Precipitation over South and East Asia in the ERA-Interim dataset has a maximum at the southern edge of the Tibetan Plateau and at the eastern border of the Bay of Bengal during April–September (Fig. ). While large regions receive more than 10 mm d-1 between April and September, the target region of this study, the lower reaches of the Yangtze River, denoted here as the Yangtze River valley, receives on average 5.9 mm d-1 of precipitation (Fig. , red box). YRV wet season precipitation shows a meridional gradient, with 4 mm d-1 in the north and up to 8 mm d-1 in the south (Fig. a). ERA-Interim's representation of precipitation shows a similar spatial pattern as the high-resolution, gridded observational dataset CN05.1 (Fig. a and b). The spatial pattern of precipitation obtained through the WaterSip method described in Sect.  is also similar (Fig. c).

The YRV has a pronounced precipitation seasonality. The six wettest months are April to September, and precipitation peaks in June (Fig. d). Both June and July are considered peak monsoon months . The monsoon precipitation and overall precipitation seasonality agree well between the reanalysis, observations, and estimated from the WaterSip method (Fig. d), with on average 5.9, 5.9, and 5.2 mm d-1 respectively. The similarity in both patterns, amount, and seasonal cycle of precipitation validates the use of the WaterSip method for this region. While there is an overestimation during most months of the year, WaterSip underestimates ERA-Interim precipitation in summer (JJA) with an average of 20.5 %. Of the precipitation estimated by the WaterSip method, 95 % is attributed to a source, while 5 % is not accounted for, for example due to moisture sources further back in time than 15 d. The remainder of this study is only based on the moisture that can be attributed to a source.

For each precipitation event in the YRV during the ERA-Interim period, we trace back to the moisture sources using the WaterSip method. The resulting average moisture sources for the YRV in April–May are shown in Fig. a. The shading can be interpreted as the contribution of evaporation to YRV precipitation in units of millimeters per day (mm d-1), equivalent to kilograms per square meter per day (kg m-2 d-1). Moisture contributions range from 0 mm d-1 in white to 1 mm d-1 in black. The maximum contribution is from southwest China, while large parts of Asia and the surrounding oceans contribute only small amounts. The 50th and 80th percentiles enclose 50 % and 80 % of the total moisture contribution by picking the grid points with largest contributions (Fig. a–c, dashed red lines). The extent of the percentiles denote the most important source regions. At the same time, the region between both red lines shows the importance of relatively moderate contributions spread out over a large area, contributing 30 % of the moisture.

Through the course of the wet season, the most intense source region to the YRV gradually moves closer and more northeast as the monsoon progresses (Fig. a–c). At the start of the wet season, in April–May, the most pronounced source regions are over the western part of South China and the Indochina Peninsula (Fig. a). In June–July the eastern part of South China becomes more important (Fig. b), and by August–September East China becomes the dominant moisture source to YRV (Fig. c). The largest source contributions for the 2-month climatologies are during June–July, when the sum of the moisture source contributions provide the YRV precipitation maximum. The mean precipitation in the YRV during June–July is 7.1 mm d-1 (Fig. d), with maxima of 8.8 mm d-1 in the south and west in the region (not shown). The overall maximum source is then over South China (Fig. b), with 0.97 mm d-1.

The 80th percentiles show equally pronounced changes (Fig. a–c). In April–May the 80th percentiles cover mostly land regions, the South China Sea, and small parts of the Western Pacific. In June–July the 80th percentile stretches out over India and the Indian Ocean, while in August–September it shifts further into the Western Pacific. The contribution from weaker moisture sources corresponds to the area between the 50th and 80th percentiles. The region between the two percentiles is the source for 30 % of the moisture, with a contribution below 0.2 mm d-1 in June–July.

