Developing a Sub-seasonal Forecasting System for Hydropower Reservoirs in Scotland 

Abstract

<p>Inflow forecasts play an essential role in the management of hydropower reservoirs. Forecasts help operators to mitigate flood risks, meet environmental requirements, and maximise the value of power generated. In Scotland, operational inflow forecasts for hydropower facilities are typically limited in range to 2 weeks ahead, which marks the predictability barrier of deterministic weather forecasts. Extending the horizon of these forecasts may allow operators to take more proactive responses to risks of adverse weather conditions, thereby improving water management and increasing profits.</p><p>This study outlines a method of producing skilful probabilistic inflow forecasts for hydropower reservoirs on sub-seasonal timescales (up to 6-weeks ahead), directly from Numerical Weather Prediction (NWP) model output. Using a case study site of a large hydropower reservoir in the Scottish Highlands, we use the European Centre for Medium-range Weather Forecasting (ECMWF) extended-range forecast to create probabilistic inflow forecasts for the reservoir. Inflow forecasts are derived by training a linear regression model for the observed inflow onto the NWP precipitation, and subsequently applying post-processing techniques from Ensemble Model Output Statistics.</p><p>We show that the inflow forecasts hold fair skill relative to climatology up to six weeks ahead. Average inflow forecasts for the period 1-35 days ahead hold good skill relative to climatology, and are comparably skilful to an average inflow forecast for the period 8-14 days ahead. Forecasts are more skilful in winter than summer, which is consistent with physical teleconnections from the tropics that operate on sub-seasonal timescales.</p><p>We further apply a stylised cost model that demonstrates the potential value of these forecasts through improved water management. The stylised cost model indicates that the sub-seasonal probabilistic inflow forecast are sufficiently reliable to improve decision making and deliver added value across all forecast horizons up to six weeks ahead, relative to climatological or deterministic forecasts.</p>