A variety of ocean-atmospheric processes controls the isotopic composition of precipitation. Since a significant amount of atmospheric water vapour is generated from the oceans, the physical conditions of the surface ocean leave their imprint on the isotopic composition of vapour and precipitation. Post-evaporation processes, such as the atmospheric circulation1, condensation of vapour into raindrops2, and their subsequent evaporation3, rain–vapour interaction4,5 also strongly affect the precipitation isotopic fractionation. Convective activity, moisture flux convergence, and cloud processes6,7 also modulate the precipitation isotopic composition. Compared to oceanic production, the continental production of moisture is relatively small; however, it may be significant in specific areas depending on the bio-geographical condition.

The oxygen isotopic (δ18O) profiles of precipitation on an intra-seasonal timescale often show a declining trend in some parts of India, such as Ahmedabad4, New Delhi, and Mumbai8, Port Blair9 in the Bay of Bengal (BoB) during the monsoon season (June to September). Such kind of declining effect is most dramatic in the northeast region of India (henceforth represented as NEI; Fig. 1a). For example, Breitenbach et al.10 observed a decrease in δ18O values ca. 15‰ at Cherrapunjee, Assam. Sinha et al.11 reported δ18O variations of Tezpur in NEI and observed a declining trend over 10‰. The cause of such depletion in δ18O is believed to be the modulation by ocean-atmospheric processes. A shift in moisture source from the north BoB to the equatorial Indian Ocean10 and large-scale planetary circulation was speculated12 to explain the reducing trend in δ18O. Recently, Cai and Tian13 invoked the concept of westerly and easterly circulations that produced higher and lower precipitation isotopic values, respectively, in the head Bay region. Their hypothesis, however, cannot explain the mechanism of generating the premonsoon high precipitation isotopic values. The reported Local Meteoric Water Line (LMWL) from NEI typically shows a slope like the Global Meteoric Water Line (GMWL; ~8) but intercept higher than 10‰10,12. This shows that a significant amount of recycled water vapour contributes to the local precipitation14. Earlier investigators mostly ignored the origin and role of this recycled water vapour.

Fig. 1: Study area and site description.
figure 1

a Map of northeast India depicting the study area. The precipitation samples collection sites are Tezpur (TEZ), Darjeeling (DAR), Kolkata (KOL), and Kakdweep (KAK; all in red dots). The Kaziranga forest site (KNP) containing the eddy-covariance tower is shown. The published data of precipitation isotopes are available from Jorhat (JOR), Cherrapunjee (CHE), and the sites in Bangladesh (Dhaka, Sylhet, Barishal, and Cox Bazar; blue dots) are also shown in this map. b The India map and its vegetation types are illustrated. As revealed in this diagram, the forest cover in northeast India is much higher than the rest of the country. c The climatological (1871–2016) variation of rainfall on a monthly scale for Assam and Meghalaya (blue bar) and Madhya Maharashtra (containing the Pune city) (dark red bar) has been shown. d The monthly variations of temperature and relative humidity for the above mentioned two regions are also shown.

Here we analysed several rainwater isotopes profiles from NEI, including a few other sites outside this region, to examine their intra-seasonal behaviour and find a more appropriate alternative explanation of this depletion. We propose that the land-surface processes mediated by plant transpiration play an essential role in controlling the precipitation isotopic behaviour during the preceding phase of monsoon in NEI. This produces a strong intra-seasonal precipitation isotopic gradient, denoted as LEMPIG (Late monsoon Early Monsoon Precipitation Isotope Gradient). The objectives of our study are: What processes control the isotopic variability of precipitation in NEI on an intra-seasonal timescale? Does plant physiology play any role in determining the precipitation isotopic composition? Whether the internal processes or the external factors play a significant role in modulating the precipitation isotope values in this region? We follow a multi-disciplinary approach to address these questions. First, we examine a comprehensive data set of precipitation isotopes from across this region and analyse the regional meteorological parameters. Second, we analyse the latent heat flux (LE) and net ecosystem exchange (NEE) measured by an eddy-covariance (EC) system, along with the reanalysis and observed rainfall data. Third, we estimate the recycled ratio (RR: ratio of recycled to the total rainfall) of rain on a seasonal scale to constrain the isotopic observations. To better comprehend the precipitation isotopic characteristic of NEI, we also present the isotopic data of another urban, semi-arid region in western peninsular India (Pune area). This region has a contrasting behaviour in terms of the moisture source, underlying land surface processes, and intra-seasonal variations of precipitation isotopes. Finally, we compare the precipitation isotopic behaviour of these two contrasting regions to further demonstrate the linkage between precipitation isotopes and biosphere–atmosphere interaction.