Comparing each 2-month period to the overall wet season mean we obtain the 2-month anomalies (Fig. d–f). These anomalies are shown with green (red) values for above (below) average contribution. The anomalies highlight the northward and eastward movement of the most important sources through the wet season. In April–May (Fig. d) the southwestern edge of China and Myanmar contribute more than average, together with the South China Sea. In June–July (Fig. e) the Bay of Bengal, Indochina Peninsula, and South China contribute more than average. At the end of the wet season, in August–September (Fig. f), eastern China and the Yellow Sea are the only regions that contribute more than average. The reason for the high uptake within the YRV region in August–September is investigated further in Sect. . During August–September, the South China continental areas, the Indochina Peninsula, and the South China Sea all contribute less than for the preceding wet months.

Changes in the total contribution from all source regions are directly reflected in precipitation amount over the YRV. Variability in moisture sources is thus intimately connected to variability of YRV precipitation. The YRV moisture sources show large seasonal variations, reflecting the large seasonal precipitation variations.

The distribution of moisture sources for the YRV found here generally agrees with a range of previous studies . However, our results are also in disagreement with studies focusing exclusively on oceanic moisture sources or studies using the e-p method . The results of this study are not necessarily in a direct contradiction to previous work, as there are different views on the effect which land recycling and the evaporation of moisture for precipitation has on the definition of moisture sources. Furthermore, each method for studying moisture sources is associated with uncertainties. An advantage of the WaterSip method is that, instead of dealing with values for p and ET obtained from model parametrization, these variables are estimated through more observation-restrained humidity changes in the atmosphere. As a first-order result, the similarities to a range of previous studies using very different methods are encouraging. The results of this study are then further compared in more detail to the literature in Sect. .

The YRV moisture sources for April–September show temporal changes, both with respect to location and amount. To examine the changes and the seasonal progression in more detail, we subdivide moisture sources into six land and four ocean regions (Fig. ). The monthly climatology of the contribution from each region shows pronounced seasonality in all regions (Fig. ).

In Fig.  the peak contribution of all regions ranges from 3.5 to 8×1011 kg d-1, which corresponds to 0.37 and 0.82 mm d-1, respectively, if distributed evenly over the YRV region.

The spread of moisture contributions in Fig.  shows the interannual standard deviation of the moisture contributed from each source region. The regions with the largest standard deviation are the South China Sea (Fig. d) and the Western Pacific (Fig. e). The role of these and other sources regarding interannual variability and their connection to dry and wet years is explored further in Sect. .

Another feature of Fig.  is the timing of contribution from the different moisture sources. For spring and the pre-monsoon season (April–May), the South China Sea (Fig. d), South China (Fig. d), and the Myanmar region (Fig. b) are important moisture sources. They provide 16.1 %, 17.3 %, and 10.1 % respectively (Table ). Combined, this moisture contributes a substantial fraction of the pre-monsoon precipitation (43.5 %).

During the monsoon precipitation peak months of June–July, contributions from the distant westernmost moisture sources of the Bay of Bengal (Fig. b), India (Fig. a), and the Arabian Sea (Fig. a) show pronounced peaks. This coincides with the strong westerlies of the Indian monsoon and provides a link between the Indian and East Asian monsoons. These short-term, distant sources contribute 10.1 %, 6.0 %, and 8.6 % respectively in June–July (Table ), a combined 24.7 % of the total June–July moisture contribution. During June–July a peak in contribution can also be seen for the Indochina Peninsula (Fig. c) and South China (Fig. d), although these regions also contribute substantially in spring. These two land regions contribute 11.0 % and 14.9 % in June–July (Table ). The two land regions lie in the path of moisture arriving at the YRV from the Indian Ocean. We hypothesize that moisture transported from the Indian Ocean, which precipitates and re-evaporates on its way to the YRV, will have the en-route land regions as new moisture sources. This will be further investigated in Sect. .

For the late part of the monsoon, during August–September, the Western Pacific and the YRV region itself (Fig. e) become more important. During August–September these regions contribute 21.5 % and 15.4 % respectively (Table ). This is a time when the region also experiences a decrease in moisture contribution from all other moisture source regions, suggesting a changeover of moisture transport mechanisms in the monsoon system. This important transition period is further investigated in the next section (Sect. ).

