Effects of stochastic parametrization on extreme value statistics

Area of Science:

• Earth System Science
• Climate Modeling
• Geophysical Event Prediction

Background:

• Extreme geophysical events pose significant risks to human life and property.
• Accurate modeling and probabilistic prediction are essential for risk assessment and loss reduction.
• Parametrizations in Earth system models simulate unresolved scales but are typically evaluated on typical, not extreme, events.

Purpose of the Study:

• To investigate the performance of two distinct parametrizations in representing extreme events.
• To compare an empirical fitting approach with a statistical mechanical approach for parametrization.
• To assess the suitability of current evaluation methods for extreme event prediction.

Main Methods:

• Utilized a modified two-level Lorenz'96 model.
• Implemented and compared two parametrizations: one empirical (Wilks, 2005) and one statistical mechanical (Wouters & Lucarini, 2012, 2013).
• Analyzed model performance specifically for extreme event representation versus bulk statistics.

Main Results:

• Both parametrizations exhibited different strengths and weaknesses.
• Model agreement with the true model was generally poorer for extreme events compared to the bulk of statistics.
• The empirical parametrization showed some advantages in capturing extremes.

Conclusions:

• Standard optimization procedures for parametrizations may not adequately capture performance on extreme events.
• Stochastic parametrizations require specific and accurate testing against extreme event data.
• Improved methods are needed to ensure Earth system models reliably predict extreme geophysical phenomena.