Large-scale circulation dominated precipitation variation and its effect on potential water availability across the Tibetan Plateau

The large-scale circulation, Indian summer monsoon (ISM), has a strong influence on the Tibetan Plateau (TP) since its onset and intensity have profound impacts on regional precipitation, which then can supply water for glaciers, lakes, rivers and social demands. Weakening monsoon intensity and longer monsoon duration seem contradictory, as a weaker monsoon tends to produce less precipitation, while a longer duration increases the probability of precipitation. Past research has focused on how ISM’s intensity affects precipitation, with little consideration of the impacts of ISM duration. Here, we investigate the long-term (1979–2100) variability in the ISM’s duration and intensity. We find a prolonged ISM from 1979 to 2018, accompanied by monsoon weakening. Different combinations of duration and intensity have resulted in different spatial patterns of precipitation across the southeastern TP. Weakening and prolonged ISM is helpful to produce more precipitation around the southeastern TP, with intensity acting as a dominant control. Afterwards, an obvious impact can be found on potential water availability. Climate projections suggest that the ISM will weaken and lengthen until 2100, thereby increasing precipitation and potential water availability across the southeastern TP. This long-term trend should receive more attentions as increased regional extreme precipitation will increase the probability of flood risks until the end of this century.


Introduction
The Tibetan Plateau (TP), called the 'Asia's water tower' , contains a large number of glaciers, lakes and rivers, and the precipitation is their main water source (Yao et al 2019). The large-scale circulation, Indian summer monsoon (ISM), is one of climate systems to influence precipitation changes around the TP, especially the southeastern TP. The onset, retreat and intensity of ISM determine the amount of rainfall during the monsoon season (Raju and Bhatla 2014) and more than 70% of the total precipitation falls around the southeastern TP during the monsoon season (figure 1). Understanding the intensity and duration (from onset to retreat) of the ISM (hereafter called ISM characteristics) is a subject of active research, which can affect agriculture and irrigation (Webster et al 1998, Syroka andToumi 2004). Previous studies have shown that total seasonal precipitation is primarily controlled by ISM's duration Figure 1. Mean percentage (unit: %) of monsoon precipitation in annual total precipitation. Shaded area represents the mean surface geopotential height (unit: m). The precipitation data and geopotential height data are derived from GPCP monthly precipitation and NCEP-DOE reanalysis 2 data, respectively. The arrow represents the values of 850 hPa wind field. over India (Raju and Bhatla 2014). However, previous studies mainly focus on the influence of ISM's intensity on TP's precipitation, and few pay attentions to ISM's duration. Understanding monsoon characteristics and predicting future variability are vital for agriculture, hydrology and social development.
There is a close relationship between the TP and the ISM by land-atmosphere-ocean interaction . ISM can influence the precipitation amounts over TP (Yao et al 2012) or watershed scales (Liu et al 2018). The earlier onset of the ISM accelerates lake expansion (Liu et al 2019) and enhances the wetting and greening of the TP (Jin and Wang 2017). ISM also can alters hydrothermal conditions through associated latent heat release through precipitation, and then exert impacts on the elevationdependence of meteorological variables (Yang et al 2018). The precipitation is the mainly source of water resources, and the runoff in the ISM domain are more than 60% of the total runoff for the 13 major rivers in the TP mountain outlets (Wang et al 2021a). The variability of monsoon rainfall not only influences local water resources, hydroelectric generation and agriculture across the TP and downstream environments, but also may trigger water-related hazards such as glacier collapse, drought or flooding (Yao et al 2019). Therefore, it is vital to quantify changes of the ISM characteristics and understand how these factors influence monsoon precipitation and surface runoff across the TP.
Previous studies mainly focus on the intensity or onset of the ISM. However, details about how the intensity and duration of the ISM influence precipitation across the TP, and the future challenges this region may face, remain unknown. Our study aims to quantitatively investigate the interannual variation of ISM characteristics in the past  and constrain changes in the future (2015-2100) with the help of the Coupled Model Intercomparison Project Phase 6 (CMIP6). We also investigate how ISM characteristics affect the amount and spatial patterns of precipitation and runoff across the TP, and project their trends until the end of this century.

