Impacts of half a degree additional warming on the Asian summer monsoon rainfall characteristics

We use data from large-ensemble (>100 members) simulations of the five AGCMs participating in the HAPPI project (table 1). Each member of AGCM provides 10 year simulation data for the current climate (2006–2015, HIST), and corresponding 1.5 °C and 2.0 °C warmer conditions (Plus15 and Plus20, respectively). The five AGCMs have different resolutions and physics but were integrated under the identical boundary conditions. The observed sea surface temperature and sea-ice conditions were prescribed in the HIST experiment. The SST boundary condition for Plus15 and Plus20 experiments were prepared by adding a SST warming pattern estimated from CMIP5 multi-models (based on RCP2.6 and RCP4.5 scenario runs from 23 models, Mitchell et al 2017) to the observed, and equilibrium forcings of aerosols and greenhouse gases were implemented following CMIP5 RCP experiments. For the sea ice, due to low reproducibility of CMIP5, linearized function of meridional temperature was used to estimate sea-ice conditions in warmer worlds over the pole regions. Details of the HAPPI experimental design are referred to Mitchell et al (2017). Before analysis, we interpolated all model data into the identical observed grid (1° × 1°, based on GPCP v1.2 data) and then the identical GPCC land mask was applied. Since different AGCMs have different ensemble sizes ranging from 100–500 (table 1), we calculate ensemble averages of each AGCM and then obtain multi-model ensemble means (hereinafter referred to as MME).

In order to check potential impacts of the air-sea coupling on our results, which is known to be important for the Asian monsoon domain (e.g. Webster 2006, Dong et al 2017), we repeat our analyses using 32 coupled global climate model (CGCM) data from the CMIP5 RCP8.5 scenario experiment (table S1 available at stacks.iop.org/ERL/13/044033/mmedia), following previous studies (Fischer and Knutti 2015, Knutti et al 2016, Schleussner et al 2016). Note that for some models, we extend the data to before 2006 using 'historical' experiment. Using single member (r1i1p1) of each CGCM, 30 year periods having 1.5 °C and 2.0 °C warming relative to the preindustrial level (1861–1880 mean) are selected (table S1). And then changes of the Asian summer monsoon rainfall under two transient warming conditions are analyzed in the same way as done for the HAPPI AGCM data.

Considering that each ensemble member of the HAPPI experiments consists of 10 year data, we define three variables of mean and precipitation extremes as follows. First, we use summer (June–July–August) mean of daily precipitation (Rm) calculated for each member. Secondly, for each member, we extract JJA maximum of daily precipitation from each year, and then calculate its 10 year average (Rx). To get more extreme events, we obtain 10 year maximum of summer daily precipitation over the whole 10 years (Rx₁₀). Rx and Rx₁₀ approximately correspond to 2 year and 10 year return values, respectively, under the independence assumption between ensemble members. They are also largely comparable to 99th and 99.9th percentiles of daily precipitation. In case of 30 year data from CMIP5 CGCM runs, we divide each 30 year data into three 10 year chunks and calculate Rm, Rx and Rx₁₀ from each 10 year chunk for consistency with the HAPPI data processing. This is also to increase the sample size of CMIP5 data. However, it should be noted that the three 10 year chunks do have different global mean warming levels (±0.4 K based on MME) due to the transient nature of the CMIP5 experiments, which makes it difficult to compare the HAPPY and CMIP5 experiments in a completely consistent fashion.

A moisture budget contribution analysis was introduced by Seager et al (2010) to examine thermodynamic and dynamic mechanisms of hydrological trends during the long-term period and this approach has been widely applied to different warming scenarios (Seager and Naik 2012, Gao et al 2012, Endo and Kitoh 2014, Li et al 2015, Lee et al 2017). Long-term mean water budget can be expressed as follows with identical unit (mm day⁻¹):

$$\begin{equation} {\rm{\delta P}} = {\rm{\delta E}} + {\rm{\delta TH}} + {\rm{\delta DY}} + {\rm{\delta NL}} + {\rm{\delta TE}} \end{equation} \tag{ 1 }$$

