Title: Distinguishing climate signals from climate noise using pattern recognition
Abstract: A primary goal of climate science is to separate the influences of external forcing and internal variability on observed climate changes, as is needed for attribution, for estimating the climate response to future changes in radiative forcing, and for characterizing and understanding internal climate variability. Ensembles of climate model simulations are commonly used for this purpose. However, much of the information gained from running large ensembles is lost in traditional methods of data reduction such as linear trend analysis or large-scale spatial averaging. I will demonstrate novel pattern-recognition methods that can efficiently separate externally forced climate change and/or predictable low-frequency variability (climate signals) from high-frequency variability (climate noise). These methods are particularly effective at filtering out spatially coherent internal variability such as El Niño and the North Atlantic Oscillation (NAO). Using pre-industrial control simulations, I show how these methods help to clarify the atmosphere-ocean mechanisms of Atlantic multi-decadal variability, isolating them from mechanisms relevant at shorter timescales. Using single-model large ensemble simulations of the 20th century, I show that these methods can identify forced climate responses with up to ten times fewer ensemble members than a simple ensemble average. This analysis elucidates forced responses in these simulations that are not otherwise apparent, such as an El-Niño-like response to volcanic eruptions and a forced trend in the NAO over the period 1950-1990. I will discuss how these pattern-recognition methods may help in identifying the forced component of observed temperature changes. Finally, I will discuss how these methods can be combined with other climate dynamics approaches to make progress on understanding how the hydrological cycle and climate extremes will change over the coming decades.