Sensitivity of Coupled Tropical Pacific Model Biases to Convective Parameterization in CESM1

Abstract

Six month coupled hindcasts show the central equatorial Pacific cold tongue bias development in a GCM to be sensitive to the atmospheric convective parameterization employed. Simulations using the standard configuration of the Community Earth System Model version 1 (CESM1) develop a cold bias in equatorial Pacific sea surface temperatures (SSTs) within the first two months of integration due to anomalous ocean advection driven by overly strong easterly surface wind stress along the equator. Disabling the deep convection parameterization enhances the zonal pressure gradient leading to stronger zonal wind stress and a stronger equatorial SST bias, highlighting the role of pressure gradients in determining the strength of the cold bias. Superparameterized hindcasts show reduced SST bias in the cold tongue region due to a reduction in surface easterlies despite simulating an excessively strong low‐level jet at 1‐1.5 km elevation. This reflects inadequate vertical mixing of zonal momentum from the absence of convective momentum transport in the superparameterized model. Standard CESM1simulations modified to omit shallow convective momentum transport reproduce the superparameterized low‐level wind bias and associated equatorial SST pattern. Further superparameterized simulations using a three‐dimensional cloud resolving model capable of producing realistic momentum transport simulate a cold tongue similar to the default CESM1. These findings imply convective momentum fluxes may be an underappreciated mechanism for controlling the strength of the equatorial cold tongue. Despite the sensitivity of equatorial SST to these changes in convective parameterization, the east Pacific double‐Intertropical Convergence Zone rainfall bias persists in all simulations presented in this study.

Evolution of the Double‐ITCZ Bias Through CESM2 Development

“…To quantitatively assess the cold tongue evolution across model versions, we use the cold tongue index from Woelfle et al (): mean SST over the central Pacific cold tongue region (180° to 140°W; 3°S to 3°N; solid black box in Figure a) minus the mean SST averaged over the greater tropical Pacific basin (150°E to 110°W; 20°S to 20°N; dashed black box in Figure a). Using this metric, no trend appears in the annual mean cold tongue bias when examined as a function of model version (Figure ).…”

Could Machine Learning Break the Convection Parameterization Deadlock?

“…While promising, SP is not without its own idealizations that also limit its predictive ability and usefulness for climate simulation. For instance, restricting explicit convection to two dimensions makes it difficult to represent momentum transport (Arakawa, ; Jung & Arkawa, ; Tulich, ; Woelfle et al, ), and the limited CRM domain extent artificially constrains vertical mixing efficiency (Pritchard et al, ). Meanwhile, the typical use of 1–4 km CRM horizontal resolution and 250‐m vertical resolution cannot resolve important boundary layer turbulence, lower tropospheric inversions, and associated entrainment that are critical to low cloud dynamics (Parishani et al, ).…”

Spatially Extended Tests of a Neural Network Parametrization Trained by Coarse‐Graining

“…There is a long-standing debate about whether convective parametrizations should be based on moisture convergence or quasi-equilibrium closures (Arakawa, 2004). Moreover, important mean state biases in climate modeling such as the double Intertropical Convergence Zone bias are sensitive to convective parametrization (Woelfle et al, 2018;Zhang & Wang, 2006). Climate models also struggle to simulate observed aspects of tropical variability such as the diurnal cycle of continental precipitation (Stratton & Stirling, 2012) and the Madden Julian Oscillation (Jiang, 2017;Jiang et al, 2015).…”