$$\begin{equation} {\rm{\delta TH}} = - \frac{1}{{\rho_{\rm{w}}g}}\mathop \smallint \limits_0^{{p_s}} \nabla \bullet \left( {{{\bar u}_{{\rm{HIST}}}}\left[ {\delta \bar q} \right]} \right){\rm{d}}p \end{equation} \tag{ 2 }$$

$$\begin{equation} {\rm{\delta DY}} = - \frac{1}{{\rho_{\rm{w}}g}}\mathop \smallint \limits_0^{{p_s}} \nabla \bullet \left( {\left[ {\delta \bar u} \right]{{\bar q}_{{\rm{HIST}}}}} \right){\rm{d}}p \end{equation} \tag{ 3 }$$

$$\begin{equation} {\rm{\delta NL}} = - \frac{1}{{\rho_{\rm{w}}g}}\mathop \smallint \limits_0^{{p_s}} \nabla \bullet \left( {\delta \bar u\delta \bar q} \right){\rm{d}}p \end{equation} \tag{ 4 }$$

$$\begin{equation} {\rm{\delta }}\left( \bullet \right) = \left( \bullet \right){\rm{Plus}} - \left( \bullet \right){\rm{HIST}} \end{equation} \tag{ 5 }$$

where P is precipitation, E is evaporation from the surface, u is the horizontal wind vector, q is the specific humidity, p is the pressure, ρ_w is the water density, g is the gravitational acceleration, and p_s is the surface pressure. Overbars indicate climatological monthly means (seasonal results are obtained by averaging monthly means). Subscripts HIST and Plus indicate historical period (HIST: 2006–2015) and future period (Plus15 and Plus20), respectively. Transient eddy term (δTE) is not considered here due to its negligible magnitude compared to the δTH and δDY terms (Gao et al 2012, Li et al 2015). Therefore, we only analyze changes (•) in precipitation (P = Rm), evaporation (E), thermodynamic (TH), dynamic (DY) and nonlinear (NL) terms, which are calculated using monthly mean values. Vertical integrations for TH, DY, and NL terms are conducted using twelve standard pressure levels (1000, 925, 850, 700, 600, 500, 400, 300, 250, 200, 150 and 100 hPa; specified in the HAPPI experiment) within the troposphere.

Spatiotemporal complex characteristics can be fairly considered by using percentiles of precipitation amounts (Freychet et al 2015). We analyze precipitation characteristics based on percentiles estimated from aggregated daily data from all ensemble members for each HAPPI AGCM (note that we use single ensemble member from each CGCM for CMIP5 analysis). For example, in case of CAM4-2°, we estimate 10th–90th percentiles of precipitation on every grid from about 460 000 data (500 ensemble members × 92 days duing JJA × 10 years). By aggregating all ensemble members, we can assess changes in extreme precipitation events above 99th percentile with enough number of samples (at least 1000 samples are needed to estimate 99.9th percentile intensity). To better identify changes in precipitation extremes, we specifically use bins of 10% width from 1st to 90th percentiles (1st percentile represents minimum precipitation intensity), bins of 1% width from 90th to 99th percentile, and bins of 0.1% width from 99.0th to 99.9th percentiles, in a similar way to Freychet et al (2015). For the CMIP5 multi-CGCM analysis, we confine our analysis up to 99th percentiles because of smaller sample size (90–92 summer days × three 10 year chunks for each model). Next, changes in frequency are also examined following Freychet et al (2015) by counting days at each grid where precipitation amount is greater than or equal to the precipitation percentile and less than the next level intensity, using the percentiles estimated from the historical period (2006–2015). For example, to get future changes in 80th percentile frequency, we count days where daily precipitation is equal to or larger than the historical 80th percentile and also less than the historical 90th percentile. In the frequency analysis, changes in dry days are additionally examined, where dry days are defined as days with precipitation less than 1st percentile.

