This study employed the European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis (ERA5) dataset, which spans the years 1979–2019 and has a horizontal resolution of 0.25°×0.25° (Hersbach et al., 2020). The ERA5 single level data and pressure levels data were used as the input to the Water Accounting Model-2layers (WAM2layers) models. The ERA5 single-level data include the 6-hourly surface pressure, total column water vapor and a set of vertically integrated moisture and flux variables (vertically integrated northward/eastward water fluxes in the forms of vapor, liquid, and ice). The hourly precipitation and evapotranspiration from ERA5 single-level data were also used as the input to the WAM2layers models. The ERA5 pressure level data include 6-hourly zonal wind, meridional wind and specific humidity at 16 pressure levels, which include 100, 200, 300, 400, 500, 600, 700, 800, 825, 850, 875, 900, 925, 950, 975, and 1,000 hPa. The surface temperature, convective precipitation and large-scale precipitation are also used to investigate the mechanism of the precipitation shift process. ERA5 has been compared to other reanalysis datasets, such as Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA2) and High Asia Refined analysis version 2 (HARv2) and it has been shown that ERA5 outperforms other reanalysis datasets for long-term trends and can reasonably reflect the summer water vapor characteristics over the TP (He et al., 2021; Yuan et al., 2021). For further information about the selection of ERA5 dataset, please see the description of Figure 1 in He et al. (2021).

To evaluate the transition from light to heavy precipitation, a new index was developed to measure the rate of change in precipitation intensity. In this study, a precipitation event is defined as a day with at least 0.1 mm/day of precipitation each day (Sun et al., 2007; Ma et al., 2015). Following previous studies (Shiu et al., 2012; Yu et al., 2022), based on precipitation percentiles at each grid, daily precipitation at each grid over the TP was ranked from light to heavy precipitation for the period 1979–2019. Then, 40 intensity bins are divided by the 0, 2.5th, 5th, 7.5th … 97.5th and 100th percentiles of daily precipitation for each individual grid, which represent an intensity interval of precipitation amount. Finally, regionally averaged precipitation histograms were constructed based on 40 intensity bins. The total number of days with precipitation events was used to calculate precipitation frequency, the average precipitation intensity of those events was used to calculate precipitation intensity, and the cumulative precipitation within each intensity interval was used to determine the total precipitation amount for each intensity bin. The slope of the histogram distribution of the precipitation amount trend was used to quantify the rate of intensity shift by least-square linear regression. This rate illustrates the transition from light precipitation to heavy precipitation. A positive slope value indicates a decrease or a relatively smaller increase in light precipitation and an increase in heavy precipitation. The negative value of slope indicates an increase in light precipitation and a decrease in heavy precipitation. Consistent with a previous study (Ma et al., 2017), we classified daily precipitation into three primary categories to make it easier to detect geophysical changes: light precipitation (less than 35%), moderate precipitation (35%–90%), and heavy precipitation (more than 90%). Due to the fixed threshold for these categories of precipitation, this classification may generate some uncertainty, but the basic characteristics of the precipitation trend across 40 intensity bins could support the conclusion.

