3.1 Variance detection of intermodal simulations
Before predicting how climate change will affect rainfall variability in MSWM using GCMs, we tested the correlation between the models' data and actual observations. We assessed the agreement between observed and EnsMean-simulated rainfall patterns by calculating climatological rainfall patterns. Spatial climatological means, categorized into annual and seasonal patterns, were derived by averaging from daily values. We measure how rainfall changes over time by looking at its spread or variability, which is described using a statistic called standard deviation. In our study, we followed a process of previous research by Akinsanola et al., (2020) and Menon et al., (2013).
First, we took out the Probability Distribution Function (PDF) of JJAS daily rainfall anomalies over the study area for both the observations and the historical simulation (1981–2014) (Fig- 2.a). Then we performed Pearson correlation and two-tail t-test analyses, and this were predominantly employed for correlation analysis in this study (Fig- 2.b). Fig- 2.b explained the significant of annual correlation of mean daily rainfall (mm/day) for observation, GCMs, and their EnsMean. The Pearson correlation test indicated a highly positive correlation between observation data and GCMs at a 95% confidence level. And to evaluate the seasonal pattern of each model by time series analysis can be seen close patterns as in Fig- 2.c.
Additionally, daily anomalies, averaged over the Mainland Indochina region, underwent detrending and band-pass filtering. The filtering included distinct bands for "synoptic" (2–10 days), "monthly" (25–35 days), "seasonal" (80–90 days), and "annual" (300–400 days) timescales, as shown in Fig- 3. The utilization of 34 years of model output imposes limitations on the gauging of decadal variability. Further details on the outcomes of the band-pass filtering method to daily anomalous rainfall are provided in Fig- 3 for the observations (average of CPC and CHRIPS observations) and the ensemble mean of the EnsMean members.
Fig- 2 (a) Probability Distribution Function (PDF) of both observed daily rainfall anomalies with Ens_Models for Historical (HIS) simulation (1981–2014). (b) The correlation heatmap of CMIP6 Ens-Model (MME) climatology of historical (1981–2014) mean daily rainfall (mm/day) in comparison to observational data (c) their seasonal time series.
Fig- 3 Outcome of band-pass filtering of daily rainfall anomalies for (a) “synoptic" (2–10 days), (b) "monthly" (25–35 days), (c) "seasonal" (80–90 days), and (d) "annual" (300–400 days) timescales.
In general, the EnsMean tend to replicate observations, with variability more closely aligned with the observations, as elaborated in the next result section. In addition, to recreate the variability of historical MSWM rainfall and examine the impact of climate change on regional rainfall variability, we examine the models' capability to regenerate the mean climatology rainfall and the extreme indices listed in Table 2.
The spatial variation of the mean MSWM rainfall (Pr), the number of consecutive dry days (CDD), the amount of consecutive 5-day maximum rainfall (RX5day), and the amount of rainfall of very wet days (R95pTOT) of CPC, CHIRPS, and EnsMean is shown in Fig- 4. In specific, CPC and CHIRPS observations (Fig- 4.a-b) reveal two most rainfall centers situated within the northern Himalayas foothill highlands around 25°–30°N, 90°–100°E (as “N” in Fig- 1.b). The second center is evident along the eastern coastal area of the Bay of Bengal (as “W + S” in Fig- 1.b). The less rainfall dry zone over central region lies between 17° and 25°N, with rainfall decreasing around from 94°E to 97°E (Fig- 4 a-c). Additionally, areas with high topography correspond with regions of greatest rainfall (Fig- 1 and Fig- 4). The spatial pattern of MSWM rainfall distribution is faithfully captured by EnsMean (Fig- 4.c) and skillfully replicate the observed rainfall patterns. This exhibiting pattern shows that correlation coefficients surpassing 0.7 to 0.9 when compared to two observation products.
As the evaluation of CDD (Fig- 4.d and f), a rainfall threshold less than 1 mm has been applied to differentiate between dry and wet days (Oo et al., 2023; Yazid & Humphries, 2015). A large quantity of dry days over the central Myanmar region (Fig- 4.d, e), with values declining southward over the coast. In general, the observations and EnsMean (Fig- 4.f) realistically replicate the CDD in patterns, boasting PCCs exceeding 0.7 in all members. Notably, the EnsMean contributes significant value over numerous grids over the region.