Regarding the causes of YRV precipitation variability, we note that different source regions are responsible for providing moisture for precipitation in the different stages of the monsoon. It is therefore conceivable that different mechanisms, such as continental recycling or long-range moisture transport, can play a role at different stages of the monsoon.

During the summer months, the YRV receives almost equal contributions from land and ocean sources. For the April–September wet season months, land sources contribute 58.4 % of moisture for YRV precipitation while the ocean sources contribute 41.6 % (Table ). We refer here to contributions from land sources as continental recycling . This term has a wider perspective than local recycling, which only includes continental recycling from within the YRV (see Sect. ). Continental recycling varies between 66 % in February and 51 % in August (Fig. ). With more than half of the moisture provided through continental recycling, this mechanism appears as important in sustaining YRV humidity and precipitation during the wet season.

Previous studies which considered land contributions to YRV precipitation reported between 30 % and 60 % continental recycling for different seasons and slightly different target regions, with a gradient of lower continental recycling to the southeast in the region and more land contributions to the northwest . As studies used different regions and time periods, continental recycling values are not directly comparable. The continental recycling fraction found in this study is nonetheless higher than what was found in previous studies. To further investigate the plausibility of this result, we compare moisture contribution from continental recycling to the total evapotranspiration (ET) due to the land surface and vegetation.

ERA-Interim mean ET over South and East Asia in April–September shows a meridional gradient, with about 3–4 mm d-1 in the south and below 1 mm d-1 in the north (Fig. a). Ocean evaporation is stronger, with maxima of over 5 mm d-1 in the Bay of Bengal and Arabian Sea. ET values are generally lower, but the pattern resembles that of precipitation over the same regions (Fig. a).

While the ET displayed in Fig. a suggests a dominating role of the oceans, our analysis highlights that moisture source contributions to the YRV are only a subset of ET at the moisture source regions. The fraction ϵ of ET that ultimately arrives as precipitation in the YRV is calculated as 2ϵ=ETYRV,PETTOT, where ETYRV,P is the amount contributed from the moisture source to YRV precipitation, and ETTOT is the total ET. The ϵ within the YRV region is sometimes referred to as the regional evaporation recycling ratio , which shows the ratio of ET that subsequently precipitates within the same region.

For the April–September mean ϵ is mostly below 50 % (Fig. b) for all source regions. Values over South China are the highest, where more than 40 % of ET in some regions results as YRV precipitation. The highest values over the ocean appear over the Western Pacific by the Yangtze River outlet and the South China Sea by the Indochina Peninsula. The values of Fig. b underline that ET is clearly sufficient to fuel the moisture sources obtained in this study. This underlines the plausibility of the continental recycling values found in this study and our moisture source results in general.

YRV precipitation provided through continental recycling can itself have a local or remote origin. In this section we examine the sources of moisture from continental recycling arriving at the YRV, found according to the description in Sect. 

We start out with the continental moisture sources to YRV previously identified from the WaterSip method (Fig. a). Next, we calculate the local fraction of continental recycling for a larger region encompassing South, East, and Central Asia (Fig. b). The Asian continental recycling fraction shows a north-to-south gradient of decreasing continental recycling, coinciding with an increasing gradient with distance from the coast. Multiplication of the YRV moisture sources (Fig. a) and the Asian continental recycling fraction (Fig. b) yields the contribution of land sources to precipitation from continental sources to the YRV, termed here the second-order continental moisture sources (Fig. c). In this calculation, monthly averages were used to allow for a possible lag between precipitation and re-evaporation. While Fig. c does not show the regional extent of the second-order moisture sources, it does provide information on the amount of YRV moisture which still has continental origin even before the last continental recycling event.

These results show that about two-thirds (in summer) to three-fourths (in winter) of the land source contributions to YRV have their origin over land, while one-third (in summer) to one-fourth (in winter) comes from the ocean. In combination with earlier results on the direct land and ocean contribution to YRV, this implies that the YRV has 41.6 % of the direct ocean contribution for April–September precipitation and 17.6 % of continental recycling which is ocean contribution recycled once on land, while the remaining 40.8 % of moisture to YRV has been recycled over land at least twice (Table ).

