of the year. This is especially important because in spring the human skin is not adapted to UV exposure. Weather changes from day to day can induce rapid ozone reductions in spring about -30% which in turn is followed by an increase in UV of about 40%. Clouds, especially the transparent cirrus clouds (high clouds consisting of ice particles) have increased in frequency during spring and fall while a decrease is observed in summer. This change in cloudiness reduces the daily UV dose in spring and fall while it is enhanced in summer. With increasing height above sea level UV rises by roughly 10% per 1000 m (rule of thumb). Snow reflects the UV-radiation by up to 80% enhancing the UV-doses at relevant conditions. Strong volcano eruptions destroy ozone in the stratosphere additionally during 1-2 years after the eruption. Therafter the ozone layer recovers. In April 1993, after the eruption of Mt. Pinatubo (1991), the UV burden was still 40% higher than average. Miniholes and streamers can appear unexpected on a short-time scale and cross over Central Europe within 1-2 days, thus enhancing UV irradiation. The human skin reacts to UV exposure depending on the type of skin. The campaign "Sonne(n) mit Verstand" of the Bavarian Ministries for Environment, for Health and for Education informs about the danger of UV radiation (see www.sonne-mit-ver-stand.de). The German Weather Service informs the public on present developments of the ozone layer and relevant topics byits ozone bulletin, which is also available via internet under (www.dwd.de/deFundE/Observator/MOHp/hp2/ozon/bulletin.htm).

<p>Many Numerical Weather Prediction (NWP) models provide the parameter total snow depth as a Direct Model Output (DMO) surface variable. In mountain regions, however, the orographic flow modification significantly influences precipitation formation and preferential settling, leading to large model biases if DMO is directly compared to fresh snow point observations. Avalanche risk forecasts in turn require calibrated deterministic and probabilistic fresh snow forecasts, as the amount of fresh snow constitutes a crucial driver of avalanche risk.</p><p>In this study, MOSMIX-SNOW, a Model Output Statistics (MOS) product based on multiple linear regression is developed. Ground-based observations and operational forecast data of the two deterministic global NWP models ICON and ECMWF form the basis of the MOS system. MOSMIX-SNOW offers point forecasts for 20 deterministic as well as probabilistic forecast variables like the amount of fresh snow within 24h, the probability of more than 30cm of fresh snow within 24h and some basic variables like 2m temperature and dew point. The unique characteristic of MOSMIX-SNOW is the large number of observation-based, model-based and empirical predictors, which exceeds 200. Furthermore, a long historical data period of 9 years is applied for training of the MOS system. Thus, local orographic effects and large scale flow patterns are implicitly included in the MOS equations by a location and lead time specific choice of predictors. To avoid unrealistic jumps in the forecast, persistence predictors, which represent the forecast value of the previous forecast hour, are included in the MOS system. All forecasts feature a maximum lead time of +48h, have an hourly forecast resolution as well as update cycle and are available for about 15 mountain locations in the Bavarian Alps between 1100m and 2400m above sea level.</p><p>The verification analysis of the winter season 2018/19 shows that MOSMIX-SNOW forecasts offer a significantly higher forecast reliability than the raw ensemble of the regional NWP model COSMO-D2-EPS. The bias of the deterministic forecast parameters is smaller for MOSMIX-SNOW, especially for heavy snowfall events. MOSMIX-SNOW turned out to be a useful tool to support the avalanche risk forecasts on a daily basis during the snowy winter of 2018/19. Furthermore, the deterministic fresh snow forecast of MOSMIX-SNOW and other meteorological parameters like 2m-temperature serve as input for operational snowpack simulations. Measurement related noise and snow drift in the observations, however, are identified as an important source of uncertainty and the application of noise reduction techniques like a Savitzky-Golay filter are expected to have a beneficial impact on the forecast quality. MOSMIX-SNOW will become operational by end of 2020.</p>