1 Introduction

From the highest peaks in the Himalayas to the largest tropical forests in the Amazon, from the biggest wetlands in Asia to the richest flora and fauna of Northern Australia, and from the grazing lands of Mexico and Africa to the semi-arid rainfed agriculture across all of these continents, the latitudinal oscillation of the Inter Tropical Convergence Zone (ITCZ) between the onset and demise of global monsoons potentially influences every livelihood of natural and human systems. Over two-thirds of the global population and nearly the world’s entire impoverished communities reside within monsoon impacted areas (Beegle et al. 2013), which receive more than half of their annual precipitation during their respective monsoon seasons. More importantly, many of these countries are among the world’s most vulnerable regions where a changing climate is already posing serious challenges. For instance, many of the top ten most impacted countries by extreme weather events are within monsoon regions (Eckstein et al. 2020).

A monsoon system is generally characterized by the seasonal wind reversal in the lower atmosphere (Ramage 1971) and a strong large-scale vertical wind shear as a result of differential heating between continents and oceans (Li and Yanai 1996; Turrent and Cavazos 2009; Webster et al. 1998). However, not every region that receives monsoon precipitation clearly witnesses a seasonal reversal of winds, such as in the South American monsoon, where trade winds dominate during the rainy season (Carvalho and Cavalcanti 2016). As the Earth rotates around the sun on its annual cycle, the ITCZ changes its location from the austral summer in the Southern Hemisphere (SH) to the boreal summer in the Northern Hemisphere (NH). Such a periodicity partially drives the regularity in the arrival and departure of the rainy season over each impacted region. The intensity, duration and spatial distribution of regional monsoon precipitation is further influenced by many regional forcing mechanisms, such as topographic controls on diabatic heating (Li and Yanai 1996; Rodwell and Hoskins 2001), the nature of land–atmosphere-ocean interactions (Ashfaq et al. 2017; Grodsky and Carton 2001; Misra 2008; Sun et al. 2019; Thorncroft et al. 2011; Yasunari 2007), the presence of east–west ocean-basin contrasts (Wang et al. 2003, 2012), and the structure of monsoonal lows (Hurley and Boos 2015). These regional-scale mechanisms interact with large-scale natural forcing, such as the El Niño Southern Oscillation, the Pacific Decadal Oscillation, and the Atlantic Multidecadal Oscillation, at varying timescales, which result in intra-seasonal to decadal variations in monsoon circulation and precipitation patterns (e.g. An et al. 2015; Webster et al. 1998).

In recent decades, the global monsoon system has witnessed heterogenous trends in the timing, strength and seasonal precipitation magnitudes at regional scales, including an earlier onset over South Asia and Southeast Asia (Bollasina et al. 2014; Bridhikitti 2019), increasing precipitation trends over Australia (Gallego et al. 2017), a delayed onset over South America (Marengo et al. 2011), and declining precipitation trends over Africa, East Asia, and North America (Cavazos et al. 2020; Zhan et al. 2018; Zhu et al. 2011). While changes over the American monsoons are partially attributed to the natural variability of the climate system (Arias et al. 2012; Castro et al. 2001; Cavazos et al. 2020), anthropogenic aerosols are considered one of the major drivers of recent fluctuations in Asian monsoons (Ganguly et al. 2012; Jiang et al. 2013; Polson et al. 2014). Similarly, drought and recovery of the West African monsoon has been attributed to aerosols and greenhouse gas concentrations (Biasutti 2019; Dong and Sutton 2015). However, a robust understanding towards the causes of variations in regional monsoon systems remain challenging due to uncertainties in the observations and deficiencies in the current generation of General Circulation Models (GCMs) (Biasutti 2019; Pascale et al. 2017; Singh et al. 2019).

GCMs have considerably improved over the years in the representation of global monsoon systems. In spite of this, significant inconsistences still exist at regional scales over a number of monsoon regions. For instance, most of the GCMs that are part of the 5th phase of the Coupled Models Inter-comparison Project (CMIP5; Taylor et al. 2012) are unable to accurately simulate the monsoon onset, the frequency and trajectory of monsoon depressions, and the spatial distribution of seasonal precipitation over the South Asian monsoon region (e.g. Ashfaq et al. 2017; Rastogi et al. 2018). Biases exist in GCM simulations of the monsoon onset over Australia (Zhang 2010), the monsoon retreat over the North American monsoon region (Cook and Seager 2013; Geil et al. 2013), the monsoon timing and variability over West Africa (Dunning et al. 2017; Roehrig et al. 2013), and the spatial precipitation distribution over the African, South American and the East and Southeast Asian monsoon regions (Mariotti et al. 2014; Sperber et al. 2013; Yin et al. 2013). Given the challenges in the simulation of regional-scale monsoon characteristics, CMIP5 models exhibit variable confidence in the simulation of future monsoon responses at regional scales (Christensen et al. 2013). While global monsoons as an aggregate are likely to strengthen in terms of precipitation intensity and weaken in terms of circulation strength (Kitoh et al. 2013; Lee and Wang 2014), there is medium to low confidence in the precipitation amounts over the American and Asian monsoons (Bukovsky et al. 2013; Christensen et al. 2013; Colorado-Ruiz et al. 2018; Pascale et al. 2017, 2019) and in the timing of the monsoon onset over the North American and West African monsoon regions (Christensen et al. 2013; Diallo et al. 2014; Mariotti et al. 2014; Pascale et al. 2019).

