CMIP X-MOS: Improving Climate Models with Extreme Model Output Statistics

Abstract: Climate models are essential for assessing the impact of greenhouse gas emissions on our changing climate and the resulting increase in the frequency and severity of natural disasters. Despite the widespread acceptance of climate models produced by the Coupled Model Intercomparison Project (CMIP), they still face challenges in accurately predicting climate extremes, which pose most significant threats to both people and the environment. To address this limitation and improve predictions of natural disaster risks, we introduce Extreme Model Output Statistics (X-MOS). This approach utilizes deep regression techniques to precisely map CMIP model outputs to real measurements obtained from weather stations, which results in a more accurate analysis of the XXI climate extremes. In contrast to previous research, our study places a strong emphasis on enhancing the estimation of the tails of future climate parameter distributions. The latter supports decision-makers, enabling them to better assess climate-related risks across the globe.

• The paper introduces the X-MOS framework that integrates deep learning to map CMIP outputs to real-world measurements for more accurate extreme event prediction.
• It employs quantile regression and attention mechanisms to enhance accuracy metrics like MAE and average precision in forecasting extreme weather conditions.
• The approach improves climate risk assessments and offers a practical enhancement for existing CMIP models, aiding policy decision-making.

Improving Climate Models with Extreme Model Output Statistics

Introduction

The study titled "CMIP X-MOS: Improving Climate Models with Extreme Model Output Statistics" introduces the Extreme Model Output Statistics (X-MOS) framework aimed at enhancing the accuracy of climate models, particularly in predicting extreme weather events. Climate models from the Coupled Model Intercomparison Project (CMIP) are crucial for understanding climate dynamics, yet often fall short when addressing the tails of climate parameter distributions, which are essential for risk assessment related to extreme weather phenomena.

Methodology

The X-MOS approach integrates deep learning techniques to map CMIP model outputs directly to real-world measurements from weather stations. This mapping focuses on improving the prediction of extreme weather events by emphasizing the tails of these distributions through deep regression techniques.

Figure 1: The X-MOS approach: GhostNetV2 is used as MOS base utility illustrated in [tang2022ghostnetv2].

Key facets of the methodology include:

1. Training with Quantiles: The model regresses various quantiles, including extreme values, helping to better focus on statistical postprocessing outcomes.
2. Attention Models: The integration of attention mechanisms enhances the spatial-channel feature extraction, allowing for more nuanced modeling of climate data.
3. Utilization of Comprehensive Datasets: The paper employs datasets such as CMIP6 outputs, E-OBS v27 observations, and GSOD data, all contributing to the robustness of the framework.

Experimental Setup

The paper details an extensive experimental setup contrasting the X-MOS model with baseline methods like direct CMIP data outputs and traditional Linear Regression approaches. The dataset spans various periods up to 2100, with explicit attention to the accuracy of extreme weather prediction.

Metrics and Hardware

Metrics used for evaluation include Mean Absolute Error (MAE) for different quantile estimations, and Average Precision for identifying extreme weather conditions based on temperature and wind thresholds. The computational experiments are conducted on high-performance hardware such as the A100 GPU, facilitating efficient training of models.

Results

The results unequivocally underscore the advances made by the X-MOS approach in accurately modeling extreme weather events globally. In comparison with baseline approaches, the X-MOS model demonstrates superior performance across both temperature and wind predictions.

Figure 2: (Left) The difference between $0.95$ quantile tx variable of E-OBS v27 and interpolated $0.95$ quantile of CMIP6 tasmax, $2018.08.03\pm14$ days. (Right) The difference between $0.95$ quantile tx variable of E-OBS v27 and interpolated $0.95$ quantile of X-MOS $2018.08.03\pm14$ days.

Discussion

The improvements achieved by the X-MOS framework suggest significant practical implications in climate risk assessment and mitigation. The enhanced accuracy in predicting extreme events supports policymakers and decision-makers in developing informed strategies to manage climate-related risks. Additionally, the methodological innovations proposed in this paper could be integrated into existing CMIP-type models, broadening their applicability and enhancing their utility in future climate research.

Conclusion

The introduction of the X-MOS framework represents a noteworthy enhancement in the field of climate modeling, particularly in its application to extreme weather prediction accuracy. By leveraging advanced model output statistics, this approach offers a promising direction for future research and implementation in climate science. Further exploration into comparative analysis with emerging advanced models could offer additional insights to refine this approach further.