South China and the Indochina Peninsula (Fig. a, dashed red lines) are the two most important exterior continental moisture sources of the YRV, contributing about 24 % of summer precipitation. As already highlighted in Sect. , the June–July peak of the Indochina Peninsula and South China moisture contribution might be connected to moisture transport from the Indian Ocean, precipitating and re-evaporating en route to the YRV. The 50th and 80th percentiles of the moisture sources for that region as a whole extend over South and East Asia, the surrounding oceans, and India (Fig. c, dashed red lines). The sources for South China and the Indochina Peninsula are therefore examples of important second-order sources of the YRV.

An advantage of the approach used here is the ability to quantify the degree to which moisture undergoes multiple recycling events (see Sect. ). The second-order continental sources show how moisture can be traced further back, sometimes back to when it evaporated from the ocean. Our results for the second-order sources emphasize the importance of the ocean in providing moisture which eventually undergoes continental recycling. They also reveal the substantial fraction (40.8 % in April–September) which is recycled on land more than once. Regarding the variability of the monsoon precipitation in the YRV, we note that the interaction with the land surface may therefore extend beyond the regions identified as first-order continental moisture sources.

Local recycling refers to the evaporation within a region contributing to precipitation within the region itself. Local recycling is therefore a subset of continental recycling (Fig. ). For the YRV, the local recycling peaks in August (Fig. a), a time when contributions from all other sources except the Western Pacific have decreased compared to earlier months (Fig. e). Local recycling in the YRV is thus important for sustaining precipitation in the later part of the summer monsoon. YRV precipitation is lower in August compared to June–July. Thereby, increased local recycling acts against a further decrease. The fractional contribution from local recycling increases from 9.8 % in July to its highest value of 15.8 % in August. A peculiar finding is that local recycling peaks 2 months after the peak in contribution from moisture sources outside the target region (Fig. b, note the different scales). In this section, we investigate the possible reasons for the peak in local recycling in August by an analysis of the seasonal evolution of characteristic variables of the YRV water cycle, including ET, soil moisture, NDVI, and local wind speed.

The ET rate within the region (Fig. c, blue) is high when local recycling peaks in August. However, ET peaks in July, when the fraction of local recycling is still quite low. The ET rate can be important for sustaining local recycling but can not in itself explain why local recycling peaks in August. Figure c (green) shows the time evolution of ϵ within the YRV, the same variable as in Fig. b. The fraction of recycled ET is relatively stable throughout the year, except for the months of July, December, and January. During 9 months of the year, including August, approximately 12.7 % of ET in the region returns as precipitation. In July, this part is reduced to 9.5 %.

The soil moisture in the region (Fig. d, pink) also shows a seasonality distinct to local recycling. The gradual increase in soil moisture from January to July can not explain the abrupt increase in the local recycling fraction from July to August. Since local recycling lags 2 months behind the precipitation peak, we explored the possibility that moisture from June or July precipitation was stored in the soil and affected August local recycling. However, interannual correlations of June or July soil moisture with local recycling in August are close to zero, both for absolute values of moisture contribution and the fraction of local recycling (not shown). We recognize that the soil moisture may participate in causing the late peak in local recycling but is not a driving factor.

The NDVI is a satellite-observed index for the density of green leaves . The NDVI average over the region shows a gradual increase from January onward (Fig. d). NDVI peaks in August and stays high in September, similar to the local recycling fraction. This means vegetation and moisture released through transpiration could help support the local recycling peak in August and perseverance in September.

Finally, to compare local recycling with the circulation in the region, we used the 850 hPa mean wind speed over the YRV as an index (Fig. e). The wind speed over the region has a marked peak in July, concurrent with the decrease in recycled ET (Fig. c). The stronger winds in July advect more moisture from distant sources, as well as a potentially stronger export of locally evaporated moisture. In contrast, weaker winds can increase the chances of locally evaporated moisture re-precipitating within the region during August and subsequent months. At the time of the local recycling peak the region experiences some of the lowest wind speeds during the year, favoring higher local recycling rates.

