Historical records are an important source of information on extreme and rare floods and fundamental to establish a reliable flood return frequency. The use of long historical records for flood frequency analysis brings in the question of flood stationarity, since climatic and land-use conditions can affect the relevance of past flooding as a predictor of future flooding. In this paper, a detailed 400 yr flood record from the Tagus River in Aranjuez (central Spain) was analysed under stationary and non-stationary flood frequency approaches, to assess their contribution within hazard studies. Historical flood records in Aranjuez were obtained from documents (Proceedings of the City Council, diaries, chronicles, memoirs, etc.), epigraphic marks, and indirect historical sources and reports. The water levels associated with different floods (derived from descriptions or epigraphic marks) were computed into discharge values using a one-dimensional hydraulic model. Secular variations in flood magnitude and frequency, found to respond to climate and environmental drivers, showed a good correlation between high values of historical flood discharges and a negative mode of the North Atlantic Oscillation (NAO) index. Over the systematic gauge record (1913–2008), an abrupt change on flood magnitude was produced in 1957 due to constructions of three major reservoirs in the Tagus headwaters (Bolarque, Entrepeñas and Buendia) controlling 80% of the watershed surface draining to Aranjuez. Two different models were used for the flood frequency analysis: (a) a stationary model estimating statistical distributions incorporating imprecise and categorical data based on maximum likelihood estimators, and (b) a time-varying model based on "generalized additive models for location, scale and shape" (GAMLSS) modelling, which incorporates external covariates related to climate variability (NAO index) and catchment hydrology factors (in this paper a reservoir index; RI). Flood frequency analysis using documentary data (plus gauged records) improved the estimates of the probabilities of rare floods (return intervals of 100 yr and higher). Under non-stationary modelling flood occurrence associated with an exceedance probability of 0.01 (i.e. return period of 100 yr) has changed over the last 500 yr due to decadal and multi-decadal variability of the NAO. Yet, frequency analysis under stationary models was successful in providing an average discharge around which value flood quantiles estimated by non-stationary models fluctuate through time.

A long-term flood record from the Buffels River, the largest ephemeral river of NW South Africa (9250 km²), was reconstructed based on interpretation of palaeoflood, documentary and instrumental rainfall data. Palaeoflood data were obtained at three study reaches, with preserved sedimentary evidence indicating at least 25 large floods during the last 700 yr. Geochronological control for the palaeoflood record was provided by radiocarbon and optically stimulated luminescence (OSL) dating. Annual resolution was obtained since the 19th century using the overlapping documentary and instrumental records. Large floods coincided in the past within three main hydroclimatic settings: (1) periods of regular large flood occurrence (1 large flood/~30 yr) under wetter and cooler prevailing climatic conditions (AD 1600–1800), (2) decreasing occurrence of large floods (1 large flood/~100 yr) during warmer conditions (e.g., AD 1425–1600 and after 1925), and (3) periods of high frequency of large floods (~ 4–5 large floods in 20–30 yr) coinciding with wetter conditions of decadal duration, namely at AD 1390–1425, 1800–1825 and 1915–1925. These decadal-scale periods of the highest flood frequency seem to correspond in time with changes in atmospheric circulation patterns, as inferred when comparing their onset and distribution with temperature proxies in southern Africa.

Long-term average river discharges as well as peak and low flows of different return periods are estimated along the entire river network of Colombia, through the conjoint use of the long-term water balance in the river basins and the framework of statistical scaling, taking the average flow field as the scaling variable. Estimation of the long-term water balance considers the spatial variability of hydrologic fields, in which drainage basins are considered the basic hydrological control volumes for estimation. A systematic effort has been made to estimate the long term average precipitation field combining rain gauge measurements with existing handmade expert maps as an input trend for a universal Kriging interpolation technique. Evaluation of estimates for actual and potential long-term evapotranspiration was implemented using diverse methods. Results were tested using the long term water balance equation against 200 streamflow gauging stations. No method for actual evapotranspiration showed significant...

This paper proposes the estimation of high return period quantiles using upper bounded distribution functions with Systematic and additional Non-Systematic information. The aim of the developed methodology is to reduce the estimation uncertainty of these quantiles, assuming the upper bound parameter of these distribution functions as a statistical estimator of the Probable Maximum Flood (PMF). Three upper bounded distribution functions, firstly used in Hydrology in the 90's (referred to in this work as TDF, LN4 and EV4), were applied at the Jucar River in Spain. Different methods to estimate the upper limit of these distribution functions have been merged with the Maximum Likelihood (ML) method. Results show that it is possible to obtain a statistical estimate of the PMF value and to establish its associated uncertainty. The behaviour for high return period quantiles is different for the three evaluated distributions and, for the case study, the EV4 gave better descriptive results. With enough information, the associated estimation uncertainty for very high return period quantiles is considered acceptable, even for the PMF estimate. From the robustness analysis, the EV4 distribution function appears to be more robust than the GEV and TCEV unbounded distribution functions in a typical Mediterranean river and Non-Systematic information availability scenario. In this scenario and if there is an upper limit, the GEV quantile estimates are clearly unacceptable.