ENSO Modeling

Climate models are essential tools for understanding ENSO mechanisms and exploring the future, either via seasonal‐to‐decadal forecasting or climate projections. Because so few events are well observed, models are also needed to help reconstruct past variability, explore ENSO diversity, and understand the roles of the background mean state and external forcings in mediating ENSO behavior. In this chapter we review the history of ENSO mod­ eling, showing the gradual improvement of models since the pioneering studies of the 1980s and 1990s and describing the existing hierarchy of model complexity. The rest of the chapter is devoted to coupled general circulation models (GCMs) and how these models perform, related model development and improvements, associ­ ated systematic biases and the strategies developed to address them, and methods of model evaluation in a multi­ model context with reference to observations. We also review how successive generations of multimodel intercomparisons help bridge the gap between our theoretical understanding of ENSO and the representation of ENSO in coupled GCMs. Much of the improved understanding of ENSO in recent decades, addressed in other chapters of this monograph, was obtained from simulation strategies in which part of the coupled ocean‐atmosphere system was either simplified or omitted, such as atmosphere‐only, ocean‐only, partially coupled, or nudged simula­ tions. We here review these strategies and the associated best practices, including their advantages and limitations. The ability of coupled GCMs to simulate ENSO continues to improve, offering exciting opportunities for research, forecasting, understanding past variations, and projecting the future behavior of ENSO and its global impacts. We list the challenges the community is facing, as well as opportunities for further improving ENSO simulations. 1 LOCEAN-IPSL, CNRS/Sorbonne University/IRD/MNHN, Paris, France; and NCAS-Climate, University of Reading, Reading, UK 2 University of Colorado, CIRES, Boulder, CO, USA; and NOAA Physical Sciences Laboratory, Boulder, CO, USA 3 LOCEAN-IPSL, Sorbonne Universités/UPMC-CNRS-IRDMNHN, Paris, France; and MARBEC, University of Montpellier, CNRS, IFREMER, IRD, Sète, France 4 Institute of Atmospheric Sciences/Department of Atmospheric and Oceanic Sciences, Fudan University, Shanghai, China 5 NOAA Geophysical Fluid Dynamics Laboratory, Princeton, NJ, USA 202 EL NIÑO SOUTHERN OSCILLATION IN A CHANGING CLIMATE Studies of coupled models began to reveal biases that had been concealed in the ocean‐only or atmosphere‐only simulations used up to then. A community of scientists, working at the interface between the ocean and the atmosphere, steadily grew and now forms the core of ENSO expertise in the tropics. A series of coupled model intercomparison projects (CMIPs) have shown steady progress in simulating ENSO and related global variability using state‐of‐the‐art cou­ pled GCMs (AchutaRao & Sperber, 2006; van Oldenborgh et al., 2005; Guilyardi 2006; Capotondi et al., 2006; Wittenberg et al., 2006; Bellenger et al., 2014; C. Chen et al., 2017). Improvements in model formula­ tion and resolution have led to better representation of many key features of ENSO; see 4th and 5th assessment reports of the Intergovernmental Panel on Climate Change (IPCC AR4 and AR5) and the Special Report on the Ocean and Cryosphere in a Changing Climate. In con­ trast to the 1990s, progress in the past two decades has been gradual. A number of studies nevertheless have pointed to key factors essential to a realistic simulation of ENSO in a coupled GCM, in particular, properly repre­ senting deep convection and clouds in the atmospheric component (which depends to a large extent on the atmo­ spheric horizontal grid resolution), and properly repre­ senting equatorial wave dynamics, upwelling, and vertical mixing in the oceanic component (a strong function of oceanic grid resolution, especially in the meridional and vertical directions near the equator). The CMIP5 models showed progress relative to their CMIP3 counterparts, as all CMIP5 models displayed some kind of ENSO‐like behavior. However, the best CMIP5 models were only marginally better than the best CMIP3 models. CMIP5 also included models with increased nonlinear behavior, stemming mostly from better‐resolved atmospheric processes, such as convective thresholds or the ability to simulate