Characteristics of the precipitation isotopic profiles

Rainwater oxygen isotopic values for Tezpur and Darjeeling are shown in Fig. 2a, b, respectively. The patterns of all these profiles show relatively high values during the premonsoon season (March-May) and lower values during the late phase of monsoon (August-September). The isotopic values show low variability in both sites until May; afterwards, they show significantly large variability. For Tezpur, the δ18O values approximately range from +2 to −18‰. Darjeeling shows a higher range, from +3.5 to −21‰. Apart from their decreasing pattern, the isotopic profiles showed pulses of high negative values (grey bars; Fig. 2).

Fig. 2: Precipitation isotope records.
figure 2

Multi-year oxygen isotopic precipitation records for Tezpur (a) and Darjeeling site (b), respectively. Lower panel c: box plot of the climatological variation of δ18O for north-east India; d meteoric water line for NEI. e Monthly variation of NEI meteoric water line slope (red dot) and intercept (blue dot). Error bars indicate standard deviation.

Figure 2c shows the general behaviour of precipitation δ18O representative of NEI (see Supplementary Note 1). Isotopic values are enriched (≈1‰) during the early part of the year and then decrease progressively. We observed a sharp decrease in June (average value: ~−4‰). The δ18O values decrease and achieve the minimum value ca. −10‰. Subsequently, δ18O rises and remains approximately −4‰ in November-December.

We have also plotted an LMWL (Fig. 2d) representing this region. The slope and intercept of this line are 8.02 ± 0.08 and 13.02 ± 0.53‰, respectively (summarised in Supplementary Table 1).

We show the slope and intercept values calculated on a monthly timescale in Fig. 2e. The area receives significant rainfall in May. As a result, the humidity level goes up (≥80%, Fig. 1d), which helps the slope to increase from a low value of ~6.9 in April to ~7.8 in May. Monsoon sets in June (see Supplementary Table 2), and the humidity is further increased, so is the slope (~8.2). Subsequently, the slope is reduced and maintains 7.8 to 7.6 in the remaining monsoon season. We quantify the LEMPIG as the difference between the average values of δ18O in March–April and September–October for a specific site.

Biosphere–atmosphere interaction

It is a fact that the photosynthesis by plants is strongly coupled with their transpiration process15. An enhanced rate of carbon assimilation usually favours an increased rate of transpiration. Our earlier work on the EC-based estimation of biosphere-atmosphere CO2 fluxes in the Kaziranga National Park (KNP) reveals maximum photosynthetic carbon assimilation during the premonsoon season16,17. In the present study, we have analysed the EC-based CO2 fluxes to determine the diurnal variation of NEE on a monthly timescale during 2016–2018 (a detailed discussion of the EC results is presented in Sarma et al.18).

Seasonal variation of recycled rain (RR)

It is known that NEI receives considerable rainfall through recycled moisture. Chowdhury et al.19 examined this behaviour on a climate scale (1979–2014) for the monsoon season. They showed that about 7% of total rainfall during the monsoon season is contributed by local moisture supply (i.e., evapotranspiration (ET)). Here we have done a similar exercise, but for each season (DJF: winter; MAM: premonsoon; JJA: monsoon; and SON: postmonsoon) separately to examine the seasonal evolution of recycled rainfall). It is clear from this estimate (Supplementary Fig. 1) that the contribution of recycled rainfall to precipitation is minimum during the monsoon season (June to September). Then it progressively increases and reaches a maximum during the winter (November–December). After that, it slightly decreases, still maintains high values until the premonsoon season. The monsoon sets in around June, and marine water vapour enters the region. As a result, the proportion of recycled moisture in rainfall decreases.