Zoom In Zoom Out Reset image size

Spatial patterns of the mean and extreme rainfall for HIST, Plus15, and Plus20 simulations, and their differences are displayed in figure 1. Compared to the observations, HAPPI HIST runs well reproduce mean (Rm) and extreme precipitation (Rx and Rx₁₀) in Asian monsoon regions but with some underestimation of extreme precipitation intensity (see below). HAPPI future runs predict overall increases in mean and extreme precipitation over the Asian monsoon region. At least four out of five AGCMs have the same sign of changes (areas hatched in figure 1), indicating good inter-model agreement. Based on MME, stronger increases are projected for extreme precipitation (5–20 mm day⁻¹) than mean precipitation (<5 mm day⁻¹), for which all five AGCMs predict its intensification over the Asian monsoon regions. Rx₁₀ generally exhibits stronger increases than Rx, suggesting stronger intensification of more-extreme precipitation as reported by previous studies (Freychet et al 2015, Lee et al 2017, Chen and Sun 2017). Differences between Plus20 and Plus15 suggest that a large part of the Asian monsoon region will experience additional intensification of precipitation, particularly extremes, due to the half a degree additional warming.

To explore regional characteristics, we define two sub-regions of South Asia monsoon (SAS) and East Asia monsoon (EAS) and examine area-mean changes in precipitations for Rm, Rx, and Rx₁₀ (figure 1, bar graphs). HAPPI AGCMs reproduce the observed precipitation (x mark) reasonably for both mean and extreme precipitation although models tend to underestimate Rx and Rx₁₀ over SAS and overestimate Rx₁₀ over EAS. Plus15 results show stronger increase in Rm over EAS (MME value: 7.7%) than SAS (2.5%). Similar increases in extreme precipitation are found between EAS and SAS with stronger changes in Rx₁₀ (SAS: 7.6%, EAS: 8.3%) than Rx (SAS: 5.8%, EAS: 6.7%). Note that these percentage changes are not much different between Rx and Rx₁₀ but Rx₁₀ has much stronger changes than Rx when considering absolute units (mm day⁻¹) as can be seen from the spatial patterns of HIST (figure 1). Plus20 results predict more increase in extreme precipitation with about 10% (based on MME) for Rx and about 12% for Rx₁₀ over both SAS and EAS regions. Differences between Plus20 and Plus15 results, which indicate potential benefits of the global warming mitigation, show relatively small (less than 2%) increases in Rm but larger increases of about 3%–5% (based on MME) in Rx and Rx₁₀ for both monsoon regions. The Rx₁₀ results in particular show good inter-AGCM agreement and suggest the robust intensification of more-extreme precipitation due to the additional warming, corroborating previous findings (Freychet et al 2015, Lee et al 2017, Chen and Sun 2017).

Zoom In Zoom Out Reset image size

CMIP5 results based on transient warming conditions from 32 CGCMs (figure S1) are very similar to those from HAPPI AGCMs, confirming larger changes in more-extreme rainfalls over the Asian summer monsoon region. As expected, inter-model differences are larger in CMIP5 results than in the HAPPI ensemble because of the larger inter-CGCM difference in internally-driven SSTs unlike the identically prescribed SSTs (see section 2 above). Nevertheless, it is notable that results from five CGCMs corresponding to the HAPPI AGCMs (table S1) spread widely in the historical climatology, future changes, and Plus20-Plus15 differences (figure S1, × marks), reasonably covering the CMIP5 ranges.

In order to understand mechanisms of precipitation changes under Plus15 and Plus20 experiments, results from the moisture budget analysis for Rm are illustrated in figure 2. All HAPPI AGCMs project increase in evaporation (E) and thermodynamic term (TH) under warming, while larger inter-model differences exist in dynamic term (DY). SAS and EAS area-mean results (figure 2, bar graphs) confirm that the increases in Rm over the Asian monsoon region are driven majorly by the increases in E and TH while the DY contribution is relatively uncertain with larger inter-model differences, which is in accord with previous studies (Endo and Kitoh 2014, Li et al 2015, Lee et al 2017). A quick check from an inter-model correlation analysis indicates that the DY increases (decreases) are largely associated with the strengthening (weakening) of the monsoon circulations (not shown). This issue on uncertainty sources warrants comprehensive analysis. NL term is much smaller than other terms, as in other studies (e.g. Seager et al 2010, Lee et al 2017). Differences between Plus20 and Plus15 indicate that E and TH terms explain the enhanced mean rainfall for both SAS and EAS regions, implying that thermodynamic mechanisms are critical to the additional increase of summer mean precipitation. Results of moisture budget analysis from the CMIP5 CGCMs (figure S2) generally support those from the HAPPI AGCMs, also confirming previous studies (Endo and Kitoh 2014).