Fig- 4 Present-day simulated map of the mean rainfall (Pr), the consecutive dry days (CDD), the consecutive 5-day maximum rainfall (RX5day), and the very wet days rainfall (R95pTOT) of (a, d, g, j) CPC, (b, e, h, k) CHIRPS, and (c, f, i, l) EnsMean during 1981–2014.
Moreover, the spatial pattern of the 5-day maximum rainfall distribution (Fig- 4.g–i) and the very wet days (Fig- 4.j–l) is realistically reproduced by the observations in conjunction with models EnsMean. Our findings indicate that the extremes simulated by EnsMean are highly in line with CPC (CHRIPS) observations. Overall, the rainfall indices exhibit better representation in EnsMean compared to observations. However, some noticeable biases persist over specific grids. Furthermore, the collective performance of EnsMean slightly surpasses that of individual models, attributed to the mitigation of spatial errors through ensemble averaging.
3.1.1 Changes in total rainfall
Future predictions of annual total rainfall under two different scenarios (SSP2-4.5 and SSP5-8.5) are shown in Fig- 5, where they are compared with the Multi-Model Ensemble (MME) simulation using time-series simulation to represent the historical time span. It is clear from this portrayal that the SSP5-8.5 scenario shows a considerable increase in future annual rainfall, whereas the SSP2-4.5 scenario shows a minor increase. To examine how these alterations have evolved throughout time. The annual cycle of total rainfall is examined, by the future period is sliced into two as the long-future (2061–2099) and the near-future (2021–2060) concerning to the present (1981–2020). When compared to the historical time span, the first few decades display a rather constant variability, while the far future displays a significant fluctuation range, especially under SSP5-8.5 simulation. These findings align with recent climate change studies and are in concurrence with the reports presented by the Intergovernmental Panel on Climate Change (IPCC).
Fig- 5 Annual total rainfall time series of historical simulation and future projections (mm/year) for (a) Monsoon Core (W + S as in Fig-1.b) region, and (b) Northern (N) region. Climatology mean (black), historical (green),, future of EnsMean simulations under SSP2-4.5 (orange), and SSP5-8.5 (red). And dash line shows the linear trend of mean values.
3.2 Assessment and anticipated modifications to rainfall variability
We analyses the climatology and the expected evolution of the variability of daily MSWM rainfall in the unfiltered state in the CMIP6 EnsMean models. As explained in the methodology (section-2), the daily rainfall figures shown here are derived from the 34-year daily climatology.
Fig- 6 Future and historical change in the spatial pattern of rainfall variability of (a) Observation [average of CPC and CHRIPS observations] and (b) standard deviation of daily rainfall anomalies for MSWM season (mm day − 1) of multimodel ensemble mean in historical simulations. Red box shows the mainland Indochina region.
Fig- 6.a and b show the results of evaluating the EnsMean historical simulations daily rainfall variability of MSWM season with the mean observations data during 1981–2014 time span. As is clear in Fig- 6.a, the 75% of the mainland Indochina region explained high variability of daily rainfall ranging from 8 to 14 mm per day with low variability over central Myanmar coastal regions below than 5 mm per day. Also, the high variability of MSWM rainfall (~ 15 mm/day) is noteworthy over the Bangladesh, north-eastern India, and southwestern China regions. When comparing the pattern of observations rainfall to EnsMean, the pattern of rainfall variability is realistically reproduced, albeit with a higher magnitude (Fig- 6.b). However, the variability of EnsMean daily rainfall is much less intense than observational rainfall.
Fig- 7. a and d illustrate the changes of daily rainfall variance from the past (1981–2014) to the future periods (2020–2054) under the SSP2-4.5 and SSP5-8.5 scenarios. In the context of global warming, EnsMean projects that exceed in rainfall variability across most of mainland Indochina, ranging from 20–60%. The direction of the predicted changes remains consistent for both scenarios, with SSP5-8.5 showing greater magnitude and spatial extent. Despite variations in magnitude, there is substantial agreement among ensemble members, with at least 90% of the models concurring on the direction of change at most gridded area over Mainland Indochina. This demonstrates that how much strong the anticipated changes are?
