3.1 Mean Pattern of Indian Summer Monsoon
Figure 1 compares the model simulated seasonal mean total, convective, and large-scale precipitation with corresponding observations. For total precipitation, observation shows the highest values (> 8 mm/day) over the north-eastern Bay of Bengal (BoB), Indo-Burmese mountains, foot-hills of Himalaya, and Western Ghats (WGs); moderate values (> 3 mm/day and < 8 mm/day) over the eastern EIO, central India, and eastern AS; and lowest values (< 3 mm/day) over the western AS, northwest India, leeward side of WGs, and western EIO (Fig. 1a). From the difference plot for total precipitation, it is noted that CAM3 simulates quite well over WGs and central India. However, there is a large wet bias over the western EIO and south-eastern coastal India and dry bias over the north-eastern BoB and Indo-Burmese mountains (Fig. 1b). While these wet and dry biases were alleviated significantly in CAM4, a new significant wet bias emerged over WGs, eastern AS, and the Himalayan region (Fig. 1c). In CAM5, the wet bias over eastern AS and WGs was alleviated to a large extent. The existing wet bias over the Himalayan region and dry bias over the north-eastern BoB was increased (Fig 1d). In CAM6, many of the existing wet and dry biases are reduced; however, the wet bias over WGs and the Himalayan region remains, and the wet bias over peninsular India has worsened (Fig. 1e).
Further splitting total precipitation into convective and large-scale components, we find that the observed spatial pattern of both components is similar to total precipitation except lesser in magnitude (Fig. 1f,k). It is also noted that both convective and large-scale components contribute nearly equally to the total precipitation over the ISM region. The difference plot for convective precipitation shows that CAM3 has a large wet bias over the south-eastern coast of India and western EIO (Fig. 1g). The wet bias over western EIO and south-eastern coast of India is greatly reduced in CAM4, but the wet bias over the Himalayan region and eastern AS is substantially increased (Fig. 1h). In CAM5, the existing wet biases are further reduced, except the wet bias over the Himalayan region, which is slightly increased (Fig. 1i). Compared to its predecessors, CAM6 shows a notable improvement in existing biases for convective precipitation (Fig. 1j). Further, the difference plot for large-scale components shows the large dry bias in CAM variants over the north-eastern BoB and Indo-Burmese mountains (Fig. 1l). While this dry bias in CAM4 was reduced substantially, a new wet bias emerged over the Himalayan region (Fig. 1m). In CAM5, these existing wet and dry biases were increased (Fig. 1n). However, in CAM6, the existing dry biases were greatly reduced (Fig. 1o). Thus, the wet bias over the Himalayan foothills and dry bias over the north-eastern BoB in the successive CAM variants are arising primarily through the wet and dry biases in large-scale precipitation, respectively.
Furthermore, for the annual cycle of total precipitation (Fig. 2a), we find a large dry bias in CAM3 and wet bias in CAM4 during June to October (highest during the peak monsoon period). In the successive CAM variants (CAM5-6), this wet bias seen in CAM4 is reduced substantially, but in CAM6 during June-July, the wet bias is slightly worsened with respect to CAM5. Past research findings have suggested that the cold and warm SST biases over EIO in a model can cause the dry and wet bias over the Indian land region and hence leading to the dry and wet bias in the annual cycle, respectively (Roxy et al. 2012; Joseph et al. 2012). From the frequency-intensity distribution of daily total precipitation rates over India during JJAS (Fig. 2b), we find that all CAM variants overestimate the frequency of light precipitation rate (<10 mm/day) and underestimate the frequency of large to extreme precipitation rate. The underestimation in the frequency of large to extreme precipitation is largest in CAM3 (which does not report any event > 90 mm/day) and is reduced relatively in CAM4 and CAM5. While in CAM6, the underestimation in the frequency of extreme precipitation rate is increased compared to CAM4 and CAM5. Such overestimation in the frequency of light precipitation rates and underestimation in the frequency of extreme precipitation rates have also been noticed in many of the CMIP5 models (e.g., Dai 2006; Deng et al. 2007; Mishra et al. 2018; Salunke et al. 2019).
