Modelling the spring Douro river flow using SST

We examine the predictability of Douro streamflow anomalies during spring using previous seasonal sea surface temperature (SST). In a first part, based on running correlation analysis, we identified two regions where previous winter SST anomalies are stably correlated with the spring streamflow anomalies for the whole period 1956-2006 under study. SST1 and SST2 indices are defined by averaging the normalized SST anomalies for these regions sited in the central part of the North Atlantic Ocean (45W-25W; 38N-42N) and in the south-western Atlantic Ocean (35W-25W; 15S-10S), respectively. These indices are used as explanatory variables for the spring Douro river flow in a model based on linear regression. This SST_model is able to explain the 50% of the total variance in spring Douro streamflow. Additionally, the SST_model presents useful forecasts skill, performing 38% better than climatology and 62% better than persistence, and it's able to modelling the phase of the streamflow with a percentage of agreement of 80%.

In a second part, an additional study is carried out over the residual time series (residual = flow – SST_model) in order to improve the modelling of the spring Douro river flow. Singular spectral analysis applied over the residual shows three significant quasi-oscillatory modes with periods around 2.4, 5 and 3 years. SSA_residual_filter, computed as the sum of the reconstructed components of these oscillatory modes, shows a strong autocorrelation pattern. This makes the SSA_residual_filter more predictable compared to unfiltered one. Additionally, significant correlation values are found between the quasi-oscillatory mode with period around 3 years and the previous winter SST in the region of El NIÑO3, and the quasi-oscillatory mode with period around 5 years and the previous spring SST in the region of El NIÑO3.4. Finally, an Auto-Regressive-Moving-Average (ARMA) model was fitted to the SSA_residual_filter series. The use of this interannual linear information considerably improves the skill of the modelling (improvements against climatology and persistence of 73% and 83%, respectively) compared to the SST_model. We conclude that the predictability for the spring Douro streamflow can be divided in two parts: the seasonal predictability associated with the Atlantic SST during the previous winter (50%) and the linear interannual predictability (17%), which is considerable lower and shows some association with El NIÑO.