When and where do ECMWF seasonal forecast systems exhibit anomalously low signal‐to‐noise ratio?

Abstract

Seasonal predictions of wintertime climate in the Northern Hemisphere midlatitudes, while showing clear correlation skill, suffer from anomalously low signal‐to‐noise ratio. The low signal‐to‐noise ratio means that forecasts need to be made with large ensemble sizes and require significant post‐processing to correct the forecast distribution. In this study, a recently introduced statistical model of seasonal climate predictability is adapted so that it can be used to examine the signal‐to‐noise ratio in two versions of the ECMWF seasonal forecast system. Three novel features of the low signal‐to‐noise ratio are revealed. The low signal‐to‐noise ratio is present only for forecasts initialized on 1 November and not for forecasts initialized on 1 December. The low signal‐to‐noise ratio is predominantly a feature of the lower and middle troposphere and is not present in the stratosphere. The low signal‐to‐noise ratio is linked to low signal amplitude of the forecast systems in early winter. Future studies attempting to examine the signal‐to‐noise ratio should focus on the extent to which this early winter variability is predictable.

Explaining and Predicting Earth System Change: A World Climate Research Programme Call to Action

“…However, there is mounting evidence that climate models may underestimate atmospheric circulation signals in sub-seasonal (Domeisen et al, 2019;Charlton-Perez et al, 2019), seasonal (Eade et al, 2014;Scaife et al, 2014;Baker et al, 2018;Lee and Ha, 2015), interannual (Dunstone et al, 2016) and decadal (Athanasiadis et al, 2020;Smith et al, 2020) predictions, and in historical simulations (Lee et al, 2014;Zhang and Kirtman, 2019;Sévellec and Drijfhout, 2019;Klavans et al, 2021;Zhang et al, 2021). This error is especially clear in the North Atlantic, although there is some ongoing debate about the potential role of non-stationarity and sampling issues (Christensen et al 2022;Weisheimer et al, 2020).…”

A Minimal Model to Diagnose the Contribution of the Stratosphere to Tropospheric Forecast Skill

“…We start from the signal-noise model for ensemble forecasts developed by Charlton-Perez et al (2019) from Siegert et al (2016):…”

“…We start from the signal‐noise model for ensemble forecasts developed by Charlton‐Perez et al. (2019) from Siegert et al. (2016): $$$Y(t)={\mu }_{y}+{\beta }_{y}S(t)+\varepsilon O(t)$$$ $$${X}_{k}(t)={\mu }_{x}+{\beta }_{x}S(t)+\eta {P}_{k}(t)\qquad \text{for}\hspace*{.5em}k=1,\dots ,K$$$ In this model, $$Y(t)$$ is the observed time series of the parameter of interest for forecasts made at different times, $$t$$.…”

A statistical perspective on the signal‐to‐noise paradox

“…Markov-type models are used in Strommen and Palmer (2019) and Zhang and Kirtman (2019) to link the signal-to-noise paradox to underestimated persistence and regime behaviour, while Hardiman et al (2022) identify missing eddy feedback as a physical mechanism. Finally, evidence for a signal-to-noise paradox in seasonal forecasts involving the stratosphere is mixed, with ongoing discussions regarding whether and how the stratosphere adds predictive skill (Seviour et al, 2014;Byrne et al, 2019;Charlton-Perez et al, 2019;O'Reilly et al, 2019).…”

“…Our findings are in line with earlier studies (Eade et al, 2014;Siegert et al, 2016;Scaife and Smith, 2018;Christiansen et al, 2022), although a different and more quantitative angle allows us to derive several complementary results. Furthermore, our analysis does not rely on assuming a specific statistical model as in Siegert et al (2016), Charlton-Perez et al (2019), and Christiansen et al (2022.…”