AUSPICE - Advancing Understanding of the Signal-to-noise Paradox and its Impacts on Climate Ensembles

The AUSPICE project will address a significant problem in climate forecasting known as the Signal-to-Noise Paradox (SNP). The SNP indicates that real-world predictability is often higher than models suggest, implying that the predictable signals in the forecast models are underestimated and posing a challenge for reliable climate forecasting. AUSPICE aims to develop innovative solutions to better understand and correct the SNP, improving climate predictions for the Euro-Atlantic region and beyond. Accurately predicting atmospheric circulation patterns over the North Atlantic is crucial for forecasting climate impacts, particularly for the UK and Europe, on seasonal, interannual and multiyear timescales. Weak forecast signals for impact-related metrics such as near-surface temperature, wind, and precipitation extremes, lead to poor decision-making in societal sectors sensitive to climate like agriculture and disaster management, and hinder the development of effective policies to address future climate change. By resolving the SNP, AUSPICE seeks to significantly enhance the reliability and usefulness of climate predictions and projections. The SNP has been most studied in the context of seasonal predictions of the wintertime North Atlantic Oscillation (NAO), a key climate driver in the region. Despite extensive research, the SNP remains unresolved. AUSPICE plans to tackle this challenge by focusing on four main objectives: O1: Develop improved and robust statistical tools to diagnose and interpret signal-to-noise errors. O2: Evaluate the contributions of predictable atmospheric circulation signals to the SNP. O3: Assess the role of local processes, feedbacks and model resolution in explaining the SNP. O4: Quantify the impacts of the SNP on future climate predictions and projections. AUSPICE will use causal networks as a novel framework for combining dynamical and physical modelling with advanced statistical methods to analyse complex interactions in the system and to address the SNP. Accurately identifying the SNP through statistical methods is challenging: The interpretability of a commonly used statistics metric, called the Ratio of Predictable Components (RPC), is severely limited by small sample sizes and theoretical assumptions and will be refined. Mitigation with brute-force larger ensembles is unfeasible due to the heavy computational costs. By leveraging machine learning, AUSPICE will develop more robust frameworks that recognize the intermittent and non-stationary nature of the SNP. Dynamical teleconnections between regional circulation and remote drivers, such as the tropical oceans, stratosphere and Arctic sea-ice, play a significant role in seasonal predictability. AUSPICE will run experiments with three state-of-the-art climate forecasting models to isolate these drivers and better understand their impact on the SNP. These targeted experiments will focus on key years when the SNP is most evident. In addition to teleconnections, AUSPICE will explore underrepresented physical processes in current models, such as interactions between small-scale atmospheric disturbances (eddies) and large-scale climate patterns, as well as air-sea interactions. Recent advancements in climate models, with increased resolution and better representations of ocean dynamics, show promise in addressing the SNP. The project will further investigate the role of physical processes like Rossby wave-breaking, the impact of precipitation on synoptic systems, and the interaction between these systems and Greenland’s topography. By integrating physical experiments and novel statistical methods into causal networks, AUSPICE will create a robust framework for understanding and overcoming the Signal-to-Noise Paradox. This will lead to more accurate and reliable climate predictions for the Euro-Atlantic region, enhancing decision-making across various sectors and advancing the scientific understanding of climate forecasting.