Datasets
The data include daily mean wind on pressure 850 hPa from the National Centers for Environmental Prediction-Department of Energy (NCEP-DOE) reanalysis 2 project (NCEP2) (Kanamitsu et al 2002), monthly precipitation from Global Precipitation Climatology Project (GPCP) (Adler et al 2003), monthly precipitation and runoff data from ERA5-Land. All datasets are resampled to 1.0 • × 1.0 • resolution for analyses. A summary of the datasets is presented in table 1.
General circulation model (GCM) outputs from CMIP6 (Eyring et al 2016) are used to project future variability of ISM characteristics and monsoon precipitation and runoff. All daily variables for the historical period  and the future scenarios (2015-2100) are selected from the Shared Socioeconomic Pathways (SSPs; O'Neill et al 2016) 126 and 585 under r1i1p1f1 initial conditions. The SSP126 is a mitigation scenario, including SSP-1 and 2.6 W m −2 of radiative forcing by the end of the 21st century, while SSP585 is a high emission scenario including SSP-5 and 8.5 W m −2 of radiative forcing (Gidden et al 2019). All simulations are resampled to 1.0 • × 1.0 • resolution and model projections are calculated for the period from 2015 to 2100 under the SSP126 and SSP585 scenarios. Detailed information about models is presented in table 2.  (Syroka and Toumi 2004). When the HWSI is positive for seven consecutive days, we define this as the beginning of the ISM. The withdrawal date of the ISM is defined as the first of seven consecutive days when the HWSI is negative. The HWSI can capture both variability of the position and intensity of the monsoon trough (Syroka and Toumi 2004), so it can also be used to describe the intensity of monsoon. The index was calculated using the NCEP2 data. In the historical period, the calculated onset values were found to be strongly correlated with the monsoon onset from Ghanekar et al (2019) with a correlation coefficient of 0.53, indicated that the HWSI can characterize the transition of monsoon well. The ISM intensity is calculated using the normalized mean HWSI values from June to September during 1979-2018. The projected timeseries of duration and intensity were calculated during the future periods (2015-2100) based on CMIP6 simulations. The surface runoff was calculated based on ERA5-Land during 1979-2018 and CMIP6 datasets during 2015-2100.

Multiple linear regression, trend analysis and correlation coefficients
The multiple linear regression is used to detect the relationship between duration and intensity and monsoon precipitation in the historical period. The regression coefficients are output to detect their relationships. The linear trends and Pearson correlation coefficients are calculated using the least-squares method. Statistical significance is tested using the Student's t-test. The cumulative anomaly is calculated by cumulatively summing the anomaly after subtracting the mean. The mean climate is calculated using data from 1981 to 2010. In addition, to identify the dominant spatial patterns of monsoon precipitation, EOF (empirical orthogonal function) analysis was performed over the study region.

Evaluation criteria of model
To evaluate the simulation capability of CMIP6 models, four statistical values are used to compare the precipitation between CMIP6 simulations and GPCP. The quantitative assessment is mainly focused on five watersheds (Brahmaputra, Salween, Mekong, Yangtze, and Yellow river basins) around the southeastern TP (figure 1). Four statistical values are mean deviation (BIAS), root-mean squared error (RMSE), spatial correlation coefficient (R) and standard deviation (SD), where, ϕ gcm is annual time-series from GCM, and ϕ obs annual time-series is from observation (here is GPCP). ϕ gcmi and ϕ obsi are spatially multi-year averaged values from GCM and observations. ϕ gcm and ϕ obs are domain-averaged values from multiyear averaged GCM and observations. ϕ i andφ are spatially multi-year averaged values (from GCM or observations). N is the total grid numbers among the five watersheds.
To quantitatively evaluate the performance of CMIP6 models, two skill scores (S 1 and S 2 ) from Taylor (2001) are used to evaluate the simulation capability around the southeastern TP. The higher the skill score, the better the simulation ability of the model, where, R 0 is the maximum correlation coefficient available (here it is 0.999), R is the spatial correlation coefficient and σ f is the variance ratio between simulated and observed values. When the variance of the simulated values is close to the variance of the observed values, R is close to R 0 , and two skill scores are close to 1. When σ f is close to 0 or infinite or R is close to −1, two skill scores are close to 0. So S 1 focuses on the performance of variance, while S 2 focuses on the spatial correlation coefficient between simulated and observed values. In our study, we evaluate the performance of CMIP6 models with the help of S 1 and S 2 .