The WAM2layers model is an updated version of the Water Accounting model, which was developed by van der Ent (Ent et al., 2014). In the original WAM2layers model, the input data are suitable for the European Center for Medium-Range Weather Forecasts interim reanalysis dataset (ERA-Interim), which has a resolution of 1.5°×1.5° (Ent et al., 2014). The time step for the original WAM2layers model is reduced to 15 min to keep the computation stable. However, considering the higher spatial resolution of ERA5, a smaller time step is needed to make the computation stable. Therefore, Xiao and Cui. (2021) modified the original WAM2layers model and reduced the time step to 10 min to fit the ERA5 data. Compared with the original model, we use the modified WAM2layers to calculate the PRR, which is derived from the moisture balance equation for atmospheric water vapor (Eq. 1; Ent et al., 2014). ∂ W ∂ t + ∂ W u ∂ x + ∂ W v ∂ y = E − P + α (1) where, as stated by Eq. 2, W represents the precipitable water in an atmospheric column, and u and v represent zonal and meridional winds, respectively. E , P , and α represent the evapotranspiration, precipitation, and residual terms, respectively. W = 1 g ∫ 100 h P a p s q d p (2) where q represents the specific humidity, g represents the gravitational speed, and p s represents the surface pressure. The relationship with the moisture from a certain target region is as follows: ∂ W Ω ∂ t + ∂ W Ω u ∂ x + ∂ W Ω v ∂ y = E Ω − P Ω + α (3) where Ω represents the target source region from which the moisture is evaporated. In addition, the precipitation is separated into two parts: the precipitation that originates in the target region ( P r ) and the precipitation that originates from the moisture that is advected into the target region ( P a ), which are called internal cycle precipitation and external cycle precipitation, respectively. The contribution of the internal cycle and external cycle was quantified by the ratio of the linear trends of internal cycle precipitation, external cycle precipitation and total precipitation. The PRR ( ρ ) can be calculated as follows: ρ = P r P r + P a (4)

The WAM2layers model can track the moisture in both forward and backward directions. Using the forward tracking approach, the PRR in each grid was calculated. In this case, the moisture from the target region evaporates E 0 into the atmosphere and is regarded as tagged water in the model. With a mixing ratio of r , the tagged water will mix with precipitable water in the atmosphere. As the model integrates with time, moisture diffuses horizontally and vertically with the wind. When precipitation P 0 occurs, the moisture from the target region contributes P 0 ⋅ r to the precipitation, and the tagged moisture will reduce the P 0 ⋅ r . This process will continue until no tagged water remains. The PRR on this day was assigned the values corresponding to convective and large-scale precipitation for a specific precipitation day that comprised both types of precipitation.

Because the atmosphere is not fully mixed, which will cause the uncertainty of WAM2layers, especially in regions with strong wind shear (Goessling and Reick, 2013), Ent et al., 2013 showed that the results of WAM2layers are similar to a detailed 3D model (Ent et al., 2013). Both the Euler and Lagrangian methods have uncertainties in moisture tracking, but the WAM2layers can track the real precipitation falling on the ground compared with the Lagrangian methods, which only track the moisture released in the air (Huang and Cui, 2015). Furthermore, WAM2layers can make the “tagged water” conserved (Zhang et al., 2017), which means all evaporated moisture can be tracked forward to the precipitation. In addition, the WAM2layers model has been used in a number of studies to examine wet and dry changes, trace the origin of moisture, and investigate the precipitation recycling process (Zhang et al., 2017; Guo et al., 2019; Zhang et al., 2019; Zhang, 2020; Li et al., 2022). Considering the above characteristics of the WAM2layers, the WAM2layers were suitable for calculating the precipitation recycling ratio (PRR).

The internal precipitation can be regarded as the product of precipitation and PRR, as stated by Eq. 5. P r = ρ ∗   P (5) where P r represents the internal precipitation, ρ represents the PRR and P represents the precipitation amount. All three variables will change to a certain amount under global warming, which is denoted by Eq. 6. P r ¯ + ∆ P r = ρ ¯ + ∆ ρ P ¯ + ∆ P = ρ ¯ ∗ P ¯ + ρ ¯ ∗ ∆ P + P ¯ ∗ ∆ ρ + ∆ ρ ∗ ∆ P (6) where the overbar represents the climatology of these three variables and ∆ represents the change in these three variables. Eq. 6 minus the climatology of these three variables and ignores the high-order term ∆ ρ * ∆ P , and we can decompose the change in internal precipitation into two parts, as shown in Eq. 7. ∆ P r = ρ ¯ ∗ ∆ P + P ¯ ∗ ∆ ρ (7) where ρ ¯ * ∆ P is the precipitation term, which represents the contribution of precipitation amount changes, and P ¯ * ∆ ρ is the PRR term, which represents the contribution of PRR changes.