The growing acknowledgement that monsoon climates are changing due to natural and anthropogenic influences has produced a mounting demand for regional climate change information over tropical regions. Moreover, within the context of negotiations to limit global warming to a 2 °C warming level, it is imperative to understand regional monsoon responses at various levels of increases in radiative forcing. Presently, due to the limitations in the representation of regional monsoons in GCM simulations (Ashfaq et al. 2017; Biasutti et al. 2018; Pascale et al. 2019; Sperber et al. 2013), regional and local-scale studies for understanding the impacts of changing monsoonal characteristics on natural and human systems increasingly rely on the dynamical downscaling of GCMs through their use as initial and boundary forcing in Regional Climate Models (RCMs; Giorgi and Gutowski 2015). There are numerous studies that demonstrate the added value of RCMs in the simulation of seasonal mean precipitation (Fotso-Nguemo et al. 2017; Giorgi 2019; Mariotti et al. 2014; Racherla et al. 2012; Torma et al. 2015), the onset and demise over monsoon regions (Ashfaq et al. 2009; Diallo et al. 2014; Sylla et al. 2013), and the intensity of hot and wet extremes (Li 2017; Qiu et al. 2019), due to their finer grid spacing and their flexibility of region-specific parametrization tuning.

In recent decades, RCMs have been extensively utilized to regionally refine GCM projections over terrestrial regions, particularly within the framework of the Coordinated Regional Downscaling Experiment (CORDEX; Giorgi et al. 2009). Among these RCMs, the International Centre for Theoretical Physics (ICTP) Regional Climate Model (RegCM) is one of the most widely used models for climate change studies (Ambrizzi et al. 2019; Ashfaq et al. 2009, 2016; Bukovsky et al. 2013; Diallo et al. 2016; Diffenbaugh et al. 2011; Fuentes-Franco et al. 2014; Gao et al. 2017; Giorgi et al. 2012; Mariotti et al. 2014; Martinez-Castro et al. 2006; Reboita et al. 2014; Sylla et al. 2015).

Recently, the version 4 of the RegCM system, RegCM4 (Giorgi et al. 2012) has been used to produce a set of regional projections over nine CORDEX domains at 25 km grid spacing under the framework of the CORDEX-CORE initiative (Giorgi and Gutowski 2015). In this study, we use this unprecedented high-resolution ensemble of simulations over seven of the nine CORDEX domains that include all major monsoon regions of the world under two Representative Concentration Pathways (RCP2.6 and RCP8.5) to investigate regional monsoon responses to increases in radiative forcing during the twenty-first century. Our analyses focus on seasonal characteristics of regional monsoons, their expected changes in onset and demise of precipitation, and on the commonality of regional monsoon responses across multiple domains. Furthermore, our comparison of simulations driven with the lowest (RCP2.6) and highest (RCP8.5) concentration pathways provide an insight into the consequences should greenhouse gas concentrations continue to escalate along their current trajectory, as well as the benefits of emission stabilization.

2 Data and methods

2.1 Datasets

We utilize dynamically downscaled simulations over seven CORDEX domains: Africa, Australasia, East Asia, Central America, South America, South Asia, and Southeast Asia (Fig. 1). For every region, we employ ICTP RegCM4 to downscale three CMIP5 GCMs that exhibit adequate region-specific skills. Table 1 shows the details of RegCM4 domains, selected parameterizations, and driving GCMs over each region. Each RegCM4 configuration utilizes 25 km horizontal grid spacing and 23 levels in the vertical over a domain that follows the latitudinal and longitudinal extent recommended by the CORDEX-CORE initiative. For GCM downscaling, RegCM4 simulations are conducted in a transient mode from 1970 to 2100 with annually varying greenhouse gas (GHG) forcing. For the historical period (1970–2005), RegCM4 uses observed GHG forcing. For the future period (2006–2100), RegCM4 is forced with projected GHG forcing RCP2.6 and RCP8.5, which represent lower and higher end radiative forcing scenarios, respectively (van Vuuren et al. 2011). Both future period integrations are initialized using restart data from the last time step of the historical period integration.

Fig. 1
figure 1

Different RegCM CORDEX domains used in the analyses. The colored land area within each domain reflects the region that has been used from each RegCM integration for spatial plots. Boxes represent areas used for various zonal average analyses

Table 1 RegCM4 configuration over various CORDEX domains

For model comparisons with observations and reanalysis, we use (1) 1° daily precipitation from the Global Precipitation Climatology Project (GPCP) version 1.2 (Huffman et al. 2016), (2) monthly precipitation from 0.5° Climate Research Unit Timeseries (CRU TS) version 4.03 (Harris et al. 2014), and (3) monthly atmospheric divergence, winds and temperatures from the 0.25° European Reanalysis 5° (ERA5) reanalysis data (C3S 2017).

2.2 Methods

We utilize the Feng et al. (2013) analysis framework to calculate a dimensionless seasonality index (hereafter seasonality), relative entropy, and the timing of centroid at each grid point. For each hydrological year (October–September), we calculate the relative entropy at each grid point using the following expression:

$${RE}_{y}=\sum_{m=1}^{12}{Pb}_{y,m}{log}_{2}\left[\frac{{Pb}_{y,m}}{{q}_{m}}\right]$$
(1)

where \({Pb}_{y,m}\) represents the precipitation probability in a month m of year y, calculated as the fraction of annual precipitation in year y falling in the month m. \({q}_{m}\) represents the uniform distribution and has value of \(\frac{1}{12}\) for all months. Relative entropy provides a measure of distance between the simulated monthly precipitation and the uniform distribution. Higher values of entropy suggest the non-uniformity of monthly precipitation in a given year, implying that precipitation is more distributed around the wet season. Subsequently, the seasonality of precipitation is calculated as the product of relative entropy and the annual mean precipitation. In order for the results to be comparable across different CORDEX domains, datasets (observations, model simulations), and simulation periods (historical, future), we normalize the results using the maximum precipitation (\({P}_{max})\) found over all datasets and simulation periods. Therefore, the expression for seasonality can be written as:

$${S}_{y}={RE}_{y} \times \frac{{P}_{y}}{{P}_{max}}$$
(2)

