Mann-Kendall test applied to hydrological data – Performance of TFPW and CV2 filters on trend analysis

This study brings an original comparison related to the performance of two filters on trend analysis regarding hydrological time series. We applied the Mann-Kendall test for trend analysis, a non-parametric test widely used in hydrological studies, and Sen’s slope in order to extract the trend magnitude. The presence of autocorrelation tends to impact on trend interpretation erroneously. As most of water resources data presents serial correlation, the use of filters is essential to achieve an accurate analysis regarding temporal variation of the dataset. The filters trend free pre-whitening (TFPW) and variance correction approach (CV2) were applied on monthly time series of precipitation, streamflow, storage and evapotranspiration, from 2002 to 2014, plus eighty synthetic time series. The comparison of the filters performances showed the TFPW filter as much superior, reducing the autocorrelation by at least 71.1%. While the CV2 filter, despite strongly reducing the variance, did not impact the serial correlation (in fact, reduced less than 1% in almost half of the performed simulations). The main difference was related to the precipitation data, from which CV2 suggested a negative trend, while TFPW, besides drastically reducing autocorrelation, showed that the time series does not have a statistically significant trend.