Improved ENSO prediction and teleconnections from reduction in coupled model bias

Abstract:

Dynamical seasonal predictions employ coupled climate models that are initialized with observationally constrained initial conditions. Model bias, which leads to model drift and contributes to initialization shock, has been a persistent obstacle to the efforts of improving seasonal predictions. By applying a prognostic bias reduction method in GFDL's new SPEAR seasonal prediction system, we demonstrated reduced model climatological prediction bias as well as improved anomaly prediction skills in the prediction of ENSO and its teleconnections. We use multiple sets of historical retrospective forecast experiments to analyze the impact of model bias on coupled seasonal predictions. The reduced climatological prediction bias has also been shown to benefit a wide range of subseasonal-to-seasonal prediction applications.