09:45 〜 10:00
[ACG38-04] CS-Colored-LIM: a data-driven linear framework for extended ENSO predictions
キーワード：Cyclostationarity、Persistence of noise、Linear inverse model、Predictability
Linear inverse models (LIMs) are powerful tools in climate sciences for understanding and forecasting climate variability. They enable researchers to investigate the dynamics of coupled ocean-atmosphere systems and predict major climate phenomena. However, traditional LIM approaches typically assume stationarity and Gaussian white noise forcing, limiting their capacity to fully capture the complex climate system.
In this study, we extend the LIM framework by incorporating cyclo-stationarity (CS), which accounts for seasonally varying dynamics, and Ornstein-Uhlenbeck colored noise, which reflects the memory of atmospheric stochastic forcing from processes such as the MJO. We refer to this approach as CS-Colored-LIM. These modifications allow for a more realistic representation of the underlying physical processes driving climate variability.
Applying both existing LIMs and the proposed CS-Colored-LIM, we investigate Niño 3.4 index forecast skill by analyzing their performance in capturing key features of ENSO and its interaction with atmospheric noise. Our analysis shows that CS-Colored-LIM effectively reproduces the seasonality of ENSO variability, providing a mechanism to explain ENSO phase locking and the spring prediction barrier. We also quantify the contribution of the colored noise to the dynamics of the system, demonstrating the non-trivial role of the persistence of noise.
Results for Niño 3.4 forecast skill show that incorporating cyclo-stationarity improves short-range (≦ 12 months) predictability by capturing the month-to-month variability, while colored noise extends effective forecast skill at longer lead times (> 12 months) by accounting for autocorrelated atmospheric forcing. Our assessment indicates that CS-Colored-LIM benefits from both enhancements, with statistically significant improvements in correlation scores compared to existing LIMs. Moreover, these modifications also yield more accurate ensemble forecasts and robust predictions of major ENSO events. Consequently, CS-Colored-LIM stands out as a simple yet reliable predictive model for long-range climate forecasting and for advancing our understanding of ENSO dynamics.
In this study, we extend the LIM framework by incorporating cyclo-stationarity (CS), which accounts for seasonally varying dynamics, and Ornstein-Uhlenbeck colored noise, which reflects the memory of atmospheric stochastic forcing from processes such as the MJO. We refer to this approach as CS-Colored-LIM. These modifications allow for a more realistic representation of the underlying physical processes driving climate variability.
Applying both existing LIMs and the proposed CS-Colored-LIM, we investigate Niño 3.4 index forecast skill by analyzing their performance in capturing key features of ENSO and its interaction with atmospheric noise. Our analysis shows that CS-Colored-LIM effectively reproduces the seasonality of ENSO variability, providing a mechanism to explain ENSO phase locking and the spring prediction barrier. We also quantify the contribution of the colored noise to the dynamics of the system, demonstrating the non-trivial role of the persistence of noise.
Results for Niño 3.4 forecast skill show that incorporating cyclo-stationarity improves short-range (≦ 12 months) predictability by capturing the month-to-month variability, while colored noise extends effective forecast skill at longer lead times (> 12 months) by accounting for autocorrelated atmospheric forcing. Our assessment indicates that CS-Colored-LIM benefits from both enhancements, with statistically significant improvements in correlation scores compared to existing LIMs. Moreover, these modifications also yield more accurate ensemble forecasts and robust predictions of major ENSO events. Consequently, CS-Colored-LIM stands out as a simple yet reliable predictive model for long-range climate forecasting and for advancing our understanding of ENSO dynamics.