Variability of ISM characteristics
The advanced ISM onset appears in figure 2(a), possibly caused by aerosol-induced nonlinear effect (Bollasina et al 2013) with modulation by weaker lowlevel jet and deficit low-level moisture supply over the eastern Bay of Bengal (Choudhury et al 2019). However, the withdrawal of ISM is seldom concentrated and we find a postponed phenomenon. Since the late 1990s, the withdrawal of the ISM has become increasingly late (figure 2(b)). The combination of earlier onset and later withdrawal has increased the duration of the ISM, with a positive linear trend of ISM duration from 1979 to 2018 (figure 2(c)). The duration anomaly displays a trend of decreasing first and then increasing during 1979-2018. The ISM intensity significantly decreased from 1979 to 2018, with obvious interdecadal changes (figure 2(d)). In short, the ISM displays significantly increasing duration and decreasing intensity during 1979-2018.

Relationship between ISM characteristics and precipitation
The occurrence of the ISM has a significant influence on the amount and spatial pattern of precipitation. The relationship between the ISM characteristics and monsoon precipitation around the southeastern TP is examined using a multiple linear regression method (figure 3). Around the southeastern TP (region A and B), the influence from the intensity and duration on the monsoon precipitation has obvious differences. The intensity displays a significantly negative correlation with monsoon precipitation in the southeastern TP. In other words, the precipitation changes in the southeastern TP and Indian Subcontinent experience opposite variation trends, mainly triggered by sea-surface temperature anomalies in the tropical southeastern Indian Ocean (Jiang and Ting 2017). However, there is a positive relation between the duration and monsoon precipitation around the southeastern TP (region A and B). The increasing duration can bring more precipitation. In the downstream of Brahmaputra (region C), both of duration and intensity exert consistent  influence on monsoon precipitation, but the correlation between precipitation and duration seems weaker compared with that between precipitation and intensity. In short, both of duration and intensity exert a certain degree of influences on monsoon precipitation across the southeastern TP, but the degree of influence is different. Weaker and longer ISM is helpful for abundant precipitation around the southeastern TP. An EOF analysis is performed on monsoon precipitation to identify the dominant spatial patterns ( figure A1). In the first mode (EOF1), the southeastern TP shows an opposite spatial pattern compared with other regions. For the second mode (EOF2), there is an opposite pattern between region A-B and region C. A good consistence appears in between figures A1 and 3. The ISM's intensity is likely to influence the precipitation around the whole southern TP, corresponding to EOF1, while the relationship between the duration and precipitation may be reflected in EOF2. There is an important moisture channel around region C, called Yarlung Tsangpo Grand Ganyon, which transports moisture from the Bay of Bengal along the Yarlung Tsangpo River, facilitating the moisture to the eastern TP (Yuan et al 2023). Due to a series of high and steep mountains, the moistures from region C have to cross the mountains to reach region A-B. The complex terrain may be one of reasons to cause the precipitation difference around this region. In short, the ISM's intensity, duration and other factors together influence the precipitation changes around the southeastern TP.
In order to further reveal the influence of ISM characteristics on monsoon precipitation, the synthetic analysis is carried out. Time series of standardized ISM characteristics are calculated (figure A2). Years greater than half of standard deviation are regarded as anomalous years chosen to analyze the precipitation difference. Four groups are classified by different intensity and duration and the corresponding precipitation anomaly is averaged at all anomalous years in each group (figure A3). When the ISM is stronger and longer, the southeastern TP receives less precipitation. There is a similar spatial pattern when the ISM is stronger and shorter. A strong dipole pattern of summer precipitation is present when the ISM intensity is strong. When the ISM is weaker but has a longer duration, the spatial pattern of precipitation shows the opposite trend compared with that during a stronger ISM, with the majority of the TP receiving more precipitation. Precipitation decreases in northern TP and regions with more precipitation locate in southeastern TP when the ISM is weaker and shorter. Different combination of ISM's duration and intensity leads to different spatial patterns of precipitation across the southeastern TP.

