On Initializing CGCMs for Seasonal Predictability of ENSO

Initializing Coupled General Circulation Models (CGCMs) for routine seasonal ENSO prediction is currently an onerous task. This is one of the main reasons on why the CGCMs participating in the Coupled Model Intercomparison Project 5 (CMIP5), which represents the state-of-the-art in climate modeling, is infrequently used for routine seasonal prediction of El Niño and the Southern Oscillation (ENSO), the largest known natural variability that affects the global climate. In this work we propose a simple ocean initialization technique that can be adopted for any CGCM for seasonal predictability studies of ENSO. The technique entails finding the best analogues from a long historical simulation of the CGCM to the targeted air-ocean initial state. Since this study is on seasonal ENSO predictability, the metrics chosen to pick the analogues were confined to a set of 4 variables in the tropical Pacific that were found sensitive to the Niño3.4 SST variations. They were Tropical Pacific SSTs, thermocline depth, time tendency of thermocline depth, and the zonal wind stress. The multiobjective optimization technique was used to optimize the overall analogue match across the four variables giving equal weighting to each. This in effect uses the minimum root mean square difference between the targeted initial state and the model states to pick the analogue from the historical simulation of the CGCM that matched the targeted initial state. The chosen analogues were then perturbed using empirical singular vectors to provide additional initial conditions to generate in total 12 ensemble members per seasonal hindcast. The methodology for ocean initialization was first tested with the Cane-Zebiak model, a two layer reduced gravity ocean model coupled to a statistical atmosphere. We found that the methodology is sensitive to the length of the library generated from the historical simulation of the model and also on the fidelity of the model in simulating the ENSO. These toy model experiments also revealed the benefit of using a multi-variate metric to choose the analogues. Before proceeding to conduct the proposed work with a CGCM, the CMIP5 historical simulations for the 20th century were analyzed for their ENSO simulation. The mean-state and ENSO variations were analyzed in both the atmosphere and ocean. It was found that most of the CMIP5 models exhibit cold (warm) biases in the equatorial (subtropical eastern) Pacific Ocean sea surface temperature that are reminiscent of the split inter-tropical convergence zone phenomenon. There is, however, a major improvement in the representation of the power spectrum of the Niño3.4 sea surface temperature variations which shows that, as in the observations, a majority of the models display a spectral peak in the 2-7 year range, have a near linear relationship with the displacement of the equatorial thermocline and exhibit a robust atmospheric response to ENSO variations. Several issues remain in the CMIP5 simulations such as erroneous amplitudes in the Niño3.4 sea surface temperature spectrum's peak and a width of the spectral peak that is either too broad or too narrow. It is also seen that most CMIP5 models unlike the observations extend the ENSO variations in the equatorial Pacific too far westward beyond the dateline and there is very little asymmetry in event duration between the warm and cold phases. ENSO variability forces a dominant mode of rainfall variability in the southeastern United States, especially in the boreal winter season. The CMIP5 exhibited a wide range of response in this metric with several displaying weak to non-existent, some showing relatively strong, and one indicating excessively zonally-symmetric teleconnection over the southeastern United States. Based on this study we choose to use the CCSM4, which displayed a reasonable ENSO simulation for our experimental seasonal hindcasts with the proposed ocean initialization strategy. The seasonal hindcasts were initiated in beginning of March of each year from 1980-2012 follows from seeking a model state that minimizes the RMS difference in SST, zonal wind stress, thermocline depth and thermocline depth tendency from a 600 year continuous integration of the CCSM4 with the corresponding metric in the Global Ocean Data Assimilation (GODAS) of the National Centers for Environmental Prediction (NCEP). The four variables are jointly optimized by multi-objective optimization of the resulting root mean squared (RMS) difference curves, essentially minimizing the normalized RMS in all four parameters. Some of the main highlights of our results from the seasonal hindcasts are: i) The deterministic skill as measured by the anomaly correlation of the monthlyensemble mean and observed SST anomalies in the Niño3.4 region at 9-month lead is 0.71. ii) The probabilistic prediction of the Niño3.4 SST anomalies at 9-month lead for warm and cold ENSO events as measured by the area under the Relative Operating Characteristic Curve is 0.7 and 0.8 respectively. Likewise the brier skill score for warm and cold ENSO events at 9-month lead for Niño3.4 SST anomalies is 0.11 and 0.21 respectively. iii) The global teleconnection patterns in SST, precipitation and 500hPa geopotential heights with Niño3.4 SST variations in the seasonal hindcast in Oct-November-December season (7 month lead) is reasonable. From these results we demonstrate that the proposed initialization strategy is viable to deploy many other existing CMIP5 models for either operational seasonal ENSO prediction or ENSO predictability studies.