Figure 2a, b show the general characteristics of precipitation isotopic variability at the Tezpur and Darjeeling sites, respectively. The range of isotopic fractionation at Darjeeling is considerably higher than that observed at Tezpur. Since Darjeeling’s (2186 m) elevation is much higher than that of Tezpur (48 m), the isotopic values are more depleted at the former site due to the ‘altitude effect’20. Figure 2c shows the climatological variation of δ18O over NEI. We observe relatively higher values of δ18O from January to May because of the ET-derived moistures; we will discuss this in the subsequent sections. With the advent of the monsoon season in June, a sharp drop in the isotopic values is observed. Figure 2d shows the LMWL representing the entire NEI region. The slope (8.02 ± 0.08) of this LMWL is the same as that of the GMWL. Still, a slightly higher value of intercept (13.02 ± 0.53) suggests a relatively higher proportion of ET-derived water vapour in this region. High humidity during the monsoon season helps maintain a slope value of 8.0 ± 0.2. During the late winter and spring, the slope is typically lower than 8, showing kinetic fractionation. During the postmonsoon season, the humidity level drops, and the proportion of ET-generated moisture increases, resulting in a slight rise in isotopic values. Hence, the slope decreases to some extent.

Similarly, the intercept shows a relatively high value in the early year, representing significant secondary moisture contributing to the rainfall. The high influx of marine water vapour during the monsoon causes the intercept and its variability to decrease. The intercept increases again in the postmonsoon because of the admixture of moisture from multiple sources. The patterns of slope and intercept of the LMWL, show consistent behaviour with the RR estimation.

During the premonsoon season, NEI receives abundant rainfall due to the thunderstorm activities21, because of this, the soil water content increases. The photosynthetic activity is enhanced and so is plant transpiration. To track the photosynthetic activity, the EC-derived net carbon fluxes are plotted in Supplementary Fig. 2, which illustrates that the maximum carbon uptake (CO2 flux from the atmosphere to biosphere ~20 μmol m−2 s−1) takes place during the premonsoon season. On the other hand, the minimum carbon transfer (~6 μmol m−2 s−1) is observed during the winter (Supplementary Fig. 2). The meteorological condition drives the seasonal carbon uptake characteristics of the KNP ecosystem. Deb Burman et al.17 analysed the photosynthetic photon flux density on a seasonal scale and concluded that the photosynthesis and ET are most and least strongly coupled in the premonsoon and winter seasons, respectively. Hence, the transpiration process of the forests in this region, interlinked with their carbon uptake, is expected to leave its signal on the precipitation isotopes of this region.

Further, the EC-based observations at KNP show that the LE is maximum during May–June16,18. The monthly LE estimated on a regional scale shows a peak in May and another slightly smaller peak around September, driven by the enhanced transpiration; (Supplementary Fig. 2; also see the Supplementary Note 2 depicting the relationship between the LE derived from the EC system and the ERA5 reanalysis data). It is to be noted that the precipitation δ18O also shows higher values during early May, which coincides with the high transpiration flux (~3.5 mm d−1) (Supplementary Fig. 3a). It is well known that an enhanced soil evaporation rate makes the remaining soil water isotopically enriched22. This isotopically enriched water, when ingested by plants, the transpired water also becomes isotopically enriched22. However, during the heavy rain condition, due to the reduced rate of soil evaporation, the isotopic composition of transpired water would be close to that of the rainwater23. A similar situation prevails in NEI during the premonsoon season.