3.2 Horizontal Wind Pattern
Figure 3 shows the seasonal mean pattern of atmospheric circulations at 850 hPa and 200 hPa from ERA-I and CAM variants. It is noted that CAM variants simulate the circulation patterns comparable to ERA-I, for example, the simulation of low-level cross-equatorial westerlies, low-level cyclonic wind over northern India, TEJ (with maximum intensity over peninsular India), subtropical westerly jet (STJ), and Tibetan anticyclone. However, from the difference plot at 850 hPa, we find that CAM3 underestimates the south-westerly wind over peninsular India and BoB, resulting in an underestimation of total precipitation over BoB and Indo-Burmese mountains due to reduced moisture transport (e.g., Swapna and Kumar 2002; Puranik et al. 2014). In CAM4, 5, and 6, we find an overestimation of the south-westerly wind over AS and peninsular India, with largest overestimation in CAM6, resulting in an increased precipitation over WGs and peninsular India (e.g., Swapna and Kumar 2002; Ratna et al. 2014).
From further analysis of the difference plot at 200 hPa, we find that CAM3 shows large underestimation in TEJ over peninsular India and EIO, which could be a reason for the weakened vertical easterly wind shear (see Section 3.3) and hence the weakening of the monsoon circulation and associated precipitation (Fig. 1). The association of the weakening (strengthening) of TEJ and decrease (increase) in tropical summer precipitation is also reported in past studies (e.g., Koteswaram 1958; Kanamitsu et al. 1972; Kobayashi 1974; Pielke et al. 2001; Sathiyamoorthy 2005; Sreekala et al. 2014). Furthermore, we find an overestimation in the core of STJ in CAM3. Modulations in the core of STJ have been reported to influence the precipitation distribution over northern India through modulation of the Tibetan anticyclone (Ramaswamy 1962). However, circulation biases seen in CAM3 have improved in the successive versions with advances in model physics.
3.3 Tropospheric Temperature Gradient and Easterly Wind Shear
Figure 4a depicts the tropospheric temperature gradient (ΔTT), computed as the difference in vertically (600-200 hPa) averaged temperature between the northern box (5°-35°N; 40°-100°E) and southern box (15°S-5°N; 40°-100°E). These two boxes represent the large-scale temperature gradient zones responsible for the seasonal reversal of winds over the ISM region due to the differential heating, primarily over the Tibetan plateau (Gill,1980; Yanai et al. 1973; Webster et al. 1998; Goswami and Xavier 2005; Xavier et al. 2007). Increased ΔTT suggests stronger monsoon circulation and increased associated precipitation. The onset and withdrawal are also established when ΔTT changes sign from negative to positive and from positive to negative, respectively. ERA-I shows monsoon onset in late May, peak precipitation in July and August, and withdrawal in early October, also noticed in other studies (Dey 1970; Indian Meteorological Department (IMD) 1972; Fasullo et al. 2003). Similar characteristics of ΔTT are noticed in CAM variants. However, the monsoon onset and withdrawal are simulated earlier than observed, by one to two weeks in CAM variants, except CAM6, which shows early onset but similar withdrawal as ERA-I (and hence a prolonged monsoon season). ΔTT during JJAS is largely overestimated in CAM6, resulting in strengthening of low-level jet (Fig. 3) and an overestimation in precipitation over peninsular India (Fig. 1).
