The use of hybrid methods for change points and trends detection in rainfall series of northern Algeria

Abstract

The aim of this research is to assess relatively new hybrid methods for changes points and trends detection on rainfall series: Dynamic Programming Bayesian Change Point Approach (BA), Şen's innovative trend method (ITM) and its double (D-ITM) and triple (T-ITM) version using the multi-scale analysis of the discrete wavelet transform (DWT) as a coupling method. Three representatives rainfall stations of northern Algeria were analysed at annual scale during the period 1920–2011. Moreover, correlation and spectral analysis (CSA) was applied for periodicity analysis. The CSA indicates the dominance of interannual to multidecadal rainfall periodicity fluctuations (2-years, 5-years and 20-years) characterising long term structured processes. Moreover, an abrupt downward trend with significant probability was detected from the 1970s with a relatively wet period between the periods 1950–1970 and 2001–2011. The latter is observed in particular in the central and eastern stations, well-explained by the BA-DWT. The results showed that the comparison results from different modelling approaches found that the hybrid models (BA-DWT, ITM-DWT, D-ITM-DWT, T-ITM-DWT) often perform better than the conventional approach (BA, ITM, D-ITM, T-ITM), where the computation time is very reasonable. The analysis revealed that information stemming from discrete wavelet spectrums significantly increased the accuracy of the methods for detecting hidden change points and trends.