Fig- 7, Standard deviation of anomalous daily rainfall (%) changed for (a, b, c) SSP2-4.5 and (d, e, f) SSP5-8.5, from historical (1981–2014) to future (2021–2054). The overall change in rainfall standard deviation (ΔSDR) has two components: ΔSDR1 represents the part entirely accounted for by mean rainfall variation, while ΔSDR2 is associated with fluctuations in the coefficient of variance. Stippling indicates grid locations where at least 90% of observations concur on the direction of change in EnsMean.
Using Equation-4, Fig- 7.a and d are further dissected into next figures (Fig- 7. b, c, e and f). The amplified variability in rainfall, attributed to the positive alteration in mean rainfall (ΔSDR1, depicted in Fig- 7.b and e), is predominantly concentrated over the Tibetan plateau, the southern Bangladesh coast sub region, and Sri Lanka, coinciding with regions of the largest ΔPr (Fig- 10.a and b). Regions with substantial negative values in ΔSDR1 are identified along the coastal areas of the Gulf of Thailand, where ΔPr lees than 0. Although the changes are more pronounced in SSP5-8.5, the directional trend of the projected variance changes are still in consistent across both scenarios.
Moreover, the alterations in rainfall variability induced by changes in coefficient of variation (ΔSDR2, shown in Fig- 7.c and f) overshadow the previously reported future rise in variability illustrated in Fig. 7.b and e. A majority of grid points over Mainland Indochina demonstrate an escalating variability primarily influenced by changes in CVR. The sign of ΔSDR2 aligns in both SSP2-2.4 and SSP5-8.5 scenarios, with SSP5-8.5 exhibiting a greater magnitude of change. Among the ensemble members, there is notably strong intermodel consensus; at least 95% of models agree on the direction of change at the maximum gridded area, especially in the MSWM monsoon area of the study region (Oo, 2023).
3.3 Regional variations in the variability of rainfall and the function of the thermodynamic element
Fig- 8 Regional variations in the variability of rainfall and the function of the thermodynamic element. Subtracting the past time slice (1981–2014) from the future time slice (2020–2054) change in the standard deviation of daily rainfall (%) for the mainland Indochina region for the (a) synoptic, (b) monthly, (c) intraseasonal, and (d) yearly timescales using band-pass filtration. Yellow (red) bars show the EnsMean for SSP2-4.5 (SSP5-8.5).
Moreover, an in-depth examination of the complete model spread is expanded to cover the regional scale of Mainland Indochina. This exploration aims to determine whether a comparable escalation in variability would manifest across various timescales. Fig- 8 illustrates the rainfall variability changes percentage (%) at various timescales across the Mainland Indochina region. Overall both temporal and spatial, the EnsMean reliably predicts an increase in the variance of daily rainfall. Notably, a noteworthy level of consistency is evident, with substantial model agreement observed across the entire region at various timescales. The most substantial strength in the variability of future rainfall is notable at monthly and annual timescales, despite the persistence of evident intermodel spread, particularly with SSP5-8.5 exhibiting higher intensity than SSP2-4.5. It is noteworthy that prior research has indicated that heightened rainfall variability could be attributed to a warmer environment and higher levels of atmospheric moisture, aligning with the Clausius–Clapeyron relation (Cook & Seager, 2013; Ge et al., 2021; Liepert & Previdi, 2009). In this case, a "null hypothesis" with direct relevance is that enhanced the variability of MSWM rainfall may be caused by exceed atmospheric moisture content due to the warmer climate. Indeed, forecasts of climate change over Mainland Indochina have already shown increased atmospheric moisture (Tang et al., 2021). Therefore, we compute a conceptualized augmentation of the daily MSWM rainfall, as detailed in the methodology explanation section. The disparity in the "CC" alteration in rainfall variability between SSP2-4.5 and SSP5-8.5 is juxtaposed with the alterations in rainfall variability, elucidating the extent to which the modeled change by the two scenarios projects might be attributed to this thermodynamic upswing in atmospheric moisture. Across all Mainland Indochina regions, the EnsMean simulation manifests noteworthy fluctuations in rainfall variability beyond what would be anticipated from the theoretical thermodynamic response across all timescales. This implies that the main factor affecting the expected variability of rainfall across Mainland Indochina may be the thermodynamic contribution.