On the other hand, a higher amount of secondary moisture does not necessarily yield enriched oxygen isotopes. This is because the evaporation of soil water releases isotopically depleted water vapour into the atmosphere22. So, the enrichment of 18O in precipitation must originate from an enriched source of moisture, which in this case can only be obtained from the high amount of transpired water vapour. The appearance of the second peak in LE (~105 W m−2) during the latter phase of monsoon (Supplementary Fig. 2) shows an increased ET (~3.8 mm d−1; Supplementary Fig. 3a), which although is unlikely to be associated with an increased number of heavier isotopes. This is clear in the progressively decreasing values of precipitation δ18O.

The increased abundance of heavier isotopes during the premonsoon season and its lower proportion during the monsoon season could also be investigated by analysing the dependence of δ18O on LE. The LE is generated when the water is vapourised, driven by evaporation and transpiration. Since the evaporation releases lighter isotopes faster from the soil surface, the precipitation isotope is expected to show an inverse, albeit weak, correlation with the LE. If the proportion of heavier isotopes were increased, the pattern of δ18O–LE correlation would change. The inverse relationship is likely to become neutral or weakly positive. Our investigation shows that, in general, the precipitation δ18O is negatively correlated with the LE averaged on a daily timescale. Figure 3 shows the scatter plots between the LE and precipitation isotope data. We chose the Tezpur site isotope record because of its proximity to our EC tower location. In this context, we like to emphasise that the δ18O and LE would show a strong correlation when the entire amount of water vapour emitted through the ET process is condensed and precipitated as rainfall, i.e., ET and precipitation, make a closed system. Since such a scenario is not possible on a local to regional scale, δ18O and LE would only show a weak correlation. Such a weak correlation characterises the monsoon season. But since the premonsoon season releases a relatively more number of 18O than the monsoon season through transpired water vapour, the LE–δ18O correlation during this time not only increases but also turns positive. The change in sign of the correlation value (Fig. 3e, f) provides firm support for our hypothesis; i.e., premonsoon season releases a greater number of 18O through the ET process.

Fig. 3: Relation between latent heat flux and precipitation δ18O.
figure 3

Scatter plots for LE and precipitation δ18O for a Kaziranga forest (the year 2016), water sampling site was close to the eddy-covariance tower (~1 km). b Tezpur site (year 2016), c Tezpur site (year 2017), d Tezpur site (year 2018). Barring 2017, in general, an inverse correlation was found in all the cases. The remaining plots in the lower panel show the same scatter plot but on a seasonal time scale combining all year data. e Premonsoon (March–April–May) and f monsoon (June–July–August) seasons. A positive trend in MAM is indicative of the presence of an increased number of heavier isotopes (18O) in the water vapour flux relative to that of the JJA season.

Further, we have 1 year (2016) of data available near the tower site (~1 km off the tower), which has also been used for this LE–δ18O analysis (Fig. 3a). Each year but 2017, shows an inverse correlation. In 2017, the EC system failed from mid-June to September. As a result, we could not get the LE data, and hence the LE, δ18O paired data were not available for the above period, yielding fewer data points during the monsoon season this year.

To examine the behaviour of LE and δ18O interrelation on a seasonal timescale, we have compiled the LE, δ18O pairs available for all the years comprising premonsoon and monsoon seasons. The winter and postmonsoon seasons are excluded because of the scarcity of data. As shown in Fig. 3e, the LE–δ18O correlation during the MAM season shows a gentle increasing trend. The same in the JJA season shows a moderately declining trend (Fig. 3f). The positive trend observed during the MAM season provides firm evidence of an increased number of heavier isotopes (18O) because of enhanced transpiration. Subsequently, the negative trend observed during the monsoon season implies a drop in the number of heavier isotopes but an increase in the lighter isotopes (16O) because of the incursion of the marine water vapour, depleted in isotopic value.

The parameters mentioned above in 2017 show a weak positive trend; moreover, most of the samples in this year belonged to April and May, some samples to June, and only a very few samples to September. So, this result attests to our hypothesis that the premonsoon time releases a relatively greater number of heavier isotopes through the ET process. It is to be noted that the role of the ecosystem in producing enriched isotopic values during the premonsoon season is also supported by the analysis of remote sensing data24.