The maximum value of seasonality can be reached if a grid point receives \({P}_{max}\) in a single month, which implies that regions with low annual precipitation will exhibit low seasonality even if the precipitation distribution across the months is highly nonuniform. Furthermore, we calculate the timing of centroid using the first moment of monthly mean precipitation, which corresponds to the timing when 50% of the annual precipitation is reached in a hydrological year. Further details of the methodology and mathematical expressions can be referred to Feng et al. (2013).

Additionally, we calculate the monsoon onset pentad (5-day average) at each grid point following a methodology adopted from Bombardi and Carvalho (2009). For each analysis time period, we use the climatological pentad time series at each grid point to calculate the sum using the following expression:

$$S\left( {pentad} \right) = \sum\limits_{{n = pentad_{1} }}^{{pentad}} {\left( {P_{n} - \mathop P\limits^{ - } } \right)}$$
(3)

where \(\stackrel{-}{P}\) represents the climatological annual mean and \({P}_{n}\) represents the nth pentad of precipitation. Subsequently, the monsoon onset and monsoon demise at each grid point are defined as the pentad after the minimum S and the pentad after the maximum S, respectively. Our approach is relatively simple and slightly different when compared to Bombardi and Carvalho (2009), yet it yields similar results when applied to the climatological data.

We utilize 1995–2014 as the reference historical period for comparisons with observations and reanalysis, and for the calculation of future changes in the twenty-first century. The reference period during 2006–2014 is taken from the RCP8.5 simulations. There is not much difference between RCP8.5 and RCP2.6 radiative forcing, and corresponding RegCM4 simulations for this time period. Therefore, no noticeable impact on future changes are expected due to this methodological choice. We use two 20-year time slices from the RCP2.6 and RCP8.5 model integrations to calculate changes in the mid-century (2041–2060) and the late century (2080–2099) period. All results are shown as an ensemble mean of three ensemble members over each domain while the robustness of future changes is being tested using the baseline variability as a threshold at each grid point. The late century period (2080–2099) changes over Africa are based on the downscaling of only two GCMs (HadGEM2-ES and NorESM1-M, Table 1), since the third ensemble member was not available at the time of analyses.

3 Results

3.1 Historical evaluation of monsoon regions

Despite regional differences among monsoon regions, such as the timing of monsoon onset and demise, the magnitude of rainy season precipitation, and the percent contribution of monsoonal rains to each annual mean, every monsoon region shares the fundamental characteristic of an uneven distribution of precipitation throughout the year. Therefore, monsoon regions are expected to display a strong seasonality in precipitation distribution. Given this fact, we analyze the seasonal characteristics of precipitation, including seasonality, relative entropy, and the timing of centroid, over each monsoon domain, both in the observations and the RegCM4 ensemble mean, during the historical period (1995–2014; Fig. 2). As expected, every monsoon region displays high values of seasonality in the observations (Fig. 2a), with the largest values over West Africa, South Asia, and parts of the Amazon. These large values of seasonality are mainly driven by the large fraction of annual precipitation during the monsoon season, which is reflected in the high magnitudes of relative entropy (Fig. 2c).

Fig. 2
figure 2

Comparison between the CRU observations and the RegCM ensemble mean during the 1995–2014 period. Seasonality (a, b), relative entropy (c, d), timing of centroid (e, f), peak season precipitation (g, h). Values are masked in the bottom three rows where seasonality is < 0.025

The timing of centroid occurs mostly around February in the austral summer over the monsoon regions in the SH and around July in the boreal summer over the monsoon regions in the NH, which generally corresponds to peak monsoon precipitation over each region. However, each monsoon region shows spatial variations in the timing of centroid (Fig. 2e), suggesting that identical periods cannot be fully representative of the peak precipitation season over a monsoon region. Therefore, instead of using a fixed period as the core rainy season over each hemisphere, we define a three-month long period as the peak of the monsoon season over each grid point spanning from a month before the peak precipitation month to the month after the peak precipitation month in the observations. Precipitation during these three months reflects monsoon precipitation over each grid point (Fig. 2g).

The RegCM4 ensemble simulates the smooth latitudinal transition of the monsoon seasons from the SH to the NH and vice versa in the Americas, Africa, and from Australia to Asia (Fig. 2f). In all cases, the RegCM4 captures the spatial patterns of the four measures of monsoon season characteristics effectively. However, there are a few exceptions where the RegCM4 simulated magnitudes differ from the observed. For instance, the RegCM4 ensemble overestimates (underestimates) seasonality (relative entropy) over higher elevations, such as the Himalayas in Asia, the Guinea Highlands in West Africa and the Andes in South America. This can be attributed to the model tendency to produce excessive convectively driven orographic precipitation. However, it should be noted that CRU may also have deficiencies over mountainous regions due to the lack of high elevation stations in those areas (Cavazos et al. 2020; New et al. 1999; Nijssen et al. 2001; Sylla et al. 2013). Relative entropy may exhibit biases if the model exhibits wet or dry biases during the monsoon season, such as lower relative entropy and drier than observed seasonal precipitation over parts of South Asia and Australia and higher relative entropy and wetter than observed seasonal precipitation over parts of South America. Additionally, South Asia exhibits earlier than observed timing of centroid over some areas, which reflects an overall dry bias in annual precipitation.

