Assessment and weighting of meteorological ensemble forecast members based on supervised machine learning with application to runoff simulations and flood warning

Abstract

The version presented here may differ from the published version or from the version of record. If you wish to cite this item you are advised to consult the publisher's version. Please see the repository url above for details on accessing the published version and note that access may require a subscription. AbstractNumerical weather forecasts, such as meteorological forecasts of precipitation, are inherently uncertain. These uncertainties depend on model physics as well as initial and boundary conditions. Since precipitation forecasts form the input into hydrological models, the uncertainties of the precipitation forecasts result in uncertainties of flood forecasts. In order to consider these uncertainties, ensemble prediction systems are applied. These systems consist of several members simulated by different models or using a single model under varying initial and boundary conditions. However, a too wide uncertainty range obtained as a result of taking into account members with poor prediction skills may lead to underestimation or exaggeration of the risk of hazardous events. Therefore, the uncertainty range of model-based flood forecasts derived from the meteorological ensembles has to be restricted.In this paper, a methodology towards improving flood forecasts by weighting ensemble members according to their skills is presented. The skill of each ensemble member is evaluated by comparing the results of forecasts corresponding to this member with observed values in the past. Since numerous forecasts are required in order to reliably evaluate the skill, the evaluation procedure is time-consuming and tedious. Moreover, the evaluation is highly subjective, because an expert who performs it makes his decision based on his implicit knowledge.Therefore, approaches for the automated evaluation of such forecasts are required. Here, we present a semi-automated approach for the assessment of precipitation forecast ensemble 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 2 members. The approach is based on supervised machine learning and was tested on ensemble precipitation forecasts for the area of the Mulde river basin in Germany. Based on the evaluation results of the specific ensemble members, weights corresponding to their forecast skill were calculated. These weights were then successfully used to reduce the uncertainties within rainfall-runoff simulations and flood risk predictions. *Manuscript Click here to view linked References

Comparison of the GPM IMERG Final Precipitation Product to RADOLAN Weather Radar Data over the Topographically and Climatically Diverse Germany

“…The gauge-adjusted quality-controlled RADOLAN RW (Radar Online Adjustment) dataset from the German Weather Service (DWD, Deutscher Wetter Dienst) is considered ground truth for the upcoming analyses. It is already widely used, e.g., for training and validation purposes in the machine learning domain [49,50], analyzing extreme flash floods [51], as well as enhancing the respective forecasts [52] and estimating the spatio-temporal variability of soil erosion [53].…”

Bayesian network model for flood forecasting based on atmospheric ensemble forecasts

“…The NCEP Global Forecast System (GFS) final analysis (FNL) data were used as the initial conditions of the WRF. The model settings were based on the Noah land surface model (Chen and Dudhia, 2001), the Rapid Radiative Transfer Model (RRTM) longwave radiation scheme (Mlawer et al, 1997), the Dudhia shortwave radiation model (Dudhia, 1989), the Yonsei University (YSU) planetary boundary layer scheme (Hong et al, 2006) and the WRF Single-Moment (WSM) three-class microphysics scheme (Hong et al, 2004). Because of the importance of cumulus parameterization for hydrological purpose, an ensemble was created by using five cumulus schemes including KF, BMJ, GR3D, MSKF and GDE cumulus scheme.…”

“…Leandro et al (2019) reduced the ensemble to the upper and lower range of the uncertainty band. Other concepts of deriving a single (deterministic type) warning indicator from ensembles are weighting of ensemble members, e.g., averaging by Bayesian model average (Raftery et al, 2005), by machine learning (Doycheva et al, 2017) or by reduction of members to create a multi-model sub-ensemble (Dietrich et al, 2009b).…”

Flood Prediction Using Machine Learning, Literature Review

“…Through an EPS model, Reference [189] aimed at limiting the range of the uncertainties in runoff simulations and flood prediction. The classifier ensembles included MLP, SVM, and RF.…”