Effects of ISM characteristics on potential water availability
TP's runoff is double influenced by large-scale atmospheric circulations and local climate variables (Chu et al 2018). Based on the significantly positive correlation between precipitation and runoff, we examine the influence of ISM characteristics on potential water availability, mirrored in the runoff (Wang et al 2021b). With the help of simulations from the hydrological models around the southeastern TP (Qi et al 2019, Wang et al 2021c, Chai et al 2022, the surface runoff from ERA5-Land is firstly evaluated. A comparable and similar distribution of surface runoff is found between ERA-Land and simulations (figure A4), with maximum in the downstream of Brahmaputra River and Ganges river and smaller values in the upper Yellow river and upper Yangtze river basins. ERA5-Land can reproduce the trends of surface runoff presented in the simulations, although accompanied by differences in some regions, such as in the upper region of the Yarlung Zangbo River basin (Liu et al 2021a). The possible reason is from snow melt and freeze-thaw of soil and glacier, which cannot be reflected in ERA5-Land. Overall, the ERA5-Land data show reasonable spatial distributions and trends of surface runoff around the southern regions of TP and can be used to detect the variation of surface runoff around this regions.
ISM characteristics correlate with surface runoff with a opposite distribution around the southeastern TP except for the downstream of Brahmaputra (figure 4). The ISM intensity has a negative influence on surface runoff as indicated by negative correlation coefficients, while the ISM duration shows a positive influence. The intensity displays a significantly and widely negative correlation with surface runoff in the southeastern TP. Both of duration and intensity exert a certain degree of influences on surface runoff across the TP through precipitation, but the degree of influence is different and opposite.
The surface runoff anomaly is also averaged at all anomalous years in four groups (figure A5). Different combination of duration and intensity brings different surface runoff. When the ISM is stronger and longer, the southeastern TP receives less surface runoff. This spatial range of less runoff becomes even more pronounced when the ISM is stronger and shorter. When the ISM is weaker but has a longer duration, the spatial range with positive runoffs becomes larger, nearly covering the whole eastern TP. Runoff decreases in nearly all the regions when the ISM is weaker and shorter. Weaker and longer ISM is helpful for abundant runoff around the southeastern TP except for the downstream of Brahmaputra.

Evaluation of CMIP6 models
To evaluate the performance of CMIP6 models around the southeastern TP, comparisons between GPCP and models are performed. We find that most of models can reproduce mean spatial distribution with less precipitation in northwest and more in southeastern TP (figure A6), indicated by high spatial correlation coefficients of 0.63-0.93 (table 3). However, most of models have a strong rain belt along the Himalayas, mainly caused by coarse resolutions of models. The precipitation bias ranges from  Table 3. Difference (Bias), spatial correlation coefficient (R), root mean square error (RMSE), ratio of the variance (õ f ), the skill scores (S1 and S2) between CMIP6 simulated precipitation and GPCP during 1980-2014.

Models
Bias R RMSEõ f S1 S2 0.28 to 2.70 mm d −1 , followed a RMSE from 0.77 to 3.80 mm d −1 . The ratios of the variance from GPCP and models are from 1.00 to 2.00, implying a comparable precipitation amount between GPCP and models. Skill score S 1 ranges from 0.22 to 0.83, and S 2 is between 0.10 and 0.64. We select models of S 1 and S 2 values greater than 0.50 as the criteria to choose models with better performance. Six models of ACCESS-ESM1-5, BCC-CSM2-MR, EC-Earth3, EC-Earth3-Veg, MPI_ESM1-2-HR and MRI-ESM2-0 are used to calculate the future variability of ISM characteristics.

