Intrinsic modulation of ENSO predictability viewed through a local Lyapunov lens
Intrinsic modulation of ENSO predictability viewed through a local Lyapunov lens
复制标题
DOI：
10.1007/s00382-013-1759-z
复制
发表时间：
2013
期刊：
影响因子：
4.6
通讯作者：
C. Karamperidou;M. Cane;Upmanu Lall;A. Wittenberg
中科院分区：
文献类型：
--
作者：
C. Karamperidou;M. Cane;Upmanu Lall;A. Wittenberg
在GFDL的CM2.1长期非强迫模拟中存在丰富的ENSO变率，促使使用动力系统理论的工具来研究ENSO可预测性的变率及其与ENSO震级、频率和物理演化的联系。本地李雅普诺夫指数（LLE）估计每月NINO 3 SSTa模式输出的特点增加或减少的可预测性。LLE描述了由于内部变化引起的无穷小扰动的增长，并且是系统相空间中任何给定点处的即时预测不确定性的度量。LLE派生的可预测性估计进行了比较，从一组重新预测实验CM2.1的误差增长。结果表明，LLE低估的误差增长短的预测提前期（小于8个月），而他们高估了较长的提前期。LLE推导的误差增长率从重新预测率的偏离是预测前置时间的线性函数，并且对用于LLE计算的时间序列的长度也很敏感。LLE推导的误差增长率更接近于4个月提前期的重新预测的估计值。在长达2,000年的模拟中，LLE导出的可预报性在4个月的提前期内的年代际变化（多）仅为9- 18%。ENSO活跃期比非活跃期更易预测，而LLE将具有规则周期性和中等量级的历元归类为最易预测的历元。西太平洋温跃层较深的事件在其峰值之前长达五年，沿着东太平洋温跃层在峰值之前几个月较早加深，被归类为更可预测的事件。此外，GCM被认为是可预测性比自然在这种措施的可预测性。
The presence of rich ENSO variability in the long unforced simulation of GFDL’s CM2.1 motivates the use of tools from dynamical systems theory to study variability in ENSO predictability, and its connections to ENSO magnitude, frequency, and physical evolution. Local Lyapunov exponents (LLEs) estimated from the monthly NINO3 SSTa model output are used to characterize periods of increased or decreased predictability. The LLEs describe the growth of infinitesimal perturbations due to internal variability, and are a measure of the immediate predictive uncertainty at any given point in the system phase-space. The LLE-derived predictability estimates are compared with those obtained from the error growth in a set of re-forecast experiments with CM2.1. It is shown that the LLEs underestimate the error growth for short forecast lead times (less than 8 months), while they overestimate it for longer lead times. The departure of LLE-derived error growth rates from the re-forecast rates is a linear function of forecast lead time, and is also sensitive to the length of the time series used for the LLE calculation. The LLE-derived error growth rate is closer to that estimated from the re-forecasts for a lead time of 4 months. In the 2,000-year long simulation, the LLE-derived predictability at the 4-month lead time varies (multi)decadally only by 9–18 %. Active ENSO periods are more predictable than inactive ones, while epochs with regular periodicity and moderate magnitude are classified as the most predictable by the LLEs. Events with a deeper thermocline in the west Pacific up to five years prior to their peak, along with an earlier deepening of the thermocline in the east Pacific in the months preceding the peak, are classified as more predictable. Also, the GCM is found to be less predictable than nature under this measure of predictability.