intraseasonal variability like the Madden‐Julian Oscillation (MJO) and westerly wind bursts (WWBs, also known as westerly wind events or WWEs). Yet as detailed in section 9.4, systematic errors still persist, decades after their first identification. In the early 2000s, once models were able to simulate ENSO properties (e.g. amplitude and frequency) closer to observed, model evaluation began to include process‐ based metrics to ensure that the right properties were simulated for the right reasons and not via error compensation (Guilyardi et al., 2004, Kim & Jin, 2011). Besides providing invaluable feedbacks to model devel­ opers, multimodel intercomparisons continue to help bridge the gap between theoretical understanding of El Niño and its representation in coupled GCMs (CGCMs) (Fedorov et al., 2003; Held 2005; Kim & Jin, 2011). Hence, thanks to this improved theoretical understanding of ENSO, more mature diagnostic tools are now available to help unravel the underlying ENSO mechanisms. ENSO model evaluation has grown into a very active area of research, and exciting steps lie ahead. 9.2. BENEFITS OF A HIERARCHY OF MODELS A hierarchy of models of increasing complexity has made it possible to simulate, experiment with, and under­ stand the dynamics of ENSO. This hierarchy includes (i) simple oscillators, which describe the cyclic nature and essential parameters of the phenomenon; (ii) intermediate models, which describe the fluid dynamics and thermo­ dynamics of the equatorial ocean and atmosphere with some simplifications; and (iii) GCMs, which describe global climate with as much resolution and comprehen­ siveness as possible on the world’s most powerful super­ computers. Each type of model serves different goals and has its own advantages and requirements. The simplest models can capture novel theoretical concepts, highlight specific mechanisms, are valuable teaching tools, and have served as sources of insight into ENSO sensitivities and sources of predictability. The simple models are easily understood, tractable, and versatile, at the cost of being mostly qualitative, limited in focus, and sometimes difficult to relate directly to observations. In contrast, general circulation models are much more detailed as they attempt to account for the full complexity of the cli­ mate system; however, due to their complexity, such models are expensive to maintain and improve and more difficult to diagnose and understand. There are also important advantages in working simul­ taneously with models of different levels of complexity. Simple models can often be used to interpret GCMs and understand their biases via process‐based metrics (e.g. An & Jin, 2004; Jin et al., 2006; Brown et al., 2011; K.‐Y. Choi et al., 2013, 2015; Graham et al., 2015; Vijayeta & Dommenget, 2018). For example, the Bjerknes stability index, a process‐based metric derived from the recharge oscillator paradigm, has allowed the identification of errors in the GCMs (Kim & Jin, 2011), with some caveats (Graham et al., 2014). Conversely, the full characteriza­ tion of ENSO’s behavior gained from GCMs can inform the development of simpler conceptual models. For in­ stance, many studies adopt a hybrid approach where GCM outputs infer the parameters or characteristics of a simpler model that is then analyzed more extensively due to its lower computational cost. Finally, the above hier­ archy is flexible to some extent, because models some­ times couple components of vastly different complexity (e.g. an ocean GCM to a statistical atmosphere, etc.). Comprehensive coupled GCMs have been described in many places (Flato et al., 2013; Guilyardi et al., 2009), so we focus in this section on the simpler range of the model hierarchy. HISTORY AND PROGRESS OF ENSO MODELING 203 9.2.1. Harmonic Oscillator Models The simplest ENSO models are harmonic oscillators constructed from ordinary differential equations that capture the oscillatory nature of ENSO with periods of 2 to 7 years. Several harmonic oscillator models have been proposed. They all share a similar mathematical form but differ greatly in the variables and processes described, as well as in the approximations made to represent the oceanic and atmospheric dynamics (e.g. Picaut et al., 1997; Clarke et al., 2007). One example is the recharge/ discharge oscillator model (Jin, 1997), which in its sim­ plest form (Burgers et al., 2005) is expressed as