A signal-to-noise paradox in climate science

Abstract

We review the growing evidence for a widespread inconsistency between the low strength of predictable signals in climate models and the relatively high level of agreement they exhibit with observed variability of the atmospheric circulation. This discrepancy is particularly evident in the climate variability of the Atlantic sector, where ensemble predictions using climate models generally show higher correlation with observed variability than with their own simulations, and higher correlations with observations than would be expected from their small signal-to-noise ratios, hence a 'signal-to-noise paradox'. This unusual behaviour has been documented in multiple climate prediction systems and in the response to a number of different sources of climate variability. However, we also note that the total variance in the models is often close in magnitude to the observed variance, and so it is not a simple matter of models containing too much variability. Instead, the proportion of Atlantic climate variance that is predictable in climate models appears to be too weak in amplitude by a factor of two, or perhaps more. In this review, we provide a range of examples from existing studies to build the case for a problem that is common across different climate models, common to several different sources of climate variability and common across a range of timescales. We also discuss the wider implications of this intriguing paradox.

Subseasonal‐to‐Seasonal Predictability of the Southern Hemisphere Eddy‐Driven Jet During Austral Spring and Early Summer

“…We have confirmed that the predictions of both of these signal/noise models are in good agreement with hindcasts that have been verified against a reanalysis. All of these results provide evidence of tropospheric predictability on S2S timescales during spring and summer from at least as early as 1 August and show no evidence of a signal‐to‐noise paradox between the hindcasts and the reanalysis (Scaife & Smith, ). We note that it may be the case that tropospheric predictability is larger in years with a more severe shift of the stratospheric seasonal cycle, with the SSW of 2002 perhaps the most extreme example of such behavior (see Thompson et al, , for a discussion of tropospheric impacts associated with the SSW of 2002.…”

“…More specifically, it allows for a continuous rather than a discrete representation of the signal. A potential disadvantage of this alternative signal‐to‐noise model is that recent work has suggested that some forecast models used in numerical weather prediction may be overdispersive (Scaife & Smith, ; but see also Weisheimer et al, ; Osman et al, ). If such an overdispersive scenario was the case for the present hindcast ensemble, then this alternative signal‐to‐noise model would offer an unduly pessimistic estimate of S2S forecast skill.…”

“…We subsequently use this simple model, along with the statistical model of Kumar (), to interpret skill scores associated with the hindcasts. This includes investigation of whether there is any evidence of a mismatch between anomaly correlation and signal‐to‐noise ratio—known as the “signal‐to‐noise paradox” (Scaife & Smith, ). We conclude with a summary of our results.…”

Drivers of the UK summer heatwave of 2018

“…() demonstrate, using a set of perturbation experiments, that the pattern of north Atlantic SSTs in May 2018 (and compared to 1976 in Figure ), sometimes referred to as the north Atlantic tripole pattern, correlates with northern European rainfall through the process of a northward‐shifted jet stream responding to the meridional temperature gradient of the north Atlantic sector. They go on to note that, although the North Atlantic SST tripole likely contributed to the observed circulation changes of summers, such as 1976 and 2018, the model produced only a very weak signal, which is often found in predictions of north Atlantic atmospheric variability (Scaife and Smith, ). In addition, the world has warmed significantly since the 1970s, and this is also partly reflected in the comparison of 2018 and 1976 in Figure .…”

Estimates of Decadal Climate Predictability From an Interactive Ensemble Model

“…Comparisons of observations and model simulations in the North Atlantic in terms of the mean state, total and decadal-scale variance, empirical orthogonal function and Atlantic multidecadal variability patterns ( Figures S1-S4) indicate that SST variability is underestimated in CCSM4, especially in the subtropics. This underestimate is possibly due to the underestimate of ocean-atmosphere interactions (Scaife & Smith, 2018) and the unresolved ocean eddy activities . The North Atlantic multimodel studies suggest that models differ greatly in the timescale, structure, and relation of ocean and atmospheric variability (Ba et al, 2014;Kavvada et al, 2013), which could lead to different estimates of predictability (Branstator et al, 2012).…”