We also compare the pentad timing of the monsoon onset and demise over each monsoon region (Fig. 3), which are generally complementary to the timing of centroid (Fig. 2e–f). The separation of the NH and SH summer monsoon regions are clearly seen in the timing of centroid (Fig. 2e–f), and in the timing of monsoon onset and demise (Fig. 3), which is simulated well in the RegCM4 ensemble mean. The earliest monsoon onset in the SH is found over parts of the Amazon, and Southern Africa in early October, which gradually progresses eastward to cover entire monsoon regions by mid-December. The earliest monsoon demise in the SH appears over parts of Australia in late February, and by the beginning of May, the rainy season fully retreats from the SH well before the start of the austral winter. This timing coincides with the earliest monsoon onset over the NH monsoon regions in Central America, Africa and Asia. By mid-July, the rainy season establishes over every NH monsoon region, with a full retreat by October (Fig. 3a, c).

Fig. 3
figure 3

Monsoon onset and demise over land in the reference period (1995–2014) in the GPCP Observations (a, c), and the RegCM ensemble (b, d). White land areas are masked where observed seasonality is < 0.025. Results are not meaningful outside the monsoon regions

The RegCM4 ensemble mean generally captures the latitudinal transition of the global monsoons. The most notable bias is found over South America, where parts of the Amazon exhibit onset dates later than observed, and over parts of South Asia, Southern Africa and Sahel where the RegCM4 ensemble exhibits an earlier onset than the observed. In addition, the RegCM4 ensemble mean displays a later than observed demise over parts of Australia and South Asia and an earlier than observed demise over parts of South and Central America (Fig. 3). Nonetheless, the RegCM4 simulations exhibit a reasonable overall skill in representing the characteristics of the monsoon during the boreal and austral summers. Moreover, the observed and simulated onset and demise dates using GPCP and RegCM4 regional ensembles are within the uncertainties of other datasets based on similar methodologies (Bombardi and Carvalho 2009).

3.2 Future changes in the monsoon regions

Figure 4 shows mid-century (2041–2060) changes in the monsoon metrics shown in Fig. 2. With the exception of a few parts of East Asia and West Africa, the RCP2.6 ensemble exhibits very small changes in seasonality (Fig. 4a). On the other hand, the RCP8.5 ensemble exhibits enhanced seasonality over a few regions, such as the Western Amazon, Eastern Sahel, parts of Central Africa, the Himalayas, and parts of East Asia. This enhancement is greater than the baseline variability over parts of Africa and the Amazon (Fig. 4b). The increase in seasonality is mostly driven by an increase in relative entropy, which exhibits a robust increase (> reference period variability) over parts of Sahel in both scenarios, and over parts of South America and Central Southern Africa (Kalahari Desert and neighboring countries) in the RCP8.5 (Fig. 4c, d). A higher fraction of annual precipitation during the rainy season (i.e. an increase in relative entropy) results in a forward shift in the timing of the centroid (up to 15 days later than in the reference period) over some regions (Fig. 4e, f).

Fig. 4
figure 4

Simulated projected changes in the near-term future (2041–2060) with respect to the 1995–2014 period under RCP2.6 and RCP8.5. Seasonality (a, b), entropy (c, d), timing of centroid (e, f), peak season precipitation (g, h). Stippling represents those regions where projected changes are greater than the reference period variability. White land areas are masked where observed seasonality is < 0.025

Most of the monsoon regions show increased precipitation (1–2 mm/day) during the peak season in both scenarios with exceptions in parts of the Amazon, West Africa and South Asia under the RCP8.5 scenario. Changes over East Asia are generally robust in the RCP2.6 scenario, while changes in the other monsoon regions are greater than the baseline variability only in the RCP8.5 scenario (Fig. 4d, h). We also note an earlier timing of centroid over the regions where monsoon precipitation decreases, e.g. over parts of South Asia, West Sahel and East Africa (Fig. 4e, f).

Stabilization of radiative forcing during the latter half of the twenty-first century in the RCP2.6 results in no major changes in the late-century (2080–2099) monsoon characteristics when compared to those in the mid-century (left columns in Figs. 4, 5), with the exception of parts of the Australian monsoon where precipitation exhibits a robust decrease in the late-century (Fig. 5g). In contrast, in the RCP8.5 ensemble, changes in the monsoon characteristics exhibit similar patterns of change but substantially strong magnitudes in the late-century relative to the mid-century (right columns in Figs. 4, 5). For instance, changes in seasonality become robust over most of the South America, West Africa, and Southeast Asia monsoon areas. Similarly, increases in the relative entropy are robust over monsoon regions in Africa and South America. Likewise, magnitudes of precipitation change nearly double everywhere (> 3 mm per day) during the peak season, which includes robust decreases over the West Sahel and Gulf of Guinea in West Africa, Southeast Africa, South Asia, North America and the Eastern Amazon, and robust increases over Central Africa, the Western Amazon, and parts of Southeast Asia, East Asia, Eastern Sahel and Northern Australia (Figs. 4h, 5h). With the exception of Australia, changes in the timing of centroid also exhibit shifts similar to the mid-century period but nearly double in magnitude (up to 30 days).

