Optimized echo state networks for drought modeling based on satellite data

Remotely sensed data obtained through satellite imaging is a useful tool for modeling environmental phenomena such as drought. In this manuscript, we apply optimized echo state networks to model and predict drought severity based on satellite images. To this end, a model is constructed in which the satellite-based vegetation index is fed as an input and drought severity index is obtained as output. We use a Kronecker-based approach to reduce the number of parameters of echo state networks to be optimized (i.e., the internal weights of reservoir). A number of evolutionary algorithms are used to optimize the parameters, of Differential Evolution results in the best performance as compared to genetic algorithms and simulated annealing. The proposed model also outperforms neural network models including multilayer perceptrons, radial basis function networks and support vector machines. © 2015 ICIC International.