Open Collections

Swift-LaPointe, Taylor

University of British Columbia

Forecasting of streamflow entering dam reservoirs is important for management of hydroelectricity operations, with economic and environmental impacts. On seasonal timescales, some prediction skill of variables that influence streamflow is derived from climate modes of variability, like the El Niño Southern Oscillation (ENSO) and Pacific Decadal Oscillation (PDO), among other factors. In this study, it is investigated how the phases of these climate modes affect inflow into the Kinbasket Lake Reservoir and Mica Dam in southeast British Columbia, a snowmelt-dominated catchment. El Niño and PDO positive phases produce greater streamflow than La Niña and PDO negative phases from April to June, with the opposite true from June to September, although the differences are small. It is investigated whether a forecast using meteorological conditions, affected by ENSO and PDO modes, to predict streamflow in this catchment achieves skill on seasonal timescales. The hybrid statistical-dynamical forecast of cumulative January to September seasonal streamflow at nine months lead time uses dynamical ECMWF SEAS5 seasonal meteorological hindcasts (retrospective "forecasts") as input into a Long Short-Term Memory (LSTM) neural network. A monthly LSTM model using 12 months of meteorological forcings outperforms a daily LSTM model using 365 days of forcings at capturing the interannual variability in seasonal streamflow volumes when forced with reanalysis (ERA5) meteorological data. Skill in the meteorological inputs is required in at least the first three months of the forecast year, reflecting that the main source of streamflow prediction skill is derived from snowpack build up prior to and at the beginning of the forecast year, rather than ENSO and PDO indices. The hybrid streamflow forecast underestimates seasonal volumes in most years due to biases present in the SEAS5 hindcasts. Three bias correction methods are investigated, and linearly shifting the mean of the SEAS5 hindcasts to that of reanalysis gives the largest improvement in skill of the streamflow forecast. This study demonstrates that it is possible to use a hybrid statistical-dynamical forecast to predict seasonal volumes in a mountainous catchment, however the skill in predicting interannual variability is largely limited by the accuracy of prediction of meteorological forcings on seasonal timescales.

Text

2024-10-15

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/89408

Atmospheric Science

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/