Single Member Selection in Ensemble Forecasting

Abstract

Ensemble forecasting is a technique devised to palliate sensitivity to initial conditions in nonlinear dynamical systems. The basic idea to avoid this sensitivity is to run the model many times under several slightly-different initial conditions, merging the resulting forecast in a combined product. We argue that this blending procedure is unphysical, and that a single trajectory should be chosen instead. We illustrate our case with a climate model. While most of the current climate simulations use the ensemble average technique as merging procedure, this paper shows that this choice presents several drawbacks, including a serious underestimation of future climate extremes. It is also shown that a sensible choice of a single estimate from the ensemble solves this problem, partly overcoming the inherent sensitivity to initial conditions of those non-linear systems with a large number of degrees of freedom.