In summary, the comparison between local recycling and characteristic variables of the water cycle in the YRV suggests that a combination of factors is responsible for causing the late peak in local recycling and maintaining late monsoon-season rainfall. Decreasing winds, high soil moisture, high green leaf area, and high evaporation rates in combination lead to a sharp rise in local recycling and a slowed decline in rainfall seasonality in August. This suggests that rainfall variability in the late monsoon season is potentially affected by each of these factors, requiring a system-oriented approach to understanding variability of the YRV hydrological cycle.

To explore the effects of local factors on the interannual variability of moisture sources, we now focus on the five driest and wettest summers out of the 37 summers between 1980 and 2016 (Table ). Four of the five driest and wettest years in ERA-Interim are matched by WaterSip as the most extreme. For all summers (JJA) the YRV average WaterSip precipitation estimate is 1.31 mm d-1 lower than in ERA-Interim. The WaterSip summer precipitation deviations range from -4 % to -33 %, with an average of 20.5 %. This is a typical bias for Lagrangian diagnostics .

The total moisture supplied from all sources reflects the precipitation of the region, with anomalies of -22 % and +28 % during dry and wet years respectively. The relative changes in local variables are smaller (Table ). YRV contribution is higher during wet than during dry summers. However, during wet summers the local recycling fraction is lower, suggesting that during wet summers the contribution from outside the region increases more than local contributions. Local ET, soil moisture, and NDVI all change less than 5 % in wet and dry summers compared to the mean. The 850 hPa wind speed over the region shows the largest changes, with 10 % higher wind speeds for wet summers compared to the average and 5 % lower wind speeds during dry summers. The lower fraction of local recycling for wet summers and the high changes in wind speeds over the region suggest that outside contribution is more strongly connected to YRV precipitation variability than local moisture sources.

The small differences in local recycling between dry and wet summers motivate a comparison of moisture contribution from the different source regions for the five driest and wettest summers (Fig. ). Contribution from all sources except the Western Pacific follow the same pattern, contributing more than average in wet summers and less than average in dry summers. The Western Pacific breaks the pattern and provides the least moisture for wet summers and slightly less than average in dry summers. The changes between contribution in dry and wet summers are smallest for the YRV. This suggests that the YRV has a more stable contribution to summer precipitation than sources outside the domain and that contribution from the YRV does not intensify interannual variability. The largest changes in contribution between dry and wet summers are seen for South China and the Indochina Peninsula. South China provides -24 % during dry summers and +32 % during wet summers compared to its average summer contribution. The Indochina Peninsula provides -28 % and +53 % in dry and wet summers respectively. As these two land regions contribute a large fraction (24 %) of summer precipitation moisture, their variability also plays a large role in the interannual variability of YRV moisture sources and precipitation.

The South China Sea is the ocean region providing the largest amount of moisture in summer (4.0×1011 kg d-1), and the Western Pacific provides the second largest amount (3.7×1011 kg d-1). However, neither of these show the largest changes in contribution between dry and wet summers. The largest absolute changes between dry and wet years occur for the Indian Ocean sources of the Bay of Bengal (2.6 to 5.4×1011 kg d-1) and the Arabian Sea (2.2 to 4.5×1011 kg d-1). The South China Sea is next (3.2 to 5.2×1011 kg d-1). The Indian Ocean sources therefore seem to play the largest role for the interannual variability of YRV moisture sources and precipitation, with the South China Sea following slightly behind.

previously found that the major contributors of moisture influxes to different regions of China are not necessarily the major contributors to precipitation interannual variability. We arrive to a similar conclusion, although different study regions and methods hinder a direct comparison of our results. According to our findings, South China, the Indochina Peninsula, and the Indian Ocean contribute the most to YRV summer precipitation interannual variability. On the other hand, the YRV region, the South China Sea, and the Western Pacific, which are some of the major moisture sources, contribute less to interannual variability.