In addition to the SCE-UA (Shuffled Complex Evolution – University of Arizona), used for the optimization of the hydrological parameters in RS MINERVE, two new algorithms have been implemented: the “Uniform Adaptive Monte Carlo” and the “Coupled Latin Hypercube and Rosenbrock”
The Uniform Adaptive Monte Carlo (UAMC) algorithm is based on the Monte Carlo experiments that rely on repeated random sampling but has been modified in order to iteratively adjust the solution space.
The algorithm randomly launches a collection of simulations (block) and finds the better results in the solution space. Afterwards, the solution space is adjusted and a new group of simulations starts. The process is repeated until the optimization converges to the best set of parameters.
The Coupled Latin Hypercube and the Rosenbrock algorithm, generates a powerful tool for optimization of complex problems. This combined algorithm can discretize a wide domain and then narrow your search to smaller sectors. Scanning of the space of possible solutions is performed by the Latin Hypercube. This algorithm allows pseudo-statistical sampling conditioned by the previous calculated solutions. The Latin hypercube is an evolution of the Monte Carlo method, with more homogeneous samples achieved with fewer samples. An important advantage of this method is that the dimension of the problem is defined by the division of the latin hypercube and not by the number of parameters.
The best results from samples become the starting points required for Rosenbrock algorithm. The advantage of this subroutine calculation lies in the speed to obtain near optimal values. This algorithm is based on a gradient search, adjusting axis changes based on the direction of maximum enhancement, thus reducing the number of evaluations of the objective function.
More details in the technical manual.