Non-stationary flood frequency analysis (NFFA) plays an important role in addressing the issue of the stationary assumption (independent and identically distributed flood series) that is no longer valid in infrastructure-designed methods. This confirms the necessity of developing new statistical models in order to identify the change of probability functions over time and obtain a consistent flood estimation method in NFFA. The method of Trimmed L-moments (TL-moments) with time covariate is confronted with the L-moment method for the stationary and non-stationary generalized extreme value (GEV) models. The aims of the study are to investigate the behavior of the proposed TL-moments method in the presence of NFFA and applying the method along with GEV distribution. Comparisons of the methods are made by Monte Carlo simulations and bootstrap-based method. The simulation study showed the better performance of most levels of TL-moments method, which is TL(η,0), (η = 2, 3, 4) than the L-moment method for all models (GEV1, GEV2, and GEV3). The TL-moment method provides more efficient quantile estimates than other methods in flood quantiles estimated at higher return periods. Thus, the TL-moments method can produce better estimation results since the L-moment eliminates lowest value and gives more weight to the largest value which provides important information.

非平稳洪水频率分析（NFFA）在解决固定假设（独立且分布均匀的洪水序列）方面起着重要作用，该假设在基础设施设计的方法中不再有效。这证实了开发新的统计模型以识别概率函数随时间变化并在NFFA中获得一致的洪水估算方法的必要性。带有时间协变量的Trimmed L-Moments（TL-Moments）方法面对固定和非平稳广义极值（GEV）模型的L-Moment方法。本研究的目的是研究存在NFFA时提出的TL矩方法的行为，并将其与GEV分布一起应用。通过蒙特卡洛模拟和基于引导的方法对方法进行了比较。仿真研究表明，大多数水平的TL矩方法，即TL（η，0），（η = 2，3，4）比所有模型（GEV1，GEV2和GEV3）的L矩方法都好。TL矩方法提供了比其他方法更高的分位数估计效率，而洪水分位数却比其他方法具有更高的回报期。因此，由于L矩消除了最小值，并赋予了更大的权重，因此TL矩方法可以产生更好的估计结果，从而提供了重要的信息。