Seasonal Precipitation Forecast Using an Ensemble of Artificial Neural Networks and Climate Oscillation Indices. A Case Study of Ceará, northeastern Brazil.

This research assesses the deterministic and probabilistic skill of an Artificial Neural Networks ensemble (EANN) for a 1-month-lead precipitation forecast. The EANN employs low-frequency climate oscillation indices to predict precipitation in the Brazilian state of Ceará, a key region for climate forecasting studies due to its high seasonal predictability. Additionally, a combination of the EANN and dynamical models into a hybrid multi-model ensemble (MME) is proposed. The EANN's forecasting ability is compared to a Multiple Linear Regression, a Multinomial Logistic Regression and North American Multi-Model Ensemble (NMME) models through leave-one-out cross-validation based on 40 years of data. A spatial comparison showed that the EANN was among the models with the highest deterministic and probabilistic accuracy, except in the southern region of the state. Moreover, an analysis of the area-aggregated reliability and sharpness diagrams showed that the EANN is better calibrated than the individual dynamical models and has better resolution than traditional statistical models for above-normal (AN) and below-normal (BN) categories. Both statistical and dynamical models depict a bad-calibrated NN category. It is also shown that combining the EANN and dynamical models improves forecast system reliability compared to an MME based only on NMME models.