Detection of long-term change in hydroelectric reservoir inflows: Bridging theory and practise

Abstract

We studied long-term monotonic trend in inflow volume to 20 reservoirs, plus a measure of system-wide hydroelectric power potentially available for generation, in British Columbia, Canada. The primary analysis involved application of a linear method with explicit signal-to-noise ratio (S:N) estimation to monthly and annual mean inflows over the period of record common to all basins, providing a broad and convenient overview of changes across the hydroelectric system. Additional analyses then investigated the sensitivity of the results to methodological choices, focussing on annual inflow volume for eight indicator basins and system power. In these auxiliary analyses, alternative data selection and processing schemes were explored: full period of record for each indicator basin; residual after Pacific Decadal Oscillation signal removal; and data from streamgauges on catchments within or near each indicator basin. Alternative analysis techniques were also applied: nonparametric Spearman rank correlation with adjustment for serial dependence; change point detection using Mann–Whitney sliding window analysis; and visual assessment of data low-pass filtered using polynomial fits of various orders. Finally, we studied changes in year-to-year water supply variability, employing a split-sample approach implemented using a standard F-test; a nonparametric Ansari–Bradley test; and a novel Monte Carlo-based test on the Shannon entropy, a measure of hydroclimatic information content. The results provided no substantial evidence for decreased annual water supply and pointed to some increases; seasonal inflow pattern shifts were also detected for many basins, in particular a wintertime flow increase. The split-sample tests yielded no evidence for shifts in the volatility and predictability of annual inflow. One of the more intriguing and perhaps broadly important findings was that methodological choices ultimately made little difference to final outcomes.