Zoom In Zoom Out Reset image size

Changes in daily precipitation characteristics are assessed for the SAS and EAS regions. First, precipitation intensity corresponding to 1st to 99.9th percentiles is estimated from HIST, Plus15, and Plus20 experiments, and then the future changes relative to HIST and the difference between Plus20 and Plus15 are obtained (figure 3). Results of HIST simulations show that HAPPI AGCMs (bars) capture the observed distribution (x marks) of precipitation intensity for both monsoon regions although some overestimations of precipitation intensity are seen at low percentiles (<50th percentile). Under warmer conditions of Plus15 and Plus20 experiments, all HAPPI AGCMs predict increase in precipitation intensity by about 3%–10% compared to the corresponding HIST values (based on MME) at high percentiles (>90th). Importantly, as the percentile gets higher (i.e. more-extreme precipitation), the stronger increase in precipitation intensity is projected for both Plus15 and Plus20 scenarios, which is more clearly observed in the SAS region. It is notable that by definition, absolute changes and associated impacts will be even larger at these high percentiles due to the stronger magnitude of climatology, which is consistent with our comparisons among Rm, Rx, and Rx₁₀ given above (figure 1). For SAS, changes in medium precipitation intensity between 10th to 80th percentiles are uncertain with mixtures of increases and decreases among models, which might be related to a stronger influence of dynamic contribution in SAS than in EAS (figure 2; Endo and Kitoh 2014, Freychet et al 2015, Vittal et al 2016, Roxy et al 2017). Light precipitation lower than 10th percentile is expected to decrease in SAS. Accordingly, distribution of difference in precipitation intensity between Plus20 and Plus15 indicates that the additional half a degree warming will strengthen extreme precipitation above 90th percentile by about 1%–4% whereas a decrease in light precipitation below 10th percentile is expected by about 3% with larger uncertainty over SAS. The EAS region exhibits different patterns of change. Precipitation intensity is projected to increase at all percentiles, indicating a shift of precipitation distribution. This is consistent with Freychet et al (2015), who found a similar overall intensification of precipitation over EAS using CMIP5 models. We also find same results from 32 CMIP5 CGCMs (figure S3). Accordingly, Plus20-Plus15 results for EAS show precipitation increases at all percentiles, and particularly, about 2%–3% increases are observed for extreme precipitation (>90th percentile) and for light precipitation (<50th percentile).

In short, for both Asian monsoon regions, HAPPI AGCM simulations consistently project further intensification of extreme precipitation due to additional half degree warming, which becomes stronger for more-extreme precipitation. This implies that possible benefits of global warming mitigation will be identified more strongly for extreme precipitation events which usually deliver larger impacts. CMIP5 results for the intensity changes under 1.5 and 2.0 degree warming scenarios (under transient world assumption) resemble those from the HAPPI ensemble for both SAS and EAS regions (figure S3), suggesting the minor role of air-sea coupling in the assessment of precipitation intensity distribution. As expected, the CMIP5 ensemble exhibits larger inter-model differences in future projections and differences between Plus20-Plus15 than the HAPPI results.