Before examining the variations in precipitation intensity, the slope and trend map of precipitation over the TP were analyzed (Figure 1). Due to geography, the Brahmaputra Grand Canyon in the eastern TP is the principal moisture channel (Figure 1A). However, the trend map indicates that the increasing trend of precipitation over the TP is dominated by the increasing trend over the western TP and Qaidam Basin (Figure 1B). The decrease in precipitation in the eastern TP, particularly over the Brahmaputra Grand Canyon, is associated with the weakening of external water vapor transportation. As described in a recent study (He et al., 2021), the water vapor flux over the southern boundary of the TP exhibits contrasting changes between western and eastern 92°E. Timeseries of precipitation over the western TP and eastern TP suggest that precipitation increases significantly over the western TP at a rate of 1.462 mm/year and decreases insignificantly over the eastern TP at a rate of −0.142 mm/year (Figure 1C). Therefore, this study investigates the effect of the precipitation recycling process on the change in precipitation over the TP, eastern TP and western TP separately.

Based on observations and simulations, a few studies have demonstrated that the intensity of precipitation is shifting from light to heavy precipitation (Na et al., 2021; Tang et al., 2022). Hence, the trends in different intensity bins of precipitation are investigated over the TP, eastern TP and western TP. For the TP and western TP, precipitation intensity shows negative trends for intensity percentile bins below the 35th percentile and positive trends above the 35th percentile (Figures 2A, C). The most significant changes are found for percentiles between the 35th and 90th percentiles over the TP and western TP. For precipitation intensity percentile bins below the 35th percentile, the decreasing trends are significant over the western TP but not significant over the TP. The magnitude of the decreasing trend is much smaller than the increasing trend. For precipitation intensity bins above the 90th percentile, remarkable increasing trends are found over the TP and western TP, albeit they do not reach statistical significance over the TP. Precipitation intensity over the eastern TP increases below the 35th percentile and decreases above the 90th percentile, which is in contrast with the western TP (Figures 2C, E). The most significant increase is found between the 35th and 65th percentiles in the eastern TP. The increasing trends of percentiles below the 35th percentile are close to zero and do not pass the significance test. Over the eastern TP, precipitation decreases obviously above the 90th percentile. For example, heavy precipitation decreases by almost 0.4 mm/year.

To quantify the transition toward heavy precipitation, the rate of change in precipitation intensity was calculated (method details of the calculation of this index are described in Section 2.1). The shift rate for the entire TP is positive, with a rate of 1.384 mm/41 years (Figure 2A). This indicates that the precipitation intensity will transition 1.384 mm from light precipitation to heavy precipitation at the end of the 41-year record. The shift rate over the western TP is similar to that over the whole TP but with larger magnitudes, with rates of 5.666 mm/41 years (Figure 2C). In contrast to the TP and western TP, the eastern TP has a negative shift rate of −3.485 mm/41 years. The negative shift rate implies that heavy precipitation will decline over the eastern TP. Considering the consistent shift rates throughout the entire TP and the western TP, the western TP is more responsible for the shift in precipitation intensity over the TP than the eastern TP.

In this study, light, moderate and heavy precipitation are classified according to the precipitation histogram over the TP to enable discussion about precipitation features. The trends of precipitation amount, frequency, and intensity for light, moderate, and heavy precipitation were examined to better understand the characteristics of the change in precipitation intensity. The frequency of light precipitation is decreasing over the TP, western TP and eastern TP, whereas the frequency of moderate and heavy precipitation is increasing (Figure 2D). This tendency is consistent with the changes in precipitation amount (Figure 2B). The intensities all exhibit an increasing trend except for the eastern TP (Figure 2F), which is primarily caused by the decline in water vapor transfer from the Indian Ocean.

