Informing Water Management by Direct Use of Snow Information as Surrogate of Medium-to-Long Range Streamflow Forecast

Monday, 15 December 2014
SImona Denaro, Politecnico di Milano, Milano, 20133, Italy, Matteo Giuliani, Politecnico di Milano, Milano, Italy and Andrea Castelletti, Polytechnic University of Milan, Milan, Italy
Medium-to-long range streamflow forecast provide a key assistance in anticipating hydro- climatic adverse events and prompting effective adaptation measures. For instance, accurate medium-long range streamflow forecasts have a great potential to improve water reservoir operation by enabling more efficient allocation of water volumes in time (e.g. via hedging). Unfortunately, these forecasts often lacks reliability and accuracy, especially when low-frequency climate forcing (e.g. ENSO) is not intense enough to improve the forecast lead time (e.g. in Europe), and might be computationally very demanding,

In this work, we explore the direct use of both rough snow data (e.g. snow depth) and snow water equivalent estimates as surrogate of medium-to-long range streamflow forecast to inform the operation of a regulated lake. The underlying idea is that snow data contains key information on current and future water availability throughout the snow melting season that might significantly improve the operation’s anticipation potential. We adopt a three step methodology: First, we compute the upper bound of the system performance by assuming perfect foresight and we assess the value of additional information as the difference between this ideal solution and current operation. Using input variable selection, we then select the most relevant snow information to explain the release trajectory associated to the upper bound operating policy. Finally, we derive the optimal policy conditioned upon the selected variables by Multi-Objecting Evolutionary Direct Policy Search.
The methodology is demonstrated on the snow-dominated Lake Como river basin, in the Italian Alps. Lake Como is a regulated lake primarily used to supply water to a large cultivated area and snowmelt from May-July is the most important contribution to the creation of the seasonal storage.

Results show that using raw data or simple SWE estimates can largely improve anticipation capability in the daily operation of the lake thus increasing the reservoir hedging potential and the overall system performance over irrigation deficit.