Separating Forced and Internal Variability in North Atlantic Sea Surface Temperature

Quantifying the relative contributions of external forcing and internal variability to North Atlantic Sea Surface Temperature (NASST) has important implications for attributing and predicting climate changes around the North Atlantic basin. Many previous methods have approached this problem by estimating the externally forced signal directly, making assumptions about forced variability for which there is no consensus. In this work, the separation of variability is approached in a fundamentally different way that does not specify the forced response's temporal evolution. We propose a dynamical adjustment method in which the internal, spatially uniform component of NASST is predicted based on patterns of NASST that are orthogonal to the spatially uniform pattern. This dynamical adjustment method is trained and validated in preindustrial simulations, where only internal variability is present. In preindustrial simulations, the dynamical adjustment demonstrates skill in reconstructing the NASST basin mean variability. Applying the preindustrial-trained dynamical adjustment to historical simulations demonstrates skill in a majority of climate models, although the skill is reduced relative to preindustrial simulations because external variability partly contaminates the predictors. The skill of dynamical adjustment is compared to several other methods which directly estimate the externally forced signal. We find that dynamical adjustment performs similarly to these comparative methods, despite the fundamentally different prediction method. However, methods based on different principles yield considerably different estimates of external and internal variability. In efforts to contend with the presence of external variability in predictor time series in historically forced simulations, several modifications to the dynamical adjustment methodology are investigated. Most directly, low-frequency filters are applied to the predictor time series to estimate and remove any external variability. We find that the filters applied in this study do not skillfully identify the external variability, and the application of dynamical adjustment using these filtered predictors does not demonstrate a consistent improvement. Modifying the dynamical adjustment by training in historical simulations or including the NASST basin mean as a predictor is also explored. While training using historical simulations does not produce a more skillful dynamical adjustment on its own, we find some variations of dynamical adjustment that include the NASST basin mean as a predictor can produce skill equal to or greater than the original, preindustrial-trained dynamical adjustment. Several variations of dynamical adjustment are applied to observational data from ERSSTv5. These estimates of internal variability are fairly consistent amongst variations, but are distinguished in both amplitude and phase with some residual estimates produced by the comparative methods that directly estimate external variability. The range of estimated external and internal variability in ERSSTv5 reinforce the sensitivity to assumptions that underly different methods.

Author

• Nedza, Douglas

• Climate Change
• Climate Model
• Machine Learning
• Oceanography
• Sea Surface Temperature
• Variability

Identifier

hdl:1920/14088

OAI identifier oai:identifier

oai:MARS:1920/14088