Enhancing the applicability of Trend-free Pre-whitening method in trend detection of autocorrelated data

Abstract:

Detecting trends in hydrometerological data via the commonly used Mann-Kendall test is challenged by the presence of autocorrelation component in the series. The positive autocorrelation inflates the probability of detecting trends when actually none exists. Trend-free Pre-whitening is one useful technique to eliminate this adverse effect of autocorrelation on the test. However, its applicability is offset by the sampling error of estimating linear trend slope in the series. Without removing the trend component, the Pre-whitening method reduces the power of the test to detect a significant trend. To remedy these weaknesses, a modified Trend-free Pre-whitening method is proposed via correcting the standard deviation of the detrended series and improving the estimation accuracy of linear trend slope. The modified method is further compared with the original one and the Pre-whitening method in simulation and application studies. The results validate the merit of the modified method. It not only reduces the probability of falsely reporting an inexistent trend but also preserves the power of detecting a strong trend. The modified method is recommended when the estimated linear trend slope is biased and unreliable.