Merging statistical and dynamical forecasts for hurricane prediction

A particular situation that arises when combining statistical and dynamical forecasts to obtain a consensus is the disparity in the size of the joint sample (forecast, observation) available for weight optimization. Some statistical methods to forecast hurricane characteristics have been created and used for many years, whereas operational dynamical models are constantly upgrading. There are several approaches to alleviate this disparity ranging from ignoring the oldest statistical forecasts to match the size of the joint sample with that of the dynamical forecasts, to producing a long dataset of retrospective dynamical forecasts to match the long sample size available in the statistical forecast methods. A constrained least squares procedure is proposed to obtain the weights as another alternative. A comparison of this approach with the simple multi-model average and other methods will be presented.