Assessment of the predictability of inflow to reservoirs through Bayesian causality

Abstract

This research assesses the predictive capacity of Bayesian causality through causal reasoning (CR), which has been successfully applied to the study of reservoir inflows. We combined autoregressive development with a causal modelling approach through a “proof of concept” that assesses the predictive capacity of the approach. The analytical power of CR revealed the logical temporal structure that defines the general behaviour of inflows, which was latent in the historical records. This allowed identifying/quantifying, through a dependence matrix, two temporal runoff fractions, one due to time and the other not. Finally, a predictive model for each temporal fraction was implemented, evaluating its forecasting skills through mean absolute error and root mean square error. This was applied to the reservoirs that supply water to the city of Ávila (Spain), whose watersheds present strong independent temporal behaviour. These results open new possibilities for developing predictive hydrological models within a CR modelling framework.