Causal Reasoning for the Analysis of Rivers Runoff Temporal Behavior

The accurate knowledge about the influence of time in the behavior of rivers systems is crucial for a proper river basin water management. Traditional techniques such as correlograms or ARMA models have been widespread used over the last decades providing the analyzer with an average behaviour of temporal influence of hydrological series. In the last decade, the development of techniques, under the discipline of artificial intelligent, have increased the range of available analytical tools. On the other hand, hydrological processes have a very strong random nature and they are driven by its high uncertainty and variability. Consequently, it is necessary to build tools, able to incorporate these peculiarities in their analytical functioning. Causal Reasoning through Bayesian Networks (BNs) allows processing and analysing hydrological series, incorporating and assessing all their variability. Causality driven by BayesA ' theorem is used here to dynamically identify, characterize and quantify the influence of time (dependence) for each time step in annual run-off series in five Spanish River basins. Therefore, BNs arise as a powerful tool for getting a deeper understanding on the knowledge of temporal behaviour of hydrological series because this analysis is dynamic and implemented specifically for temporal iterations (decision variables). Implications and applications of this research are largely aimed to improve and optimize the design and dimensioning of hydraulic infrastructures, as well as reducing the risk of negative impacts produced by extreme events such as several droughts or floods, among others.