Are we near the predictability limit of tropical Indo‐Pacific sea surface temperatures?

Abstract

The predictability of seasonal anomalies worldwide rests largely on the predictability of tropical sea surface temperature (SST) anomalies. Tropical forecast skill is also a key metric of climate models. We find, however, that despite extensive model development, the tropical SST forecast skill of the operational North American Multi‐Model Ensemble (NMME) of eight coupled atmosphere‐ocean models remains close both regionally and temporally to that of a vastly simpler linear inverse model (LIM) derived from observed covariances of SST, sea surface height, and wind fields. The LIM clearly captures the essence of the predictable SST dynamics. The NMME and LIM skills also closely track and are only slightly lower than the potential skill estimated using the LIM's forecast signal‐to‐noise ratios. This suggests that the scope for further skill improvement is small in most regions, except in the western equatorial Pacific where the NMME skill is currently much lower than the LIM skill.

El Niño Southern Oscillation (ENSO) and Health: An Overview for Climate and Health Researchers

“…Given the expectations of the broad ENSO forecast user community related to ENSO forecasts as a panacea for climate risk management problems, much effort has been invested in improving predictability [120][121][122] with seasonal health forecasting scheme developers conscious that validation of predictions is a requisite part of the forecasting development process [104,123]. Further to the issues of predictability, other constraints related to seasonal health forecasting may well bear implications for the operationalisation of ENSO (climate)-sensitive disease early warning systems.…”

Observed El Niño‐La Niña Asymmetry in a Linear Model

“…If the deterministic dynamics are close to linear and the asymmetry is generated due to rapidly decorrelating unresolved nonlinear processes, then the prospects of predictability of these events is compromised; however, incorporation of state‐dependent noise would improve an ensemble (probabilistic) forecast by better characterizing the range of possible outcomes. Evidence for the role of state‐dependent noise in generating the asymmetry may be found in the similar skill that linear and nonlinear inverse models in the Niño3.4 area have (Chen et al, ), as well as further evidence suggesting that at seasonal time scales, the Tropical Pacific forecast skill is very close to linear, except perhaps in the far eastern Pacific (Ding et al, ; Newman & Sardeshmukh, ). In this case, further model development may not significantly improve the forecast skill of the largest EN events beyond what can be explained by a linear system forced by CAM noise.…”

“…Past analyses have used a Linear Inverse Model (LIM; Penland & Sardeshmukh, , hereafter PS95), to empirically extract a linear dynamical system forced with state‐ independent (i.e., additive) white noise from the covariance statistics of Tropical Pacific anomalies. Such a LIM can still explain ENSO irregularity (PS95), the (nonnormal) cycle of ENSO decay and growth (PS95, Vimont et al, ), the spectral characteristics of the main ENSO indices (Ault et al, ; Newman et al, ) and has also been shown to have forecast skill comparable to that of fully coupled, nonlinear General Circulation Models (Newman & Sardeshmukh, ), suggesting that it provides a good approximation to the Tropical Pacific deterministic dynamics at time scales relevant for ENSO (PS95; see also Penland, ). Moreover, inverse models augmented with deterministic nonlinearity do not seem to improve forecast skill over their linear counterparts (Chen et al, ; Kondrashov et al, ), which suggests that nonlinear processes contribute mostly to the unpredictable part of the system.…”

Enhancing ENSO Prediction Skill by Combining Model‐Analog and Linear Inverse Models (MA‐LIM)

“…The matrix is determined by an error variance minimization procedure as , where is the covariance matrix at lag ( month in this study) and the angle bracket denotes the expected mean. Following Newman and Sardeshmukh (), our LIM computes temporal evolution of a state vector in the EOF space, so represents the leading 16 and 9 principal components (PCs) of monthly SST and SSH anomalies, which explain 76% (80) and 61% (64) of the total variance, respectively, for the GCM simulations (observations). The number of EOF modes retained was chosen by trial and error to maximize the cross‐validated forecast skill, but the skill is relatively insensitive to this choice.…”