Articles | Volume 14, issue 3
https://doi.org/10.5194/hess-14-407-2010
https://doi.org/10.5194/hess-14-407-2010
05 Mar 2010
 | 05 Mar 2010

Flood trends and variability in the Mekong river

J. M. Delgado, H. Apel, and B. Merz

Abstract. Annual maximum discharge is analyzed in the Mekong river in Southeast Asia with regard to trends in average flood and trends in variability during the 20th century. Data from four gauging stations downstream of Vientiane, Laos, were used, covering two distinct hydrological regions within the Mekong basin. These time series span through over 70 years and are the longest daily discharge time series available in the region. The methods used, Mann Kendal test (MK), ordinary least squares with resampling (OLS) and non-stationary generalized extreme value function (NSGEV), are first tested in a Monte Carlo experiment, in order to evaluate their detection power in presence of changing variance in the time series. The time series are generated using the generalized extreme value function with varying scale and location parameter. NSGEV outperforms MK and OLS, both because it resulted in less type II errors, but also because it allows for a more complete description of the trends, allowing to separate trends in average and in variability.

Results from MK, OLS and NSGEV agreed on trends in average flood behaviour. However, the introduction of a time-varying scale parameter in the NSGEV allowed to isolate flood variability from the trend in average flood and to have a more complete view of the changes. Overall, results showed an increasing likelihood of extreme floods during the last half of the century, although the probability of an average flood decreased during the same period. A period of enhanced variance in the last quarter of the 20th century, estimated with the wavelet power spectrum as a function of time, was identified, which confirmed the results of the NSGEV.

We conclude that the absence of detected positive trends in the hydrological time series was a methodological misconception due to over-simplistic models.