Estimation approach for T-year hydrological events using non-stationary data

As stated in The IPCC Sixth Assessment Report, heavy rainfall events of unprecedented scale have occurred in recent years increasingly in terms of both frequency and intensity because of global climate change. As a matter of course, greater attention must be devoted to flooding caused by heavier-than-ever rainfall events. This flooding includes both levee breach and inland water rise effects.

In Japan, T-year hydrological events, such as 100-year-rainfall events with a return period of 100 years as estimated from frequency analysis, have been used conventionally as targets of river improvement plans. In fact, "Guidelines for Small and Medium-Sized River Planning” have been consulted when hydrological quantities are estimated. Nevertheless, the flow chart in the guideline drawn by the MLIT (*) has been discounted completely in work by Kuzuha et al. (2021, 2022a,b,c). In fact, it is most inappropriate to use the SLSC as the criterion for validating stochastic models; it is also inappropriate for usage of the Jack-knife or bootstrap method. Mizuki and Kuzuha (2023) present related supporting details.

As described in this paper, we intend to present other issues which must be urgently resolved: The fact that the precipitation population has not been stationary. It must be regarded as non-stationary because of global climate change.

Explanations of frequency analysis based on the non-stationarity of the precipitation population have been presented in the literature by Hayashi et al. (2015) and by Shimizu et al. (2018). We have considered different approaches than theirs. Ours predict future T-year hydrological events under the condition of non-stationary precipitation population, as presented below. In other words, those approaches can be adapted to recent quite heavier rainfall data.

• We use d4PDF data (2015) data. In fact, d4PDF data were calculated using climate simulations of 50 ensemble members. Each ensemble member has climate data obtained during 1951–2010: we can use annual maximum rainfall of 3,000 years. We specifically examined the area around Kumano city, Mie prefecture and analyzed the annual maximum around Kumano.
• First, we calculated the annual maximum 1-hour precipitation at Kumano described above.
• For example, there are 50 annual maximum 1-hour precipitation events in 1951, because there are 50 ensemble members. Therefore, we can estimate 100-year rainfall in 1951 using 50 data and the Gumbel distribution. We can estimate time-variational 100-year rainfall during 1951 and 2010.
• The blue line in the figure shows the time variational 100-year rainfall between 1951 and 2010.
• The orange line represents future 100-year rainfall calculated using the triple exponential smoothing method.

At the presentation, we intend to show other approaches which can be useful to predict future 100-year precipitation.

 

* MLIT: The Ministry of Land, Infrastructure, Transport and Tourism, Japan