Projected variability of ISM characteristics, precipitation and potential water availability
The ISM duration and ISM circulation have been unanimously predicted to increase and weaken under both scenarios during 2015-2100, more significantly under the scenario of SSP585 than under SSP126 ( figure 5). Under the SSP126 scenario, the spatial correlation between ISM intensity and monsoon precipitation is similar to the historical period shown in figure 3(a), with a strong negative correlation across the southeastern TP, indicating more precipitation will occur with the weakening ISM. However, under the SSP585 scenario, the negative correlation covers almost the whole TP, and is especially pronounced for the southeastern TP, possibly leading to increased precipitation across this region. Sabade et al (2011) pointed out that weakening ISM was likely to cause an increase in precipitation under different climate scenarios around TP. Assuming that the ISM will weaken and lengthen in the future, the precipitation across the TP is likely to increase, especially under the SSP585 scenario. More precipitation is identified under the SSP585 than SSP126 scenario, especially along the Himalayas to the southeastern TP ( figure 6(a)). Under the scenario SSP126, variance of precipitation explained by the changes of ISM intensity and duration is small. Under the scenario SSP585, explained variance is increasing, nearly covering the whole TP ( figure 6(b)). The influence of duration on precipitation is less than that from the intensity, as indicated by little explained variances, mainly along the Himalayas. The ISM intensity likely causes greater influence on precipitation than the ISM duration around the Himalayas and the southeastern TP under the scenario SSP585.
The major river systems of the TP are expected to be very vulnerable to climate change and variations in the environmental conditions (Immerzeel et al 2010). Future changes in the potential water availability have become increasing important to water resources management on the TP. It is better to reveal the variation of runoff using the precipitation data due to a better relationship between precipitation and runoff  . The mean ratio of runoff to precipitation is derived from ERA5-Land ( figure A7(a)). In most of the southeastern TP, the ratios are between 0.6 and 0.8, which turns out that 60%-80% of the runoff comes from precipitation. Even in some regions the ratio can be as high as 0.8-1.0. In other words, there is a strong and close connection between surface runoff and precipitation around this region. And this connection is relatively stable and the interannual variability of ratio is small, demonstrated by smaller standard deviations ( figure A7(b)).
We assume the ratio of runoff to precipitation is stable till to future period around the southeastern TP. So with the help of the ratio and projected time-series of precipitation from CMIP6 models, we calculate the time-series of projected surface runoff around TP and focus on the trends of surface runoff around the southeastern TP. Future projections show large increases in the runoff for the rivers originating on the TP (Lutz et al 2014), but displays a non-monotonic changes (Cui et al 2023). Compared with trends of precipitation, similar spatial pattern of surface runoff is found (figure 6(c)). We find an increasing runoff belt along the Himalayas to southeastern TP. Under the scenario SSP585, the increasing trend is more significant than that under the scenario SSP126, agreed well with the results from Khanal et al (2021). Except for temperature, precipitation is the another main factor to influence the hydrology in TP (Khanal et al 2021).

Discussions
In our study, the duration of ISM is predicted to lengthen under both scenarios derived from CMIP6 models. However, results from Sabeerali and Ajayamohan (2018) suggest that the ISM will shorten under RCP8.5 from CMIP5, while Lee and Wang (2014) found that northern hemisphere monsoon will lengthen induced by advanced onset and delayed withdrawal derived from CMIP5 models. This contrasting result may be caused by the use of different datasets and ISM indices and requires further discussion and validation. In this study, we projected changes of ISM's characteristics from the perspective of two extreme scenarios, SSP126, a mitigation scenario, and SSP585, a high emission scenario. Due to the sensitivity of precipitation to radiation forcing, the projected results are anticipated to reveal the possible changes of ISM's characteristics and its influence on precipitation under possible scenarios (SSP126-SSP585) in the future.
The TP is the source region of many large Asian rivers (Wang et al 2021a) and includes areas of glaciers (Yao et al 2012) and lakes , which are all fed by precipitation (Bibi et al 2018). In the last 50 years, the TP has experienced significant warming and frequent extreme climate events (Chen et al 2015). The rain days and cumulated precipitation amounts greater than 1.0 mm d −1 in the future (2015-2100) are projected to increase around the southeastern TP, and this phenomenon will become more significant under the scenario SSP585, followed by expanded area and increased values (figure A8). In the future, the weakening and lengthening of the ISM against the background of global warming will directly increase potential water availability by influencing precipitation changes. Part of the runoff and rainfall also infiltrates the soil and contributes to groundwater. The TP is highly vulnerable to geohazards like landslides, glacial-lake outburst floods and ice disasters (Yao et al 2019) and is sensitive to increasing population exposure to global extremes . Understanding how changes to monsoon precipitation will impact these geohazards is vital to help mitigate risks in the future.

Conclusions
Our analysis draws the following conclusions. The duration of the ISM lengthens from 1979 to 2018 due to an earlier onset and delayed withdrawal, followed by a weakening. Different combinations of duration and intensity cause variable spatial patterns of monsoon precipitation and the corresponding potential water availability. Weakening and lengthening ISM helps to produce more monsoon precipitation and potential water availability across the southeastern TP. The influence of the intensity of the ISM is significantly larger than that of duration. In the future, the ISM is projected to be weaker and longer until 2100, leading to more precipitation and more potential water availability across the southeastern TP. Therefore, an improved understanding and accurate prediction of the onset, withdrawal and intensity of the ISM is vital for water resource management and flood-risk monitoring.
All data that support the findings of this study are included within the article (and any supplementary files). Figure A3. Spatial distribution of the ensembled annual monsoon precipitation anomaly (mm) for four groups. The mean precipitation anomaly was calculated as the mean annual precipitation of the specific year, relative to the precipitation climatology during 1981-2010.