Fig. 5
figure 5

Simulated projected changes in the far future (2080–2099) with respect to the 1995–2014 period under RCP2.6 and RCP8.5. Seasonality (a, b), entropy (c, d), timing of centroid (e, f), peak season precipitation (g, h). Stippling represents those regions where projected changes are greater than the reference period variability. White land areas are masked where observed seasonality is < 0.025

We note a robust tendency towards a late arrival of monsoons in response to an increase in GHG forcing, as illustrated by a late arrival of the earliest onset everywhere, except over parts of the East and Southeast Asian and North American monsoon regions (Figs. 6a–d, 9, 10). This shift in the start of the rainy season is progressive as higher GHG forcing levels result in further delays. The delay in the timing of the onset is visible in both future scenarios; however, it is spatially more robust in the RCP8.5, which exhibits magnitudes > 3 pentad shifts over most areas by the mid-century and > 6 pentads by the late century. The strongest delays in the start of the rainy season are found over the Sahel and parts of South Asia (> 8 pentads). A spatially robust delay in the monsoon demise is also exhibited over South Asia and West Africa in the RCP8.5 in both time periods (~ 3 to 5 pentads), however, other regions show mixed responses (Fig. 6e–h). For instance, a later demise of the North American monsoon only becomes spatially robust in the late century under the RCP8.5. Similarly, some areas over the Northern Amazon in South America and Australia (Southeast Asia and South Africa) that exhibit a delayed (earlier) demise during the mid-century period shift to an earlier (delayed) demise in the late century period. Nonetheless, we can conclude that monsoon regions in general exhibit a prevailing shrinking of the monsoon season as the delays in the monsoon onset, in most cases, are more pronounced than the delays in the monsoon demise.

Fig. 6
figure 6

Simulated projected changes in the monsoon onset (ad) and demise (eh) in the RCP2.6 (left column) and the RCP8.5 (right column). Changes are shown for the near-term (2041–2060) and the long-term (2080–2099) with respect to 1995–2014. White land areas are masked where observed seasonality is < 0.025. Results are not meaningful outside the monsoon regions

4 Discussion

4.1 Drivers of projected changes in regional monsoons

A holistic review of intra-annual land-atmospheric characteristics is required to identify the drivers of projected changes in regional monsoons. Complex land-atmospheric processes prepare a monsoon region for the transition to the rainy season while the ITCZ inter-hemispherically oscillates towards warmer continents. Several months before the start of the rainy season (pre pre-monsoon), each monsoon region exhibits a stable atmosphere and dry conditions (Figs. 7, 8; first column), with deep atmospheric subsidence in the zonally averaged wind vectors of divergence and vertical velocity, and low magnitudes of zonally averaged precipitation (Figs. 9 and 10). Around the time of monsoon demise in one hemisphere, land begins to warm in the other, which initiates shallow overturning in the lower troposphere (Figs. 7, 8; second column) through sensible heating. During that time, surface temperatures reach the annual maximum, which drives the development of heat lows before the monsoon onset. Every monsoon region receives a noticeable amount of pre-monsoon precipitation (Figs. 11a), which helps to warm up the upper troposphere (Fig. 12) and induce deep overturning through latent heat release in the atmosphere (Figs. 7, 8; third column). Pre-monsoon diabatic heating thus prepares the region to transition into a highly conducive regime for atmospheric convection during the monsoon onset.

Fig. 7
figure 7

Latitude-height cross-section of divergence (1/s) and vertical pressure velocity (Pa/s) in ERA5 (left three columns) and RegCM (right three columns), shown as wind vectors averaged over West Africa (a1a6; 15 W–15 E, 2–20 N), South Asia (b1b6; 70–95 E, 5–35 N), South America (c1c6; 60–40 W, 0–30 S), and Southeast Africa (d1d6; 15–45 E, 5–30 S). Colored contours represent vertical pressure velocity. Black arrows represent the direction of the monsoon along the latitude. Both vertical pressure velocity and divergence have been exaggerated by 50 and 106 in all plots. Vertical pressure velocity is multiplied by − 1 so that positive values represent an upward motion. The three panels in each case represent averages over pre pre-monsoon, pre-monsoon and monsoon periods. The Pre-monsoon period represents the average over the two months prior to the monsoon season. The pre pre-monsoon period represents the average over two months prior to the pre-monsoon period

Fig. 8
figure 8

Same as in Fig. 7 but over Australia (a1a6; 130–150 E, 10–20 S), East Africa (b1b6; 32–50 E, 2–20 N), East Asia (c1c6; 110–140 E, 20–40 N), Southeast Asia (d1d6; 95–110 E, 10–25 N), and North America (e1e6; 114–104 W, 22–40 N)

Fig. 9
figure 9

Zonally averaged 5-day mean precipitation (1995–2014) in the GPCP observations (first column) and in the RegCM simulations (second column) over West Africa (a, b; 15 W–15 E, 2–20 N), South Asia (e, f; 70–95 E, 5–35 N), South America (i, j; 80–40 W, 10 N–35 S), and Southeast Africa (m, n; 15–45 E, 5–30 S). Mid-term (2041–2060) and long-term (2080–2099) future changes in zonal average 5-day mean precipitation under RCP8.5. Africa (c, d), South Asia (g, h), South America (k, l), and South Africa (o, p). The vertical lines in each panel represent the approximate climatological timing (in pentads) by when the region receives earliest monsoon precipitation in the observations as shown in Fig. 3

Fig. 10
figure 10

Same as in Fig. 9 but for Australia (ad; 130–150 E, 10–20 S), East Africa (eh; 32–50 E, 2–20 N), East Asia (il; 110–140 E, 20–40 N), Southeast Asia (mp; 95–110 E, 10–25 N), and North America (qt; 114–104 W, 22–40 N)