We also show the annual cycle of easterly wind shear (Fig. 4b), which is defined as the difference in the zonal wind between 850 hPa and 200 hPa, averaged over peninsular India (0 - 15°N; 50°-90°E) (Webster and Yang 1992; Jiang et al. 2004). This is known to cause vertically integrated moisture to propagate northward (EIO to central India), affecting atmospheric instability and convective activity over central India (Zhou and Murtugudde 2014). ERA-I shows a change from negative to positive wind shear in late April, with peaks in July and August, then positive shear until November (Fig. 4b). From our simulations, we found that CAM3 fails to accurately simulate this shear, although the subsequent CAM variants show improvements but still underestimate it, except CAM6. CAM6 simulates this shear close to ERA-I (compared to its predecessors), with slight overestimation in June to August, resulting in a better monsoon simulation and moisture transport towards BoB. This overestimated easterly wind shear in CAM6 (through enhanced low-level wind strength) could be a reason for an increase in precipitation over peninsular India (Fig. 1e).
3.4 Monsoon Intra-seasonal Oscillations (MISO)
One of the most significant modes of ISM variability is the 30-60 day oscillations of northward propagating convection anomalies over the ISM region (e.g., Goswami et al. 1998; Sharmila et al. 2012; Joseph et al. 2012). This northward propagating convection anomalies from EIO to the Indian subcontinent from June to September is referred to as the monsoon intra-seasonal oscillations (MISO). It is reported to explain more than 20% of the total JJAS rainfall variance over the Indo-Pacific region (Goswami et al. 1998). During the summer monsoon, the active and break spells have also been linked to MISO (e.g., Joseph et al. 2009; Krishnan et al. 2009; Goswami et al. 2011). In Figure 5, we show the space-time evolution of MISO from day -20 to +10 for observations and CAM variants. It is computed as a time-series of normalized area-averaged filtered (20-100 day) precipitation anomalies over central India (15°-25°N; 70°-90°E), with 20-100 day filtered precipitation anomalies regressed at different time lags during JJAS. The precipitation maximum over central India is thus on day 0 of MISO. From observations, the convection initiation occurs over central EIO on about day -20, and it spreads eastward (day -15) and then moves north-eastward to the Indian subcontinent by day -5. Around day 0, MISO has a strong eastward tilted convection band over the monsoon trough region and suppressed convection over EIO (Annamalai and Slingo 2001). This convection band shifts to Himalayan foothills by day +10, accompanied by corresponding north-eastward movement of the negative anomalies from EIO. Simulations show that CAM3 fails to capture the MISO pattern, but the successive CAM variants show the initiation of organized convection over central EIO and subsequent north-eastward movement comparable to observations. However, in CAM4, there is early suppression of convection (about -15 to -10 days) followed by enhanced convection (from day 0 onward) over EIO, but this is greatly improved in the subsequent variants, namely, CAM5 and CAM6. In addition, the observed north-eastward tilt in the convection band is underestimated in CAM simulations (lower eastward tilt), although with a larger spatial extent.
Further, from the Hovmöller diagram of MISO (Fig. 6), we find that the north-eastward propagation of convection from central EIO to the Indian subcontinent in CAM6 is consistent with observations but shows a weaker eastward component. CAM4 and CAM5 also capture northward propagation, but convection initiates over southern peninsular India instead of central EIO between day -30 to -20. Furthermore, the observed eastward propagating component of MISO is simulated westward in CAM4 and CAM5 (i.e., the north-eastward propagation of convection anomalies is simulated north-westward). This indicates that MISO simulations have improved over time in the subsequent CAM variants, with CAM6 showing the highest improvement.
Further, the important atmospheric processes are analyzed to understand how the MISO has improved in the subsequent CAM variants. Previous research has highlighted the role of atmospheric internal dynamics along with easterly wind shear and meridional asymmetry in specific humidity on the underlying mechanism of MISO (e.g., Webster 1983; Wang and Xie 1997; Jiang et al. 2004, 2011; Abhik et al. 2013). Previous studies have also suggested that improvement in the simulation of seasonal mean climatology from the equator to 15°N, as well as improvement in the movement of convection band from the equator to monsoon trough, are linked to improvements in the simulation of seasonal mean heat source in the EIO region and its interaction with regional and planetary-scale circulations (Attada et al. 2014). Hence, we next analyze the regional Hadley and planetary-scale Walker circulations and the model's internal dynamics in the following subsections.
