Towards operational subseasonal hydrometeorological ensemble predictions in mountainous catchments

Abstract

Subseasonal hydrometeorological predictions have received increasing attention within the last decades. Significant advances in meteorological ensemble forecasting have led to skilful predictions of meteorological variables beyond the medium-range forecast horizon. Using these meteorological ensemble forecasts to force hydrological models can provide valuable probabilistic streamflow forecasts. Such predictions at the subseasonal time horizon can have a great impact for planning purposes in various economic and public sectors. Currently, the scientific basis for the predictability of subseasonal forecasts is being investigated and various meteorological organizations are running numerical weather predictions models for the subseasonal time-scale. In order to improve our knowledge to predict streamflows in small to medium sized mountainous catchments at the subseasonal time scales this PhD thesis focuses on the interface between the meteorological and the hydrological predictions and the assessment of the performance thereof. Thus, the current state of subseasonal hydrometeorological predictions is explored and statistical bias correction and downscaling methods are investigated to optimally combine meteorological and hydrological prediction models bridging the gap of application scales. Furthermore, a hydropower optimization system making use of the resulting streamflow predictions has been explored. In a first step the historical forecasts from subseasonal ECMWF Integrated Forecasting System have been analysed in terms of their performance to predict temperature and precipitation at 1637 measurement stations across Europe. Aside of the uncorrected direct model output, different post-processing techniques have been applied and their effect on the forecast performance is analysed. The results clearly demonstrate the need for post-processing the subseasonal predictions to achieve skilful subseasonal forecast for point observations. Post-processed forecasts indicate positive skill with respect to climatology for up to 19-25 days lead time (corresponding to forecast week 3) in case of weekly mean temperature. Forecast skills of weekly precipitation sums stay positive up to 5-11 days lead time (corresponding to forecast week 1) and clearly outperform the uncorrected forecasts. These statistically corrected subseasonal meteorological predictions with daily resolution are used to generate streamflow forecasts with the hydrological model PREAVH in small to medium size mountainous catchments (with areas between 80 and 1700 km2) of different hydroclimatic characteristics. Furthermore, the performance of the resulting streamflow forecasts is compared with the forecast performance of a traditional ensemble streamflow prediction (ESP) approach based on historical meteorological observations. The study clearly demonstrates the superiority of the subseasonal NWP-hydro prediction system. It is found that the benefits are most pronounced in the snow-dominated catchments and this underlines the importance of snow-related processes in subseasonal hydrometeorological predictions. In an additional analysis, the ensemble streamflow predictions were assessed in terms of their benefits for water resource management. To this end, the uncorrected meteorological predictions are used to generate streamflow forecasts and the effect of different hydrological post-processing techniques on the forecast performance is analysed. Besides the total runoff itself, multiple hydrologically relevant variables are analysed for 370 sub-catchments covering entire Switzerland: areal catchment precipitation, total baseflow and soil moisture storage. The results stress the importance of persistence and memory effects on the performance of streamflow forecasts. Subsurface processes were found to show a delayed response of one week in forecast performance. This can be of particular importance for applications in water resource management. Finally, the resulting subseasonal ensemble streamflow predictions were used in a hydropower optimization setup in the Swiss Alps to further assess the potential of the forecast in terms of their economic value. We found that the gain in forecast performance can indeed further translate into monetary benefits, i.e. economic value of a forecast. Depending on the optimization scheme used, the average additional gain of up to 4% could be expected in terms of revenues if such predictions were used in an operational manner instead of the climatological predictions (corresponding to an increase of 0.2 Mio EUR/yr in the Verzasca catchment investigated in this study).