Fig. 11
figure 11

Pre-monsoon precipitation during 1995–2014 in a CRU observations, b RegCM ensemble. Projected changes w.r.t. 1995–2014 in the pre-monsoon c, d total precipitation, and e, f convective precipitation during c, e 2041–2060 and d, f 2080–2099 under RCP8.5. The pre-monsoon precipitation represents the average of the two months before the peak monsoon season at each grid point

Fig. 12
figure 12

Changes (2080–2099 minus 1995–2014) in the latitude-pressure cross-section of pre-monsoon relative humidity (%) under RCP8.5 over a West Africa, b South Asia, c South America, d Southeast Africa, e Australia, f East Africa, g East Asia, h Southeast Asia, and i North America. The colored lines in each panel represent atmospheric boundary layer height in 1995–2014 (blue) and 2080–2099 (red). The height of atmospheric boundary layer is represented in terms of corresponding atmospheric pressure. All domains are identical to the ones used in Figs. 9 and 10. The pre-monsoon period represents the average over two months prior to the monsoon period

The latent heat, driven by atmospheric warming from the pre-monsoon to the monsoon phase, can be explained through the monthly anomalies of zonally averaged upper-tropospheric (500–200 mb) temperatures with respect to annual means (Figs. 12, 13; first column). These figures demonstrate the meridional differential heating in the upper troposphere that is responsible for large-scale atmospheric shear during the monsoon season and precipitation progression along the summer hemisphere latitudes (Held and Hou 1980; Lindzen and Hou 1988). Zonally averaged upper tropospheric temperatures over each monsoon region reaches higher than the annual mean before the earliest monsoon onset (Figs. 12, 13; first column, zero-line contours). Intense latent heating of the atmosphere during the monsoon is reached at the peak of the season, which coincides with the maxima of seasonal precipitation (Figs. 9, 10; first column), meridional differential heating (Figs. 12, 13; first columns), and deep vertical ascending motion (Figs. 7, 8; third column). We note that the inter-seasonal overturning of atmospheric circulations (Figs. 7, 8; last three columns), the distribution of pre-monsoon precipitation (Figs. 9, 10; second column) and the meridional differential heating (Figs. 12, 13; second column) are represented quite well in the RegCM4 ensemble over each monsoon region. The most noticeable systematic biases occur over South and East Asia, and South America, where the model exhibits weaker monsoon circulations, which in turn explain the dry biases over parts of these regions.

Fig. 13
figure 13

Departure of zonally averaged monthly upper tropospheric temperatures from the annual mean over West Africa (ac), South Asia (df), South America (gi) and Southeast Africa (jl). The first and second column show comparisons between ERA5 and RegCM in the reference period (1995–2014). The third column shows the projected changes in the late-century (2080–2099) period under RCP8.5 with respect to the reference period. All domains are identical to the ones used in Fig. 7, except for West Africa, which extends further down to 5S. Black contours represent the zero line. The grey dotted line represents approximate timing of the earliest onset over land and the black arrows represent the direction of the monsoon progression along the latitude over each region

Given the progressive nature of simulated changes in regional monsoon characteristics as a function of GHG forcing (Figs. 4, 5, 6), we restrict our focus only to the late twenty-first century changes under the RCP8.5 scenario to highlight the potential mechanisms responsible for the projected variations in monsoon onset and precipitation characteristics. Because the magnitudes of changes are highest during the late twenty-first century period under RCP8.5 (Figs. 5, 6), we expect that clear causal connections can be established between precipitation responses and variations in the underlying physical processes.

The projected delay in the monsoon onset is deeply rooted in pre-monsoon precipitation dynamics. With the exception of East and Southeast Asia, the two regions where the monsoon onset does not exhibit a spatially robust delay, changes in zonally averaged precipitation exhibit a decrease in every region (Figs. 9, 10; right two columns). Spatially, this pre-monsoon decrease corresponds very well with the changes in monsoon onset (Figs. 6d, 11d). The observed pre-monsoon precipitation is mostly convective in nature over the monsoon regions, in contrast to the mature monsoon season where the fraction of stratiform precipitation also increases due to the presence of organized precipitation systems (Schumacher and Houze 2003). The predominance of convective precipitation during the pre-monsoon season also occurs in our simulations, given the remarkable resemblance between the simulated future changes in the pre-monsoon total and convective precipitation (Fig. 11c–f). We note that those monsoon regions where convective precipitation is suppressed during the pre-monsoon also exhibit an increase in the height, or depth, of the atmospheric boundary layer and a decrease in relative humidity, or the saturation level, in the lower troposphere (Fig. 12). In present climate conditions, the convective environments over the monsoon regions during the pre-monsoon exhibit a sharp contrast from their respective monsoon seasons, with a relatively deeper atmospheric boundary layer, highly variable atmospheric moisture and a predominant thermally driven convection due to warmer land surface and limited soil moisture conditions (Thomas et al. 2018). Two potentially unfavorable developments in the pre-monsoon convective environments take place in the future climate conditions. First, as the land surface warms in response to future increases in radiative forcing, the atmospheric boundary layer grows even deeper due to a greater partitioning of the surface energy flux towards sensible heating. A deeper atmospheric boundary layer requires an additional buoyancy force to lift air parcels to their level of free convection. Second, future warming of the overlying atmosphere increases the required amount of moisture needed for the atmospheric boundary layer to become convectively unstable (Chou and Neelin 2004; Langenbrunner et al. 2019). While the atmosphere is expected to be more moist in future climates, the additional moisture requirement for convective initiation is challenging to satisfy during the pre-monsoon due to the limited moisture supply as winds predominantly blow from the dry land regions—an upped-ante mechanism that leads to reduced precipitation (Chou and Neelin 2004; Neelin et al. 2003). Neelin et al. (2003) suggest that warming induced by the increase in greenhouse gas forcing raises the amount of moist static energy requirements for convective instability in the atmospheric boundary layer, which is analogous to the game of poker where stakes get higher when a player ups the ante. This additional moisture requirement is difficult to fulfill in subsidence zones where atmospheric flow in the lower levels is from non-precipitating regions. Overall, these unfavorable developments in the future period increase atmospheric stability and reduce convective precipitation during the pre-monsoon months (Figs. 11, 12).

