BSCALE: Downscaling of precipitation: development, calibration and validation of a probabilisitc Bayesian approach.
Hydrogeology, Hydrology, Limnology, Urban Water Management, Water Chemistry, Integrated Water Resources Management
Final Report Abstract
Downscaling of atmospheric model output in weather forecasting, especially precipitation, is required to map the meteorological quantity from the low-resolution spatial grid of weather models (e.g., 35km x 35km) to a selected location (e.g., a ground site) where the corresponding quantity is needed for operational use. Such applications include, but are not limited to, precipitation and runoff forecasts, climate forecasts, or weather forecasts for renewable power generation (temperature, wind speed). Typically, this involves establishing and calibrating stochastic relationships between historical observations and weather model outputs as predictors to subsequently estimate the future site-specific quantity, the predictor, as best as possible. Usually, linear regressions or probabilistic methods such as quantile matching are adopted. In any operational forecast, it is also relevant to determine the forecast uncertainty, expressed as the conditional probability density of future weather events on model forecasts, since predictors are not deterministic quantities. Especially when dealing with precipitation, there is more over the challenge of capturing extreme values, which cause extreme floods, such as experienced 2021 in the Ahr valley, Germany. In terms of its stochastic representation, precipitation is to be understood as a mixed process, which is composed of a binary (precipitation/no precipitation) process and a continuous process (precipitation depth). The BSCALE project proposes to establish these complex relationships using a Bayesian approach that first transforms the meteorological variable into Gaussian space, then assumes the relationship between predictand and predictors as multivariate normal, and then transforms this relationship back into the original space. This approach has been shown to satisfactorily reproduce the predictive density and the tails representing the extremes, as confirmed by the verification. The application of the proposed method is not limited to in weather forecasting, but can also be applied to climate projections. This was illustrated in BSCALE as a spin-off application using a post-processing of temperature forecasts for the 21st century over northern Italy from the CMIP5 climate model ensemble and two carbon scenarios RCP4.5 and RCP8.5. In this context, the forecast uncertainties obtained with the procedure developed in BSCALE were compared against those obtained with the established Reliability Ensemble Averaging (REA) procedure. As a final research task in BSCALE, inspired by the Ahr flood in summer 2021 we performed an analysis of flood risk and hazard due to extreme precipitation. In doing so, a statistical extreme value analysis for the River Ahr was used to show that existing HQ100 extreme value statistics based on 1-year runoff extreme values greatly underestimate the risk of extraordinary flood events. The extension and application of the extreme value statistics to longer periods, for example 50, 100 or 200 years, confirm that the prevailing risk of extraordinary flood events is considerably higher than assumed by the public side, and thus a new approach of water management risk assessment would actually be necessary.
Publications
- (2019) A Bayesian processor of uncertainty for precipitation forecasting using multiple predictors and censoring, Monthly Weather Review, 147(12)
Reggiani, P. and O. Boyko
(See online at https://doi.org/10.1175/mwr-d-19-0066.1) - (2019). Probabilistic precipitation analysis in the Central Indus River basin, In Indus River Basin: Water Security and Sustainability edited by T. Adams and S. Khan, Elsevier Science
Reggiani, P., A. Boyko, T.H.M. Rientjes and A. Khan
(See online at https://doi.org/10.1016/b978-0-12-812782-7.00005-9) - (2021). Assessing uncertainty for decision-making in climate adaptation and risk mitigation. International Journal of Climatology; 1-22
Reggiani, P., E. Todini, O. Boyko and R. Buizza
(See online at https://doi.org/10.1002/joc.6996)
