The American Monsoon System in HadGEM3.0 and UKESM1 CMIP6 simulations

The simulated climate in the American Monsoon System (AMS) in the CMIP6 submissions of HadGEM3.0 GC3.1 and the UKESM1 is assessed and compared to observations and reanalysis. Pre-industrial control and historical experiments are analysed to evaluate the model representation of this monsoon under different configurations, resolutions and with and without Earth System processes. The simulations exhibit a good representation of the temperature and precipitation seasonal cycles, although the historical experiments overestimate summer temperature in the Amazon, Mexico and Central America 5 by more than 1.5 K. The seasonal cycle of rainfall and general characteristics of the North American Monsoon are well represented by all the simulations. The models simulate the bimodal regime of precipitation in southern Mexico, Central America and the Caribbean known as the midsummer drought, although with a stronger intraseasonal variation than observed. Austral summer biases in the modelled Atlantic Intertropical Convergence Zone (ITCZ), Walker Circulation, cloud cover and regional temperature distributions are significant and influenced the simulated spatial distribution of rainfall in the South 10 American Monsoon. These biases lead to an overestimation of precipitation in southeastern Brazil and an underestimation of precipitation in the Amazon. El Niño Southern Oscillation (ENSO) characteristics and teleconnections to the AMS are well represented by the simulations. 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 UKESM has the same performance as the lower resolution simulation of HadGEM3.0 GC3.1 and no significant difference for the AMS was found between the two 15 model configurations. In contrast, the medium resolution HadGEM3.0 GC3.1 N216 simulation outperforms the low-resolution simulations in temperature, rainfall, ITCZ and Walker circulation biases. Copyright statement. TEXT

. Summary of the datasets used in this study. For each dataset, the acronym used hereafter, the period of coverage, the field used and the horizontal resolution are shown. Some datasets extend further back in time, but only the satellite-era period is used in most of the datasets. The variables used are: precipitation, surface temperature (T surf ), sea-level pressure (SLP), SSTs, the x and y components of the wind (u, v), the lagrangian tendency of air pressure (ω), outgoing longwave radiation (OLR) and specific humidity (q).  (Kennedy et al., austral summer in South America has a significant bias in the easterly flow coming from the equatorial and subtropical Atlantic.

135
The biases in the low-level winds suggest a weaker easterly flow into southeastern Brazil but also a strong southward flow from northern to southern South America. For example, the Bolivian Low-Level Jet, which is the strong southward flow observed in Figure 1a in Bolivia, is stronger in the simulations.
During boreal summer (Figures 1d, f), positive biases are found in southwestern North America (> 3.5 K), which are higher in UKESM1-hist than in GC3-hist. The flow in the western coast of Central America has a bias in UKESM1 in the easterly  similar bias in the circulation in South America, particularly for GC3 N96-pi. The South American low-level circulation of GC3 N216-pi has the smallest biases with respect to ERA5 amongst all the simualtions. UKESM1-pi was found to be almost indistinguishable from GC3 N96-pi, which is why in this and the following sections only GC3 N96-pi results are shown. Figure   2e, f show the difference between the historical and piControl experiment of GC3, which illustrates the response to historical forcing in GC3. This temperature response in South and Central America was of about 1.5 K whereas in JJA in North America, 150 temperatures were 4 K higher in the historical experiment than in the piControl. A very similar temperature pattern response to historical forcing was observed for UKESM1 (not shown) although of slightly different magnitude. The only dynamical response to forcing seems to be the easterlies in the East Pacific Ocean during JJA.
The seasonal evolution of temperature in key regions of the AMS is shown in Figure 3 which provides a better comparison of the temperature field in these experiments. The strongest seasonal contrast in surface temperature is in the North American