Zoom In Zoom Out Reset image size

Changes in frequency of daily precipitation are similarly examined for the percentiles from 1st to 99.9th percentiles that are estimated from the HIST experiments (figure S4). All HAPPI AGCMs agree on more frequent extreme rain events above 90th percentile, supporting the intensity changes (figure 3). Non-linear increases in frequency are notable particularly above 99th percentile, reaching more than 60% increase at 99.9th percentile under the 2 °C warmer world. This behavior seems to be related to stronger moisture constraint on more-extreme precipitation (Allen and Ingram 2002, see below). On the other hand, three or four out of five HAPPI AGCMs predict more frequent dry days for the two monsoon regions, which are consistent with previous projections based on CMIP5 models (Kitoh et al 2013, Freychet et al 2015, Lee et al 2017). Differences between Plus20 and Plus15 indicate a sharp increase at extreme precipitations above 99th percentile. We also analyze CMIP5 CGCMs models for frequency changes under 1.5 °C and 2.0 °C warming scenarios and obtain overall similar results (figure S5).

Previous studies suggested that extreme precipitation changes follow the Clausius-Clapeyron (C-C) relation at global scales (e.g. Allen and Ingram 2002, Min et al 2011, Kharin et al 2013) and also at regional scales (e.g. Pall et al 2007, O'Gorman and Schneider 2009, Lee et al 2017). In order to explore physical mechanisms behind the stronger responses (or benefits) in more-extreme precipitation, inter-model relationship is examined between area-mean temperature changes and precipitation intensity changes using HAPPI AGCMs. The five AGCMs under the Plus15 and Plus20 scenarios provide ten samples (ensemble mean of each model). To consider the small number of models of HAPPI experiments as well as the possible limitation of AGCM experiments based on a single SST boundary condition and with no air-sea interactions, we also examine the temperature-extreme precipitation relationship using 32 CGCMs from the CMIP5 ensemble, which provide 64 samples representing both 1.5 °C and 2.0 °C warmer conditions in the transient world assumption (table S1). HAPPI results show that AGCM responses in temperature and precipitation over the two Asian monsoon regions are quite different across models, although identical SST conditions are prescribed (figure 4). This suggests considerable influences of the differences in model physics such as convective parameterization and land surface schemes (cf. Song and Zhou 2014, Lee et al 2016, 2017).

Statistically significant relationships are observed between regional-mean temperature changes and extreme precipitation exceeding 99th percentiles in the both SAS and EAS regions from HAPPI AGCMs (figure 4). EAS region exhibits a significant temperature-precipitation relation at 90th percentile as well. The resultant linear regression slopes of extreme precipitation (99th and 99th percentiles) onto temperature are generally well in accord with the C-C relation (7%/°C) for both monsoon regions, representing the important role of moisture increase under warming in the intensification of extreme precipitation. When checking the temperature-extreme precipitation relationship using the individual HAPPI models, the regression slopes deviate a lot from the C-C scaling (3%–20%/°C for 99th and 99.9th percentiles), but most of the models generally support the C-C relation. CMIP5 results show statistically significant links between extreme precipitation and temperature at both 90th and 99th percentiles, supporting the HAPPI-based results. Noticeable difference includes the weaker regression slope (5.3%/°C) in CMIP5 over EAS region, which may represent stronger influence of air-sea interactions in CGCMs on this region, compared to SAS region (e.g. Dong et al 2017). Nevertheless, regression slopes are greater at 99th percentile than at 90th percentile in both regions and from both AGCM and CGCM ensembles. Likewise, regression slopes at 99.9th percentiles (10.13 and 6.86%/°C for SAS and EAS, respectively) are also slightly greater than at 99th percentiles (9.36 and. 6.73%/°C) in HAPPI GCMs. This clearly indicates the stronger contribution of moisture increase (C-C relation) on the strengthening of more-extreme precipitation over the Asian monsoon region, which has not been demonstrated before. This regression analysis based on Plus20 and Plus15 data also implies that global warming mitigation from 2 °C to 1.5 °C target temperatures would provide potential benefits as weaker increases in heavy precipitation intensities, for example, by 3%–5% at the 99.9th percentile values (figure 3). On the other hand, regression lines for the 99th and 99.9th percentiles in figure 4 pass very close to the origin but not exactly through it, which indicates that the temperature-extreme precipitation relation could be in fact nonlinear due to other effects including dynamic contribution on extreme precipitation (Freychet et al 2015, Lee et al 2017, Pfahl et al 2017).