A recent study indicated that the precipitation recycling process has a significant impact on the summertime water vapor over the TP (He et al., 2021). Therefore, the spatial-temporal distribution of total precipitation and the internal and external cycles of the precipitation recycling process for the three types of precipitation were further examined to determine the effect of the precipitation recycling process on the precipitation shift (Figures 3, 4). The results indicate that moderate precipitation and heavy precipitation dominate the precipitation over the western TP and eastern TP and that moderate precipitation contributes more than heavy precipitation (Figure 3). More specifically, the amount of moderate precipitation is approximately 150–250 mm/year and 250–350 mm/year over the western TP and eastern TP, respectively, and the amount of heavy precipitation is approximately 80–140 mm/year and 100–250 mm/year over the western TP and eastern TP, respectively (Figures 3D, E, G, H). Light precipitation contributes no more than 30 mm/year and 45 mm/year over the western TP and eastern TP, respectively (Figures 3A, B).The internal and external cycles both significantly influence the tendency of light precipitation to significantly decrease over the western TP. The internal cycle of precipitation greatly increases the amount of moderate and heavy precipitation over the western TP rather than over the eastern TP. In contrast, the external cycle of precipitation recycling causes a significant increase in moderate and heavy precipitation over the western TP and a substantial decrease over the eastern TP (Figures 3C, F, I; Figures 4G–I). The opposite influence of the external cycle results in a weak and insignificant impact on heavy precipitation over the TP. Further investigation of the spatial distribution of the three types of total precipitation and the internal and external cycles of the precipitation recycling process exhibits consistent results with Figure 3, except for the northern part and southern part of the eastern TP (Figures 4G–I). Specifically, the increase in light precipitation consists of an increase over the southern part of the eastern TP and a decrease over the northern part of the eastern TP (Figure 4A). The internal cycle contributes to a general increase in light precipitation over the eastern TP, and the external cycle contributes to a slight increase in light precipitation over the eastern TP (Figure 3C). For moderate precipitation, internal and external cycles of precipitation both contribute to the increase in moderate precipitation over the eastern TP. The heavy precipitation is opposite to the increase in light precipitation, and the decrease in heavy precipitation is composed of an increase over the northern part of the eastern TP and a decrease over the southern part of the eastern TP (Figure 4C). The internal cycle precipitation generally contributes to the increase in heavy precipitation over the eastern TP, but the external cycle exhibits a spatial distribution similar to that of heavy precipitation.

To investigate the change in the internal and external cycles of the precipitation recycling process to the shift in precipitation intensity, the histogram of the trend in internal precipitation and external precipitation is calculated for the TP, western TP and eastern TP (Figure 5). The histogram of internal precipitation and external precipitation is similar to the histogram of total precipitation over the western TP and TP (Figures 5A–D). Both internal precipitation and external precipitation show obvious increasing trends above the 35th percentile, while they decrease below the 35th percentile over the TP and western TP. The magnitude of the decreasing trend is smaller than the magnitude of the increasing trend. However, for the eastern TP, the histograms of internal precipitation and external precipitation show contrasting distributions (Figures 5E, F). The internal precipitation increases across all percentile bins (Figure 5E). However, external precipitation increases slightly below the 80th percentile, which does not exceed the significance test, and decreases above the 80th percentile, especially above the 90th percentile.

Subsequently, to quantify the relative contribution of the internal and external cycles of the precipitation recycling process to the shift in precipitation intensity, the shift rate index is further calculated for the TP, western TP and eastern TP (Figure 5). The contribution was calculated by dividing the shift rate induced by internal and external cycles to the shift rate in total. This result suggests that the shift rate induced by the internal cycle is 1.207 mm/41 years (∼87.2%) and contributes more to the shift in precipitation intensity than the external cycle, which shifts at a rate of 0.177 mm/41 years (∼12.8%) (Figures 5A, B). Contribution was determined by dividing the shift rate of the internal or external cycle by the shift rate of precipitation. The shift rates of the internal cycle over the western TP and eastern TP show similar TP trends, with rates of 1.725 mm/41 years and 0.745 mm/41 years, respectively. However, the shift rates of the external cycle over the western TP and eastern TP are opposite, with rates of 3.941 mm/41 years and −4.23 mm/41 years, respectively.