3.4.1 Hadley and Walker Circulations
The regional Hadley and the planetary-scale Walker circulations are crucial elements of ISM circulation (Oort and Rasmusson 1971; Rao 1976; Sikka 1980; Krishnamurthy and Kinter 2003; Gadgil 2003; Annamalai et al. 2007; Lau et al. 2015; Fan et al. 2017). The movement of the equatorial heat source influences the seasonal mean Walker circulation, while the intensity and position of the monsoon heat source can affect the regional Hadley circulation (e.g., Goswami et al. 1999). Thus, both the atmospheric circulations associated with the heat sources can strongly affect the distribution of seasonal mean precipitation during ISM. From Figure 7 for Hadley circulation, we find the observed subsidence over the southern Indian ocean beyond 10°S and ascent over 10°S to 25°N. CAM3 shows ascending motion over the equator through 25°N (with the strongest ascent over the equator and 12°N), likely associated with higher precipitation over EIO and AS (Fig. 1), leading to weaker northward propagation of equatorial convection to the Indian subcontinent (Sharmila et al. 2013). This stronger than observed ascending motion over the equatorial region seen in CAM3 is improved in CAM4, but the ascent over 12°N is slightly increased with an increased precipitation bias over AS, although a small improvement in northward propagation of convection from the equator is noted from improved ascent over EIO. Subsequent CAM variants (CAM5 and CAM6) show improved simulations of ascent over 12°N, leading to improved northward propagation and spatial pattern of precipitation (Fig. 1). Similar to Hadley circulation, from Figure 8 (Walker circulation), we find observed subsidence over 40°-65°E and ascent over 65°-160°E (Walker 1924; Sikka 1980; Power and Kociuba 2011). In CAM3, the subsidence is largely underestimated (almost absent), and the ascent over 60°E is highly overestimated, likely associated with precipitation overestimation over WEIO and poor MISO simulation. Subsequent CAM variants improve this overestimation in ascending motion and underestimation in subsidence. Specifically, CAM6 simulates the Walker circulation closer to observation. Thus, we can speculate that the east-west heat source associated with Walker circulation and the monsoon heat source associated with Hadley circulation, which was poorly simulated in CAM3, have improved in subsequent CAM variants.
3.4.2 Atmospheric Internal dynamics
Figure 9 shows the JJAS mean meridional variation of total precipitation, vertical easterly wind shear (U200-U850; m/s), and specific humidity at 1000 hPa, averaged over the longitudinal domain of 70°-90°E for ERA-I and CAM variants. It is noted that the magnitude of total precipitation is greatly overestimated over 10°S and underestimated over Indian latitudes in CAM3 (Fig. 9a). While in subsequent CAM variants, this overestimation in precipitation over 10°S is improved, the underestimation over Indian latitudes is slightly overestimated. Further, from the vertical easterly wind shear, we find that it is highly underestimated in CAM3 from the southern Indian ocean (10°S) to Indian latitudes (up to 20°N), with the highest underestimation over EIO (Fig. 9b). This underestimation is also improved in subsequent CAM variants, with CAM6 showing the highest improvement. Furthermore, the meridional gradient of specific humidity is simulated comparable to ERA-I in all CAM variants, except an overestimation by 2g/kg across the latitudes from 20°S to Indian land (20°N) in CAM3. However, it is noted that the meridional gradient of specific humidity from ocean to land is improved in subsequent CAM variants, with the highest improvement in CAM6, likely an underlying reason for the large improvement in the northward propagation of MISO (e.g., Jiang et al. 2004; Drbohlav and Wang 2005). Improvements in meridional gradient of specific humidity and hence MISO contribute to improvements in JJAS seasonal mean rainfall in successive CAM variants.