Machine learning for downscaling: the use of parallel multiple populations in genetic programming