Since the map of the precipitation intensity shift rate is more informative and crucial for authorities, the spatial distribution of the shift rate associated with the internal cycle and external cycles are further investigated in Figures 5G, H. The results show that the internal cycle strongly influences the shift in precipitation intensity over the western TP and the Qaidam Basin, while the shift rate generated by the external cycle increases west of the southern boundary but decreases east of the southern boundary (Figures 5G, H). The spatial maps of shift rates are consistent with the changes in light, moderate and heavy precipitation associated with internal cycle precipitation and external cycle precipitation (Figures 4D–I).

Considering the non-negligible contribution of internal precipitation to the shift in precipitation over the TP, western TP and eastern TP, the factors that affect the change in internal precipitation are further examined. As Section 2.4 states, the internal precipitation can be decomposed into the PRR term and precipitation term based on Eq. 5. Therefore, it is essential to comprehend the contribution of PRR and precipitation as a function of precipitation intensity to understand the change in internal precipitation under global warming.

The results suggest that the PRR term contributes more to heavy precipitation than to light precipitation (Figures 6, 7). For precipitation percentile bins above the 90th percentile, the PRR term and precipitation term both contribute to the increase in internal precipitation, and the magnitude of the PRR term exceeds the magnitude of the precipitation term over the TP (Figures 6A, B). More specifically, the PRR term contributes much more in the high percentile than in the low percentile, which indicates that the PRR term promotes the increase in heavy precipitation. For precipitation percentile bins between the 35th percentile and 90th percentile, the PRR term and precipitation term exhibit opposite contributions to the increase in internal precipitation over the TP. However, internal precipitation is more affected by the precipitation term since its magnitude is larger than that of the PRR term. For precipitation percentiles below the 35th percentile, both the PRR term and precipitation term cause a decrease in internal precipitation over the TP. For the western TP, the PRR term decreases across all precipitation intensity bins (Figure 7A), and the precipitation term decreases below the 35th percentile bins and increases above the 35th percentile bins (Figure 7B). Considering the magnitude of the PRR term and precipitation term, the precipitation term dominates the change in internal precipitation over the western TP. For the eastern TP, the PRR term increases across all precipitation intensity bins, and the precipitation term increases below the 80th percentile and decreases above the 80th percentile. In summary, the precipitation term dominates the decrease in internal precipitation in light precipitation, while the PRR term dominates the increase in internal precipitation in heavy precipitation over the TP.

In the summer, the average PRR over the TP is usually lower than 40% (Hua et al., 2015; Guo et al., 2018; Zhao and Zhou, 2021), showing that the internal cycle is far less than the external cycle. Non-etheless, the contribution of the internal cycle to the shift in precipitation intensity is larger than that of the external cycle. Therefore, it is essential to comprehend the variation in PRR as a function of the precipitation intensity to understand how the precipitation recycling process influences the shift.

Because surface temperature and evaporation are directly related to the precipitation recycling process (An et al., 2017), the linear trend of surface temperature and evaporation was evaluated as a function of precipitation intensity bins (Figures 8A, B). The result demonstrates that the surface temperature is increasing across all bins, with light precipitation experiencing the greatest increase. However, evaporation decreases with light precipitation and increases with moderate and heavy precipitation. The smaller increase in temperature during heavy precipitation can be partly explained by the evaporative cooling effect. The results also indicate that the precipitation recycling process weakens in light precipitation and strengthens in heavy precipitation, as shown in Figure 8C. More specifically, in the climatology of PRR, the climatological mean PRR of light precipitation (∼51%) is larger than that of heavy precipitation (∼35%), which indicates that the dependence of light precipitation on local evaporation is greater than that of heavy precipitation. However, the trend of PRR for light precipitation is negative, while it is positive for heavy precipitation, indicating that the dependence of light precipitation on local evaporation is decreasing, while the dependence of heavy precipitation is increasing (Figure 8C). Therefore, there exists an issue: why are the responses of light and heavy precipitation in contrast as the precipitation recycling process is strengthened?

