Bat optimisation neural networks for rainfall forecasting: case study for Kuching city

Abstract This paper presents a novel metaheuristic artificial neural network (ANN) model, named as Bat optimisation neural network (BatNN), for spatial downscaling of long-term precipitation. This novel BatNN was developed due to the inefficiency of traditional ANNs in spatial downscaling of large-scale outputs from climate models. Input data are predictors from three climate models including HadCM3, ECHAM5 and HadGEM3-RA combined with observed precipitation collected from Kuching airport rainfall station. The output is the forecasted precipitation. Data from 1961 to 1990 were used for model training, while data from 1991 to 2010 were used for validation. Square root of correlation of determination (r), root mean square error (RMSE), mean absolute error (MAE) and Nash–Sutcliffe coefficient (E) are used to evaluate the models' performance. Results showed that through global and local searches, BatNN is able to avoid local optima trappings. The average r, RMSE, MAE and E for three climate models were yielded to 0.96, 1.69, 1.40 and 0.84, respectively. This reveals that BatNN is able to optimise and forecast long-term precipitation accurately.