Below normal pre-monsoon precipitation in the future period drives a potential decrease in tropospheric latent heat release, which is displayed by a shift in the warming pattern of the upper troposphere in the future period (Figs. 13, 14; right column). The lack of latent heat driven diabatic heating delays the deep overturning of atmospheric subsidence during the pre-monsoon phase, particularly over West, East and Southeast Africa, South Asia and South America, which exhibit strong delays in the monsoon onset (Fig. 15). It should be noted that each of these regions exhibit a strengthening of shallow overturning in the lower atmosphere over land (Fig. 15; positive values in the lower atmospheric levels), which is indicative of an increase in sensible heating due to warmer surface temperatures in the future period. However, increases in sensible heating alone are unable to compensate the diabatic heating deficit in the upper troposphere, which are caused by decreased pre-monsoon precipitation. Similar mechanisms have been shown to cause a delayed onset bias over South Asia in the CMIP5 GCMs (Ashfaq et al. 2017). Such an anomalous deep subsidence is not visible over East and Southeast Asia, and North America due to the lack of a clear spatial pattern of a delay in the monsoon onset (Figs. 4d, 15). On the other hand, an increase in pre-monsoon precipitation may cause a decrease in the strength of sensible heat driven heat lows, which is reflected as an anomalous sinking motion in the lower levels, such as over East and Southeast Asia and East Africa (Fig. 15f1–g1). Weaker than normal precipitation during the monsoon season is reflected in the anomalous subsidence during the monsoon season, which is more prominent over parts of West and Southeast Africa, South Asia and South America (Fig. 15; brown color). A substantial late season warming in the upper troposphere, such as over the North American monsoon region (Fig. 10f), is indicative of a robust delay in the monsoon demise (Fig. 6h).

Fig. 14
figure 14

Same as in Fig. 13, but for Australia (ac), East Africa (df), East Asia (gi), Southeast Asia (jl) and North America (mf). All domains are identical to the ones used in Fig. 7, except for Australia, which extends further north to the equator, and North America, which extends further down to 10 N. The black contours represent the zero line. The grey dotted line represents the approximate timing of the earliest onset over land and the black arrows represent the direction of the monsoon progression along the latitude over each region

Fig. 15
figure 15

Changes in latitude-height cross-section of divergence (1/s) and vertical pressure velocity (Pa/s), shown as wind vectors averaged over latitudes used in Fig. 7 and 8. West Africa (a1, a2), South Asia (b1, b2), South America (c1, c2), and Southeast Africa (d1, d2), Australia (e1, e2), East Africa (f1, f2), East Asia (g1, g2), Southeast Asia (hh2) and North America (i1, i2). Colored contours represent the vertical pressure velocity. Both vertical pressure velocity and divergence have been exaggerated by 50 and 106 in all plots. Vertical pressure velocity is multiplied by − 1 so that positive values represent an upward motion. The two panels in each case represent the averages over the pre-monsoon and monsoon periods. The pre-monsoon period represents the average over two months prior to the monsoon season

The delay in monsoon onset and demise are not the only major changes in regional monsoon characteristics. Rainy seasons across the monsoonal belt exhibit shrinking in response to higher levels of GHG forcing. The increase in precipitation over parts of the monsoon regions (Fig. 5g, h) in connection with a shorter length of the rainy season points towards an increase in the intensity of the monsoon rains and a decrease in the fraction of annual precipitation outside of the monsoon seasons (Fig. 5d).

Many of the findings in our analyses are supported by earlier studies over a number of regional monsoons. The late twenty-first century precipitation increases over many areas are consistent with the CMIP5 multi-model analyses (e.g. Hsu et al. 2012; Kitoh et al. 2013). A potential delay in the monsoon onset over South Asia and parts of Africa has been reported previously by high-resolution RCM based studies (Akumaga and Tarhule 2018; Ashfaq et al. 2009; Diallo et al. 2016; Kumi and Abiodun 2018; Mariotti et al. 2014). Similarly, the lack of evaporation, the changes in the atmospheric boundary layer structure and the decrease in moisture convergence are linked to pre-monsoon drying in the future projections of CMIP5 models or in enhanced CO2 experiments (Langenbrunner et al. 2019; Seth et al. 2013). Likewise, earlier studies support a late demise over North America (Colorado-Ruiz et al. 2018; Cook and Seager 2013; Seth et al. 2013), and a decrease in the meridional tropospheric thermal contrast over South Asia and North America (Ashfaq et al. 2009; Mei et al. 2015; Torres-Alavez et al. 2014). Moreover, dipolar latitudinal changes in precipitation over South America (where dry conditions persist north of the Equator and wet conditions to the south) are consistent not only with previous modeling studies using GCMs (Jones and Carvalho 2013) and RCMs (Llopart et al. 2019; Reboita et al. 2014), but also in the observations (Sena et al. 2018).

