Using the Firefly optimization method to weight an ensemble of rainfall forecasts from the Brazilian developments on the Regional Atmospheric Modeling System (BRAMS)

• Published in: Advances in geosciences Vol. 35; pp. 123 - 136
• Main Authors: dos Santos, A. F., Freitas, S. R., de Mattos, J. G. Z., de Campos Velho, H. F., Gan, M. A., da Luz, E. F. P., Grell, G. A.
• Format: Journal Article
• Language: English
• Published: Katlenburg-Lindau Copernicus GmbH 17.09.2013; Copernicus Publications
• Subjects: South America
• ISSN: 1680-7359; 1680-7340; 1680-7359
• DOI: 10.5194/adgeo-35-123-2013

In this paper we consider an optimization problem applying the metaheuristic Firefly algorithm (FY) to weight an ensemble of rainfall forecasts from daily precipitation simulations with the Brazilian developments on the Regional Atmospheric Modeling System (BRAMS) over South America during January 2006. The method is addressed as a parameter estimation problem to weight the ensemble of precipitation forecasts carried out using different options of the convective parameterization scheme. Ensemble simulations were performed using different choices of closures, representing different formulations of dynamic control (the modulation of convection by the environment) in a deep convection scheme. The optimization problem is solved as an inverse problem of parameter estimation. The application and validation of the methodology is carried out using daily precipitation fields, defined over South America and obtained by merging remote sensing estimations with rain gauge observations. The quadratic difference between the model and observed data was used as the objective function to determine the best combination of the ensemble members to reproduce the observations. To reduce the model rainfall biases, the set of weights determined by the algorithm is used to weight members of an ensemble of model simulations in order to compute a new precipitation field that represents the observed precipitation as closely as possible. The validation of the methodology is carried out using classical statistical scores. The algorithm has produced the best combination of the weights, resulting in a new precipitation field closest to the observations.