Abstract

When the climate system undergoes changing external forcing (e.g. from increases in greenhouse gas concentrations), there are two inherent limits on the gain in skill of decadal climate predictions that can be attained from initializing with the observed ocean state. One is the classical initial-... Show moreWhen the climate system undergoes changing external forcing (e.g. from increases in greenhouse gas concentrations), there are two inherent limits on the gain in skill of decadal climate predictions that can be attained from initializing with the observed ocean state. One is the classical initial-value predictability limit that is a consequence of the system being chaotic, and the other corresponds to the forecast range at which information from the initial conditions is overcome by the forced response. In the first part of this presentation these two limits are quantified for Community Climate System Model, version 3 (CCSM3) with several 40-member climate change scenario experiments. Predictability of the upper-300m ocean temperature, on basin and global scales, is estimated by relative entropy from information theory. Overall, information from the ocean initial condition exceeds that from the forced response for about seven years. After about a decade the classical initial-value predictability limit is reached. In the second part we quantify the classical initial-value predictability limit for several other models, including GFDL's CM2.1, the Kiel Climate Model, MIROC3.2, and CCSM4. We estimate this limit by examining long control integrations of these models. Significant model-to-model variations in predictability are found. In the third part we address the question of whether prominent intrinsic modes of the coupled system may have predictability beyond that found for basin-wide measures. In general we find the PDO does not have above average predictability while AMOC predictability lasts a few years longer than basin-wide predictability. However, in CCSM3 and 4, AMOC's predictability is associated with only weak atmospheric signals. Show less