Flood magnitudes in the Tagus River (Iberian Peninsula) and its stochastic relationship with daily North Atlantic Oscillation since mid-19th Century

This paper presents a new methodological approach in the analysis of the influence of the North Atlantic circulation on the flood magnitude of the Tagus River, the largest Atlantic draining river of the Iberian Peninsula. This methodology is based on Correspondence Analysis viewed as a qualitative regression tool. A daily scale database (sea level atmospheric pressure, river discharge, rainfall) was used for this study. The selected streamflow station, Vila Velha de Rodão (Portugal, near the Spanish border), holds the longest continuous daily river discharge register of the Iberian Peninsula (over 160 years, since 1852). The annual maximum flood magnitudes were classified into seven categories according to their specific recurrence intervals (T). The qualitative regression approach was used to relate annual peak flood occurrence with the North Atlantic Oscillation (NAO) index mode (positive or negative) registered, during the preceding 40 days (divided in 8 successive 5-day periods). Large floods (categories 1–2 of T > 50 years and category 3, T 25–50 years) were found to be associated with a very high frequency of the negative NAO mode during the initial 20–25 days (within a total 40 days period length) before de flood peak. The lack of significant association of these flood categories with a predominant NAO mode during the immediately preceding 15 days, prior to the flood, suggest a time lag of 15 days before the peak. Minor flooding (category 7, T < 2 years) show a high degree of association with the presence of a positive NAO mode during the previous 20–25 days. The analysis of flood response under natural and dam-regulated regimes (before and after the construction of dams ca 1960) revealed no changes in the behaviour of major floods (responding to a period of 25 days with a dominant negative NAO index mode). Moderate flooding of category 4 (T: 10–25 years), that during the pre-dam construction period were linked to the existence of an initial 5–15 days of negative NAO mode, were not documented during the post-dam period, probably due to flood peak discharge attenuation by reservoirs. The clear climatic control (rainfall accumulation and number of persisting NAO negative mode days) in the onset of flood category 3 (T: 25–50 years) was also statistically blurred during the post-dam period, due to the effective role of dam operation in flood management during wet winters. This dam management effect does not seem to mask the meteorologically related largest flood magnitude events (T > 50 years), indicating that catastrophic events will likely occur under the NAO negative daily and seasonal patterns defined in this work. The robustness of the model was assessed by the phenomenological analysis of misclassified Flood/NAO events. Results also indicate that flood management practises may benefit from improved climate scenarios and forecasting of the NAO index.