A Non-Parametric Mann-Kendall and Sen’s Slope Estimate as a Method for Detecting Trend Within Hydro-Meteorological Time Series: a Review

Abstract

Climate and water resources are interconnected in a complex way such that a change in any one induces a change in another. Trend analysis is usually employed to assess and understand the long term pattern of climatic and hydrologic (hydro-meteorological) time series data in order to assess its impact on the environment, particularly water resources. Parametric and non-parametric statistical methods were employed at various times for trend tests depending on the nature of the data at hand. Even though the parametric method was observed to be more robust in making a decisive conclusion, there are certain conditions that need to be met by the data and it was observed that Hydro-meteorological data does not meet most of the conditions. As such, non-parametric procedures for detecting trends were found to be suitable for hydro-meteorological time series. This paper found the rank based Mann-Kendall as one of the most commonly method employed in detecting the trend of hydro-meteorological and Seasonal Kendall Slope (Sen’s slope) for detecting the magnitude of the trend. It was observed that hydrometeorological data are sometimes serially dependent therefore the problem of serial correlation and seasonality will make the application of Mann-Kendall test to have limited applicability, hence, application of pre-whitening procedure is recommended before subjecting the data to the test. But Monte Carlo Simulation Investigation reveals that effect of serial correlation is dependent upon sample size and trend magnitude, when the sample size and trend magnitude are large, the serial correlation will no longer affect Mann-Kendall test. It is concluded that Mann-Kendall and Sen’s Slope Estimate method to be a suitable method of assessing trends within hydro-meteorological time series and authors were
advised to make the sample size hydro-meteorological time series to be analyzed to be large enough to take care of the effect of serial correlation.