155
Monsoon region, where wintertime temperatures are roughly 12 • C and June temperatures are close to 27 • C. Although colder than observed in the piControl and warmer in the historical experiments throughout the whole year, the models accurately reproduce the seasonal cycle of this region, which may be relevant for the simulated onset timing and strength of the monsoon (Turrent and Cavazos, 2009).
The piControls show a colder-than-observed winter in southern Mexico and northern Central America whereas the historical 160 experiments show a warming signal of about 1.5 K in winter and 2 K in the summer when compared to the piControls. In spite of these biases, both types of experiments follow closely the seasonal cycle in North and Central America. However, the temperature cycle in South America is poorly represented in the simulations (Figures 3 c, d). The models seem to reproduce a stronger than observed seasonal cycle, as observed by the 4 K temperature difference between late austral winter and spring, whereas the annual cycle of temperature varies by less than 1 K in the observations. The warmer than observed Amazon ( Fig.   165 3 d) bias peaks in austral spring (SON), during the development of the monsoon (Marengo et al., 2012). In southeastern Brazil, the seasonal cycle is reasonably well reproduced but with a significant cold bias throughout the year which is significantly larger during austral winter (JJA), as models (e.g. UKESM1) simulate a temperature of 292 K which is 4 K lower than the 6 https://doi.org/10.5194/wcd-2020-8 Preprint. Discussion started: 19 March 2020 c Author(s) 2020. CC BY 4.0 License.
observed 296 K. In all panels of Figure 3, the historical experiments show a larger warming signal as a response to forcing in UKESM1 than in GC3.

The ITCZ and the Walker circulation
The AMS is intertwined with the seasonal migration of the East Pacific and Atlantic ITCZ and associated with the Walker circulation through teleconnections (Zhou et al., 2016). This section validates the modelled ITCZs and Walker circulation. Figure 4 shows the observed and modelled climatological rainfall and the ITCZ climatological position in the East Pacific and Atlantic Oceans. Three simulations are shown: the ensemble-mean UKESM1-historical, GC3 N96-pi and GC3 N216-pi. 175 We choose this set of simulations because the simulations of lower resolution, both historical and piControl, showed very similar ITCZ characteristics, as will be described below, and relatively few differences were found across, e.g., GC3 N96-pi and GC3-hist. summer. The boreal winter position of the modelled ITCZs is displaced with respect to the observations. The simulated ITCZ crosses south of the equator during boreal winter to a region with rainrates above 12 mm day −1 that are found between 10-0 • S.
After boreal spring, the modelled ITCZ crosses back north of the equator and matches the observed ITCZ reasonably well for boreal summer and fall.

200
Low-level wind biases are also found near the ITCZ, for instance, between days 1 and 100, Figures 5f and h show that north of the equator the models show a stronger than observed northward wind, and a stronger than normal southward wind south of average around August 16th. However, winter-time rainfall, before monsoon onset and after monsoon retreat, is overestimated by all the simulations, particularly the higher resolution GC3.1 N216 which has a positive bias of 2 mm day −1 in early winter.
The strong monsoonal seasonal cycle of precipitation in eastern Brazil is characterised by a very wet summer (∼8 mm 270 day −1 ) compared to a very dry (∼0.2 mm day −1 ) winter. The austral summer rainfall in the observations consistently shows that maximum rainfall is found in early January (∼8 mm day −1 ). Rainfall in this region decreases to ∼6 mm day −1 by late March as the monsoon migrates northward, to then, sharply descend in austral fall (April). The models (Figure 9c) show a positive bias at the peak stage of the monsoonal rainfall. This bias was found to be of +4 mm day −1 and +2.5 mm day −1 for the low and high resolution simulations, respectively. The bias in the seasonal cycle is consistent with the seasonal mean bias 275 shown in Figure 7, which showed that rainfall in southeastern Brazil is overestimated in all the simulations, but this bias is smaller in the GC3 N216-pi simulation. In spite of this positive bias in the magnitude of precipitation, the seasonal evolution of rainfall is very well represented by the simulations, as the onset and retreat dates are in close agreement with the observations.
Finally, the simulated rainfall in the Amazon is in very good agreement with the observations during austral winter ( Figure   9d). The models also show a good representation of the transition from winter to summertime rainfall by representing with 280 relative skill the smooth transition from 4 mm day −1 in September to 6 mm day −1 in November and close to 8 mm day −1 in late December. However, peak summertime rainfall in January and February is underestimated by all the simulations. The low resolution simulations, after simulating an annual maximum of rainfall in December, simulate a decrease in precipitation for January and February, whereas the observations show the opposite behaviour. Rainfall in the Amazon from January to March, in both TRMM and CHIRPS, is close to 10 mm day −1 , yet the low resolution simulations present rainfall rates of 8 mm day −1 285 or even less in mid-February. GC3.1 N216 shows a better agreement with observations but still underestimates summertime rainfall by 1 mm day −1 , outside of the uncertainty range of TRMM.