Fig- 9 Correlation scatter plot between change of standard deviation (%) and change of mean rainfall value (%) of MSWM rainfall for the (a) synoptic, (b) monthly, (c) intraseasonal, and (d) annual time scales by the different of HIST (1981–2014) to future (2020–2054) of SSP2-4.5 and SSP5-8.5. The Mainland Indochina region's SSP2-4.5 (SSP5-8.5) area average is represented by orange (red).
The variation in rainfall patterns shows a notable and consistent correlation with the variability of historical rainfall across Mainland Indochina regions in both scenarios. These correlations are exhibited positive values and statistically significant at the 95% confidence level across all timescales. Particularly, there are high and significant positive correlations found at the synoptic scale. Overall, alterations in change of standard deviation (%) exhibit a significant positive correlation with the change of mean value (%) across the study region at various timescales (Fig- 9). This robust and statistically significant relationship suggests that modifications in mean rainfall contribute to explaining specific aspects of the intermodel extend in the changes over the daily rainfall variance.
3.4 Projected variations in average and anomalous rainfall
Fig- 10 Changes of the rainfalls in the future. Modification Projected multi-model changes for SSP2-4.5 (a, c, e, g) and SSP5-8.5 (b, d, f, h) in terms of (a, b) daily rainfall, (c, d) consecutive dry days, (e, f) 5-day maximum rainfall, and (g, h) total rainfall equal or more than daily 95th percentile. Stippling shows over the grid sites is denoted that the change in EnsMean is agreed with the observation at least 95% significant.
We are motivated to analyses changes in average and extreme rainfall indicators further by the observable and robust rise in the variability of MSWM rainfall in the analyzed EnsMean simulations. Fig- 10 displays the spatial patterns of MSWM total rainfall from the 95th percentile (R95pTOT), 5-day maximum rainfall (RX5day) consecutive dry days (CDD), and the mean rainfall (Pr), under the influence of global warming. This comparison is made by the difference between the twenty-first-century future period (2020–2054) and the historical timeframe (1981–2014) under the SSP2-4.5 and SSP5-8.5 scenarios.
In general, the upcoming amplification of MSWM rainfall variability across the study area is unified that the 5–10% positive change in average rainfall in the Mainland Indochina regions, a 15–20% increase over Bangladesh and central Myanmar, and a 30–40% upturn over Sri Lanka and the Tibetan foothill regions (see Fig- 10.a, b). All regions exhibit a similar pattern of escalating intensity under the SSP5-8.5 scenario. Notably, there is only a slight decrease in average rainfall over the Gulf of Thailand region in both scenarios. These findings are consistent with previous research highlighting future changes in the zonal dipole of mean rainfall over Mainland Indochina. (Kim et al., 2008; Sengupta & Nigam, 2019; You et al., 2017).
In contrast to ∆Pr, alterations in the ∆CDD display a noticeable upswing (or downturn) compared to the corresponding pattern in Fig- 10.a and b, excluding certain areas in the central Tibetan region. However, the intensity registers only a 5 to 15% change (Fig- 10.c,d). Furthermore, both ∆Rx5day and ∆R95pTOT indicate a projected increase of approximately 10–20% in the future (Fig- 10.e,f). The presented forecasts exhibit notable robustness for mean rainfall and ∆R95pTOT, with a consensus of at least 95% of EnsMean members aligning on the direction of change in most grid points (Fig- 10.g,h). Nevertheless, there is a considerable intermodel extend in changes in ∆CDD, with fewer grid points reflecting a 95% model agreement.
Furthermore, as previously noted, SSP5-8.5 has larger projected increase and drop magnitudes, which are probably due to its higher forcing (Oo et al., 2023). In conclusion, enhancing the consecutive dry days and the extreme rainfall events are linked to the predicted exceeding the variability of MSWM rainfall over Mainland Indochina. These developments have important ramifications for the region's future water supplies, agricultural, hydroelectric power production, and social stability.