The American monsoon system in HadGEM3 and UKESM1

Abstract

Abstract. The simulated climate of the American monsoon system (AMS) in the UK models HadGEM3 GC3.1 (GC3) and the Earth system model UKESM1 is assessed and compared to observations and reanalysis. We evaluate the pre-industrial control, AMIP and historical experiments of UKESM1 and two configurations of GC3: a low (1.875∘×1.25∘) and a medium (0.83∘×0.56∘) resolution. The simulations show a good representation of the seasonal cycle of temperature in monsoon regions, although the historical experiments overestimate the observed summer temperature in the Amazon, Mexico and Central America by more than 1.5 K. The seasonal cycle of rainfall and general characteristics of the North American monsoon of all the simulations agree well with observations and reanalysis, showing a notable improvement from previous versions of the HadGEM model. The models reasonably simulate the bimodal regime of precipitation in southern Mexico, Central America and the Caribbean known as the midsummer drought, although with a stronger-than-observed difference between the two peaks of precipitation and the dry period. Austral summer biases in the modelled Atlantic Intertropical Convergence Zone (ITCZ), cloud cover and regional temperature patterns are significant and influence the simulated regional rainfall in the South American monsoon. These biases lead to an overestimation of precipitation in southeastern Brazil and an underestimation of precipitation in the Amazon. The precipitation biases over the Amazon and southeastern Brazil are greatly reduced in the AMIP simulations, highlighting that the Atlantic sea surface temperatures are key for representing precipitation in the South American monsoon. El Niño–Southern Oscillation (ENSO) teleconnections, of precipitation and temperature, to the AMS are reasonably simulated by all the experiments. The precipitation responses to the positive and negative phase of ENSO in subtropical America are linear in both pre-industrial and historical experiments. Overall, the biases in UKESM1 and the low-resolution configuration of GC3 are very similar for precipitation, ITCZ and Walker circulation; i.e. the inclusion of Earth system processes appears to make no significant difference for the representation of the AMS rainfall. In contrast, the medium-resolution HadGEM3 N216 simulation outperforms the low-resolution simulations due to improved SSTs and circulation.

A wavelet transform method to determine monsoon onset and retreat from precipitation time‐series

“…The precipitation output from the latest reanalysis from the European Centre for Medium‐Range Forecasting: ERA‐5, which has been shown to exhibit a relatively good representation of the temporal characteristics of rainfall in monsoon regions (e.g., García‐Franco et al ., 2020). Other variables from ERA5 used to diagnose changes to the circulation associated with monsoon onset were geopotential height at 500 hPa and wind speed ().…”

“…(2016). These simulations were recently evaluated in the North American Monsoon and the MSD region, showing a good representation of the seasonal cycle in these regions (García‐Franco et al ., 2020).…”

“…The pre-industrial Control experiments use constant forcing estimated for 1850 whereas the historical experiments aim to represent timevarying greenhouse emissions, volcanic eruption and solar signals in the historical period (1850-2014) Eyring et al (2016). These simulations were recently evaluated in the North American Monsoon and the MSD region, showing a good representation of the seasonal cycle in these regions (García-Franco et al, 2020).…”

“…The Coupled Model Intercomparison Project (CMIP) assessments of monsoon onset and retreat typically use precipitation threshold methods due to the lack of high temporal or vertical resolution output from all models to estimate vertically integrated quantities need for some methods (e.g., Geil et al ., 2013; Zou and Zhou, 2015; Ha et al ., 2020). Threshold methods have multiple shortcomings for CMIP assessments as the persistence and threshold parameters are tuned for observations with a specific seasonal cycle but models have a range of biases in the seasonality, magnitude and spatial distribution of rainfall (Pascale et al ., 2019; García‐Franco et al ., 2020). The use of pre‐defined threshold values may also not be suitable to compare different model experiments with changes in forcing where the climatological mean rainfall or the seasonal cycle may change within the model run.…”

“…The majority of existing methods to diagnose bimodal signals in this region use geometric or statistical measures of the monthly mean rainfall that measure the difference between the months of maximum rainfall and the drier months. However, this approach fails to capture the shorter‐scale changes that have been shown to occur in both observations and model data, as the MSD does not start or end exactly on given calendar months (Magaña et al ., 1999; García‐Franco et al ., 2020). Zhao and Zhang (2020) review and compare several methods to detect and measure the MSD, finding that using monthly‐mean data and prior assumptions of the dates of the first and second peaks can lead to errors.…”