OLR and q
The seasonal cycles of out-going longwave radiation (OLR), vertical velocity (ω) and specific humidity (q) are key features of a monsoon since these quantities characterise the strength and height of deep convection and the mid-level moisture. Figure   290 10 shows the pentad-mean annual cycle of OLR, q and ω at the 500-hPa level in four regions of the AMS, as in section 4.2.
For the North American Monsoon the seasonal cycle of OLR, q and ω is relatively well represented in the simulations. During late boreal winter and early spring, OLR increases steadily as a result of surface warming. However, in early June, close to monsoon onset (Douglas et al., 1993;Geil et al., 2013), OLR sharply decreases reaching a minimum value of 246 W m −2 by mid-July. The vertical velocity decreases steadily from January to a minimum in August, indicating ascent from May 1st 295 until September 15th. The models show similar seasonal cycles but overestimate the summertime OLR by ≈ 6 W m −2 and underestimate mid-level moisture by 0.3 g/kg and ω by 0.01 Pa s −1 . After convective activity decreases in late August in ERA-5, OLR increases to a local maxima of ∼ 271 W m −2 on mid-September, q decreases significantly and ω turns positive. The simulated shallower convection and drier mid-troposphere may be compensated by stronger ascent in the mid-troposphere.
In the MSD region, OLR and q show signs of convective activity from mid-April, as OLR decreases and moisture increases 300 steadily from April until June. The characteristic MSD bimodal distribution can also be observed as two peaks of low OLR, The near-surface air temperature and SLP response to El Niño and La Niña events is shown in Figure 11 for The SLP response to ENSO events is mostly observed in the subtropical high pressure systems (Vera et al., 2006;Marengo et al., 2012). In particular, the response in the northern Pacific and Atlantic, known as the Pacific North-American pattern 345 provide insight into the Rossby wave source and the effect on the midlatitude jet arising from ENSO events. A weakened North Pacific Subtropical High is observed in ERA5, with an SLP anomaly of -4 hPa off the coast of California. The models show a similar but smaller SLP response in the same region. In the North Atlantic, positive ENSO events produce a negative NAO response, with opposite response for negative ENSO events. While the models seem to be able to capture this response of Atlantic SLP, the simulated response is weaker in the low resolution simulations.

350
The rainfall anomalies to ENSO events are shown in Figure 12. Three regions in the AMS have a significant precipitation response to ENSO events in the observations and simulations. In southern North America, rainfall increases (decreases) during El Niño (La Niña) events due to the effect of Rossby waves on the subtropical jet and wintertime midlatitude disturbances (Vera et al., 2006;Bayr et al., 2019). The GPCP dataset (Figure 12a, b) shows significant boreal winter rainfall increases in southeastern US and the Gulf of Mexico during El Niño events, and an opposite response to La Niña phases. All the simulations 355 sensibly reproduce this teleconnection rainfall pattern.
The core Amazon basin shows the strongest response to ENSO events in the observations. This teleconnection works through the perturbation of ENSO to the Walker circulation (Vera et al., 2006;Cai et al., 2019). Strong positive (negative) rainfall anomalies during the negative (positive) phases of ENSO in northern South America are observed in GPCP. All the simulations show a very similar and statistically significant response. The strongest simulated response is that of GC3 N96-pi, 360 especially over eastern Brazil and the equatorial Atlantic Ocean. The models also show the observed response in the third region, southeastern south America, which shows an opposite sign response to ENSO events to that of the Amazon (Vera et al., 2006). The climate hazards infrared precipitation with stations-a new environmental record for monitoring extremes, Scientific data, 2, 150 066,

2015.
Gamble, D. W., Parnell, D. B., and Curtis, S.: Spatial variability of the Caribbean mid-summer drought and relation to north Atlantic high