Flood Frequency Analysis in an ephemeral river in Spain: inference from instrumental and historical data

The aim of this study is to evaluate the probability of flood discharges for the Almanzora River (SE Iberian Peninsula) which has a short instrumental series but information on historical events. The maximum peak discharge of the instrumental records that affected the Almanzora River occurred in 1973 and was over 5.000 m³s^-1. This extreme event can be considered the most important of the 20th century, but not of history. The main questions we aim to answer are: what is the 100-yr flood for the Almanzora River if we ignore or account for historical events? And, what is the return period of the extreme 1973 flood in the area? We reconstruct the flow rates of the maximum historical floods using the descriptions retrieved from four historical archives within the Almanzora catchment since 1500. To perform the frequency analysis, we establish several thresholds: a) we assume that in the period 1500 -- 1850, no other flood exceeded the perception threshold of 3600 m³s^-1, apart those reconstructed; b) during the period 1850 -- 1962, no other flood exceeded the perception threshold of 1300 m³s^-1, apart those reconstructed; and c) in the instrumental period, 1963 -- 2016, we just consider as exactly known maximum annual discharges higher than 30 m³s^-1. Bayesian inference is applied to fit a GEV distribution and calculate the return periods of flood discharges in the Almanzora watershed. The results show that the Q100 is 3560 m³s^-1, with 95% credible bounds ranging from 2700 to 5800 m³s^-1. There were at least two flood discharges (much) higher than Q100 from 1500 to 1900, in 1580 and in 1879. In both cases, the descriptions from historical sources support this assumption. Also, we estimate the return period of the 1973 flood as 250 years (with 95% credible bounds from 100 to more than 1000 years). Comparing the results using or ignoring historical floods we obtain that when the analysis is done using just instrumental data, the return period is rather underestimated: Q50 is estimated as 661 m³s^-1, while with historical data is over 2000 m³s^-1. FFA with historical data gives us better knowledge about the possible hazard in the area, and future river management should consider these new results.