The research question centers on improving the accuracy and communication of seasonal climate forecasts, specifically concerning extreme events. The study is driven by the limitations of standard forecasting methods revealed by the devastating 2022 Pakistan floods. While seasonal forecasts, coordinated by the World Meteorological Organization (WMO), exist and utilize coupled ocean-atmosphere modes like El Niño Southern Oscillation (ENSO), their analysis using tercile probability maps often underestimates the likelihood of extreme events. The 2022 Pakistan floods, exceeding previous extreme events in 2010, highlight this problem. These floods caused unprecedented damage, and the existing forecasting system failed to convey the severity of the predicted rainfall. This study aims to understand why existing systems misrepresented the risk and propose improved methodologies for predicting and communicating extreme climate events. The importance of this research lies in enhancing societal resilience by providing more accurate and actionable early warnings of climate extremes, thus enabling better disaster preparedness and mitigation strategies.

Existing literature establishes the operational use of dynamical global climate models for seasonal climate prediction, coordinated by the WMO. Studies have linked past extreme rainfall events in Pakistan, such as those in 2010, to strong summer La Niña events and resulting atmospheric circulation shifts. These studies demonstrated the influence of ENSO teleconnections on South Asia's climate and the impact of extratropical influences. However, the challenge remains in evaluating the conditional skill of seasonal predictions given the limited number of ENSO events within typical hindcast periods. The use of tercile probabilities, while providing some discrimination, often masks the potential for more extreme events, leading to underestimation of risk, as seen in the 2022 Pakistan case. This study builds upon previous research by focusing on methods to identify and communicate the high probability of extreme events during these 'windows of opportunity', which are periods when predictable large-scale drivers such as ENSO significantly influence regional climate variability.

The study employs a combination of observational data and model output to analyze the 2022 Pakistan floods. Observed rainfall data comes from the Global Precipitation Climatology Project (GPCP) and Global Precipitation Climatology Centre (GPCC). Wind and sea level pressure data are from ERA5 reanalysis. The primary prediction system used is the Met Office DePreSys3 (DP3), a coupled climate model with a 43-year hindcast period and a large ensemble size. This system was chosen for its long hindcast period, enabling analysis of rare events, and its suitability for perturbation experiments. Multiple operational seasonal prediction systems from the Copernicus Climate Change Service (C3S) archive were also analyzed for comparative purposes. The methodology involves several steps: 1) Analyzing observed and predicted rainfall anomalies, focusing on probabilities beyond tercile levels (quintiles and deciles) to quantify the extreme nature of the 2022 forecast; 2) Examining the physical mechanisms driving the extreme rainfall through analysis of low-level (850 hPa) and upper-level (250 hPa) circulation anomalies, comparing model predictions to observations and historical teleconnections; 3) Conducting perturbation experiments using DP3 to isolate the influence of the strong summer 2022 La Niña on predicted rainfall and atmospheric circulation, by comparing forecasts with and without the La Niña signal; 4) Developing a simple empirical model using multiple linear regression based on indices representing the West Pacific Subtropical High (WNPSH) and the subtropical Asian jet meridional shift to understand the interaction of multiple drivers influencing rainfall; and 5) Developing and evaluating an interactive tool to identify and analyze extreme forecast signals across predefined global regions, incorporating historical performance and analysis of physical drivers to improve forecast confidence.

The 2022 Pakistan floods were an unprecedented extreme event (4.9 standard deviations above the climatological mean rainfall). While operational forecasts indicated an increased risk of above-normal rainfall, they failed to adequately convey the high probability of extreme rainfall. Analysis of DP3 and other operational systems revealed significantly raised probabilities for extreme rainfall quantiles, far exceeding tercile probabilities and eclipsing even previous wet years. The DP3 ensemble mean showed a +3.5σ anomaly for summer 2022 rainfall over Pakistan. A key driver was the strong summer 2022 La Niña, which the model successfully predicted. Perturbation experiments confirmed La Niña's significant role in driving the extreme rainfall forecast. Analysis of low-level and upper-level circulation anomalies showed excellent agreement between model predictions, observations, and historical teleconnections. The WNPSH and subtropical Asian jet shift indices, both skillfully predicted by DP3, played a crucial role in the extreme rainfall. A simple empirical model incorporating these indices successfully reproduced past rainfall variability and predicted the 2022 extreme. A novel interactive tool demonstrated the potential for automated identification and analysis of extreme forecast signals, allowing for further examination of physical mechanisms and improved confidence building. Evaluation of this tool over a 44-year hindcast period showed a much higher frequency of correct outer quantile outcomes than their climatological probabilities.

The findings demonstrate that strong signals for extreme rainfall preceding the 2022 Pakistan floods were present in operational seasonal prediction systems, though not effectively communicated. The strong summer La Niña acted as a window of opportunity, significantly enhancing the predictability of extreme rainfall. The study's success in attributing the extreme rainfall to predictable large-scale drivers, coupled with skillful prediction of relevant atmospheric circulation features, points to the potential for improved forecasting. The limitations of solely relying on tercile probabilities are underscored, necessitating a more comprehensive analysis of forecast probability distributions, including higher quantiles, to accurately assess risks. The proposed interactive tool offers a promising approach to identify, analyze, and improve confidence in forecasting extreme climate events by combining automated identification of signals with analysis of physical drivers. While the tool's effectiveness needs further real-time testing, the study's results suggest its substantial potential.

This study highlights the existence of 'windows of opportunity' for improved seasonal forecasting of climate extremes, exemplified by the 2022 Pakistan floods. The strong summer La Niña, effectively predicted by the models, played a pivotal role in creating these favorable conditions. The authors demonstrate the need to move beyond tercile probabilities to include higher-order quantiles for better representation of extreme events. A novel interactive tool is presented, integrating automated identification of extreme signals with assessment of physical drivers. This approach promises to enhance forecast confidence and lead to more effective early warnings. Future research should focus on real-time testing of the tool, exploring the influence of other drivers, and integrating social science perspectives to improve communication and user confidence.

The study's reliance on a single prediction system (DP3) for the perturbation experiments, while representative, may limit generalizability. The effectiveness of the proposed interactive tool needs to be fully evaluated in real-time operational forecasting. The analysis focuses primarily on rainfall, and other relevant factors, such as soil moisture and topography, could be considered to refine the predictions. Finally, the study acknowledges the importance of social science research in understanding how to communicate effectively and build user confidence, which are crucial aspects of effective early warning systems.