The development and evaluation of a hydrological seasonal forecast system prototype for predicting spring flood volumes in Swedish rivers

Hydropower makes up nearly half of Sweden's electrical energy production. However, the distribution of the water resources is not aligned with demand, as most of the inflows to the reservoirs occur during the spring flood period. This means that carefully planned reservoir management is require...

Full description

Bibliographic Details
Published in:Hydrology and Earth System Sciences
Main Authors: Foster, Kean, Uvo, Cintia Bertacchi, Olsson, Jonas
Format: Article in Journal/Newspaper
Language:English
Published: European Geophysical Society 2018
Subjects:
Online Access:https://lup.lub.lu.se/record/1a5b1055-a7ee-4e23-b212-b5f750b994b7
https://doi.org/10.5194/hess-22-2953-2018
Description
Summary:Hydropower makes up nearly half of Sweden's electrical energy production. However, the distribution of the water resources is not aligned with demand, as most of the inflows to the reservoirs occur during the spring flood period. This means that carefully planned reservoir management is required to help redistribute water resources to ensure optimal production and accurate forecasts of the spring flood volume (SFV) is essential for this. The current operational SFV forecasts use a historical ensemble approach where the HBV model is forced with historical observations of precipitation and temperature. In this work we develop and test a multi-model prototype, building on previous work, and evaluate its ability to forecast the SFV in 84 sub-basins in northern Sweden. The hypothesis explored in this work is that a multi-model seasonal forecast system incorporating different modelling approaches is generally more skilful at forecasting the SFV in snow dominated regions than a forecast system that utilises only one approach. The testing is done using cross-validated hindcasts for the period 1981-2015 and the results are evaluated against both climatology and the current system to determine skill. Both the multi-model methods considered showed skill over the reference forecasts. The version that combined the historical modelling chain, dynamical modelling chain, and statistical modelling chain performed better than the other and was chosen for the prototype. The prototype was able to outperform the current operational system 57% of the time on average and reduce the error in the SFV by ∼ 6% across all sub-basins and forecast dates.