To address this issue, convective and large-scale precipitation were separated from precipitation (more information is supplied in Supplementary Text S1). Convective precipitation has a larger PRR than large-scale precipitation and is a strong activator of the precipitation recycling process (Figure 8D). Consequently, Supplementary Figure S1 examines the influence of convective and large-scale precipitation on the shift in precipitation intensity. The results indicate that convective precipitation is more influential than large-scale precipitation on the shift. The trend of PRR for convective and large-scale precipitation over intensity bins (Figure 8D) reveals that convective precipitation has a consistent trend distribution in PRR with total precipitation, whereas large-scale precipitation does not differ significantly across intensity bins. More specifically, the large-scale precipitation PRR decreases across all intensity bins, and the convective precipitation PRR increases in heavy precipitation and decreases in light precipitation. Therefore, the changes in the PRR trend between light precipitation and heavy precipitation are mainly caused by convective precipitation.

The mechanism responsible for the precipitation shift over the TP is summarized in the schematic diagram (Figure 8E). Global warming causes an increase in atmospheric moisture and water vapor flux around the TP, which leads to increasing humidity over the TP through the external cycle of the precipitation recycling process. As discussed in (He et al., 2021), the contrast trend between the western TP and eastern TP is positively correlated with the changes in moisture flux at the southern boundary. Moisture flux mainly flowing over the Brahmaputra Grand Canyon brought by the India Summer Monsoon (ISM) has declined (Bingyi, 2005; Zhao and Zhou, 2021). This has reduced the external cycle of heavy precipitation over the eastern TP. Due to the strengthening of the heating source (Xie and Wang, 2019) and the “up-and-over” process (Dong et al., 2016), the moisture flux near the southern boundary of the western TP is increasing (He et al., 2021). In addition, the enhanced warming over the TP has caused glacier melting, lake expansion and increasing evaporation, all of which promote the intensification of the precipitation recycling process. Increasing moisture induced by both the internal and external cycles would help the transition of light precipitation to heavy precipitation, resulting in a decrease in light precipitation and an increase in heavy precipitation.

Because precipitation dominates summer evapotranspiration over the TP (Wang et al., 2018; Ma and Zhang, 2022), decreased light precipitation and increased heavy precipitation would result in a different response to the precipitation recycling process. Positive feedback is formed when a decrease in light precipitation leads to a decrease in evapotranspiration, which in turn leads to a decrease in convective precipitation and a further decrease in light precipitation. The decreasing convective precipitation during the aforementioned process also decreases the PRR in light precipitation, thereby decreasing the dependence of light precipitation on local evaporation, despite the intensification of the precipitation recycling process under global warming. The convective precipitation that is enhanced by an increase in evaporation leads to an increase in heavy precipitation, forming positive feedback to increase heavy precipitation. Increasing convective precipitation during heavy precipitation also increases the PRR, indicating that heavy precipitation is increasingly dependent on local evaporation. The contrast changes in PRR between light and heavy precipitation have a positive effect on the shift in precipitation intensity.

Therefore, an intensified precipitation recycling process could facilitate the transition to heavy precipitation by increasing evaporation and adjusting the dependence of light and heavy precipitation on the precipitation recycling process. However, the contrast changes in the external cycle between the eastern and western TP result in a weaker contribution of the external cycle over the TP. Although the internal cycle only amplifies the effect of the external cycle on the shift, its contribution even exceeds that of the external cycle. Further divide the internal precipitation into the PRR term and precipitation term and find that the PRR term contributes more than the precipitation term, especially during heavy precipitation, which heavily amplifies the effect of the external cycle (Figure 6). The TP is projected to warm significantly in the future, which will significantly intensify the internal cycle; consequently, the influence of the internal cycle on the shift is anticipated to increase.

The internal cycle and external cycle are important components of the hydrological cycle over the TP. A better understanding of the internal and external cycles over the TP can contribute to the early response to disasters for facilities over the TP. The decrease in light precipitation may contribute to more frequent drought over the TP. It is important to note the amplification effect of the internal cycle, which amplifies heavy precipitation by increasing evaporation over the TP. The amplification of heavy precipitation may also increase the occurrence of extreme precipitation and flood disasters, which could damage the infrastructure over the TP. As internal precipitation amplifies the precipitation shift over the TP, we suggest that the government take measures to mitigate the adverse effects associated with this shift in a warming climate.