Abstract. Wind resource assessments predict future production levels from historical data. To characterize how year-to-year variability in past wind speeds affects the certainty of future predictions, we analyze 62-year wind speed records of 60 weather stations in Canada, and compare the actual levels of each station's final 20 years to "predictions" made from previous periods of varying duration. We estimate both median (P50) and 10 % quantile (P90) levels using historical means and standard deviations, validating estimator performance on statistically-independent "control" sequences made by randomly permuting the 62 annual values of each station's record. Errors of estimates made from the control sequences always decline with record length; the central half of the stations’ exceedances falls within ranges of 44–55 % (P50) and 85–95 % (P90) for 42-year estimates. For the actual chronological records, on the other hand, error is lowest when estimates were made from short records (4–5 years) and increases with length after 15 years; for 42-year estimates the corresponding ranges are 0–45 % (P50) and 36–100 % (P90). The strong biases reflect a nearly nationwide downward trend in recorded wind speeds, but even a near-zero-trend subset of 30 stations exhibits interquartile ranges of 24–73 % (P50) and 80–100 % (P90), both twice as large as expected. These findings show that serial correlation in wind speeds can persist across decades, and, if ignored, results in substantial overconfidence in estimated resource levels.;