4.2 Implications of changes in regional monsoons

The projected variations in the onset, duration, and demise of the monsoons and the resulting changes in seasonal precipitation have strong implications for regional planning, especially for energy, health, agricultural and water resource sectors. For instance, 75% of energy production in Brazil is through hydropower, where a projected shift in rainfall timing will have a detrimental impact on the adequacy of energy production during certain times of the year (de Queiroz et al. 2019). Additionally, hydropower is a significant source of energy in India and Pakistan, where significant delays in monsoon onset and decreases in summer precipitation is projected. Similarly, agriculture has a major footprint on the human lives and economies of Asian, African and American monsoon regions. Economic impacts on agricultural production due to monsoon variations in these monsoon regions can be seen throughout the world (Rojas et al. 2019). As an example, over half of the global arabica coffee supply is produced in Brazil and over 70% of cacao production, the ingredient used to make chocolate, comes from Ghana and the Ivory Coast in West Africa. Both coffee and cacao are highly rain dependent and during times without rain, such as those that occur during a delayed monsoon onset or drought conditions, result in decreased yields and increased prices (de Camargo 2010; Schroth et al. 2016). Likewise, a delay in monsoon onset and/or variations in rainy season precipitation has been a known source of vector borne diseases (e.g. cholera, malaria, dengue) in Africa, South Asia and South America (Benitez 2009; Dash et al. 2013; Dhiman et al. 2010). Extreme weather events are already on the rise in many countries across the monsoon belt (Zhang and Zhou 2019). A shrinking of the rainy season and strong wet and dry projected variations in precipitation are indicative of further strengthening of the prevailing trends under enhanced GHG forcing, which may push highly vulnerable socioecological systems in the monsoon impacted developing world beyond their elasticity.

The implications of projected changes in global monsoons at higher levels of radiative forcing under the RCP8.5 are unprecedented. However, our analyses demonstrate the possibility that a substantial change in global monsoons can be avoided under the RCP2.6, highlighting the urgent need for steps towards emissions stabilization. Unfortunately, observations suggest that we may already be running out of time as the maximum observed CO2 levels in 2019 (415.7 ppm; https://www.co2.earth/daily-co2) were comparable to the annual average CO2 levels in the RCP8.5 (412.8 ppm), and the maximum radiative forcing levels in the RCP2.6 will be reached before the 2030s in the RCP8.5 (Meinshausen et al. 2011). Nonetheless, significant deviations from the current emission trajectories should still be able to avoid major catastrophes.

5 Summary

In this paper, we have provided, for the first time, a global view of changes in monsoon characteristics based on a large ensemble of high-resolution RegCM4 experiments for two different GHG forcing scenarios. In total, we analyzed nine regional monsoons, including three in the SH (Australia, Southeast Africa, and South America) and six in the NH (East Asia, Southeast Asia, South Asia, East Africa, West Africa, and North America). The evaluations of the RegCM4 ensemble mean demonstrates the ability of the model to reasonably reproduce the inter-hemispheric transition of the monsoon seasons and the evolution of the seasonal monsoon characteristics in different regions.

Monsoon systems around the world are projected to experience unprecedented changes in monsoon precipitation characteristics, including shrinking of the rainy season, delays in monsoon onset and demise, and substantial changes in the magnitude of seasonal precipitation, especially under the high-end RCP8.5 scenario. For this scenario, most of the projected changes become spatially robust and greater than the baseline variability towards the end of the twenty-first century. For instance, presently, the monsoon onset in the SH varies from late October over South America to late November over Southeast Africa and mid-December over Australia. In our simulations the monsoon onset is projected to be delayed by two to eight pentads, with the strongest delay over South America and the weakest over Australia. Similarly, presently, the monsoon onset in the NH varies from early May over East Asia and Southeast Asia to early June over the rest of the NH monsoon regions, and this is projected to have no spatially robust changes over East Asia and Southeast Asia but to undergo substantial delays over the rest of the NH monsoon regions, with the strongest delays of five to more than eight pentads occurring over South Asia and Africa. A delay in the monsoon demise is also projected over many regions that experience late monsoon onset, however, the delay in monsoon demise is less than the delay in monsoon onset, which reflects a shrinking of the monsoon rainy seasons. These changes in the regional monsoons are also accompanied by an increase in their seasonality, which is mainly driven by a higher fraction of the annual precipitation during the monsoon seasons. Collectively, these projected changes point towards a substantial drying outside of the monsoon seasons and an intensification of monsoon precipitation within the monsoon seasons.

A robust relationship between the projected pre-monsoon drying and delays in the monsoon onset exists across regional monsoons, as a weakening of latent heat driven atmospheric warming during the pre-monsoon period delays the overturning of atmospheric subsidence in the monsoon regions. Pre-monsoon precipitation is predominantly convective in nature, and it is suppressed as a result of the deepening of the atmospheric boundary layer and a reduction in the relative humidity levels in the lower troposphere. In response to the RCP2.6 forcing, which corresponds to a scenario of strongly reduced GHG emissions by the end of the century, most of the regional monsoons exhibit small changes within the baseline variability. This illustrates the strong added value in reducing emissions for global economies, in particular, of currently poor and more vulnerable tropical countries.

While regional monsoon projections in this study are one of the most detailed to date, there remain a number of outstanding issues that illustrate the need of future work to address them. These include potential impacts of land-use change (Pielke 2005), irrigation systems and water management practices (Devanand et al. 2019), land–ocean-atmosphere feedbacks (Chou et al. 2001), model internal biases (Giorgi and Bi 2000), sensitivity of simulated responses to the choice of driving GCMs (Kjellstrom et al. 2011), and the inter-comparison of RCMs. Finally, finer scale observational datasets are needed to evaluate regional models, particularly in complex terrain areas and in regions where the largest impacts are expected.