The climatological variations of ET and T (transpiration) (Supplementary Fig. 3a) show that the T is maximum during late April and early May; subsequently, T decreases steadily. As mentioned earlier, an enhanced rate of soil water evaporation leads to the enrichment in soil water isotopic composition and is manifested in the transpired water vapour. Supplementary Fig. 3b shows the monthly climatology of the soil water content at the surface (0–10 cm) and the root zone (10–800 cm) for this area (obtained using the Global Land Evaporation Amsterdam Model (GLEAM)25 output). As depicted in this figure, the soil water content during March shows a minimum value; subsequently, it increases sharply, facilitating rapid evaporation. With the advent of the monsoon season, the soil water evaporation rate reduces; hence the transpired water vapour produces relatively depleted isotopic values. Afterwards, soil water decreases moderately during the postmonsoon season. The occurrence of heavy rain during the monsoon season and subsequent retention of a high amount of soil water even during the postmonsoon season does not allow soil water to be isotopically enriched. Considering the rainfall recycling, a higher fraction of evaporated water vapour than the transpired water vapour causes relatively higher RR during the postmonsoon season. The transpired water vapour fraction using the GLEAM data indeed supports this assertion. Hence, the transpired water vapour during the monsoon and postmonsoon seasons cannot produce isotopically enriched rain compared to the premonsoon season.

We have calculated yet another parameter, the canopy-atmosphere decoupling coefficient (Ω, dimensionless) to describe the relative roles of biospheric and environmental processes in controlling the total water vapour exchange at our study site. The ET primarily comprises transpiration, a biologically controlled water vapour flux governed by plants’ stomatal opening and closure. The second component is evaporation, a physical vapourisation of water. In the absence of the separate measurements of these components, we use Ω to infer the relative controls of these two parameters on the ET at KNP. Ω has been defined as the ratio of environmental to physiological controls of ET15 (vide the ‘Methods’ section for details). Lower values of Ω signify the greater contribution of transpiration in ET and higher values of Ω stand for the increased role of evaporation in ET. As detailed in Deb Burman et al.26, Ω is lower (~0.25) in premonsoon (MAM), highlighting increased transpiration during this season. Further, Ω is higher (0.35) in monsoon (JJA), showing that evaporation is the principal contributor to LE in this season. Overall, these findings support our isotope-based observations.

As mentioned earlier, several lows characterise the isotopic values during the monsoon season. Such periodic low values have been observed at several places and are believed to be associated with intense convective activities27,28. But the atmospheric controls of precipitation isotopes on intra-seasonal timescale at NEI and other Indian sites, such as the coastal region of Kerala28 may not be the same. The mean residence time of moisture in the atmosphere is over a week29; the cumulative outcome of these convective processes imparts a ‘memory effect’ resulting in a declining trend in the precipitation isotopes in NEI. But unlike NEI, a strong monsoonal circulation heavily influences the coastal and inland areas of Kerala28. As a result, the memory effect of the isotopic depletion is not sustained, which helps to get back the isotopic ‘background’ value relatively quickly. So, the condition to have a significant LEMPIG is that the area under consideration should be relatively free of external atmospheric influences.

Distinct low values also characterise the isotopic rainfall values in the head BoB region and possess a declining trend. However, it may be noted that the sea surface temperature forcing provides the major amount of moisture and drives the convective activities in the north BoB region30,31. On the other hand, orographic influence and land surface processes strongly affect the precipitation in the continental area of NEI. Because of these different feedback mechanisms, the isotopic depletion observed in the coastal region (Kolkata and Kakdweep) is characteristically different from that found in NEI. The interaction between the oceanic moisture and that generated through the land-surface processes and its modulation by the regional-scale convective activities and moisture recycling determine the isotopic composition of precipitation in NEI.

In NEI, convective events are triggered when the environment is sufficiently moisture-loaded and strong vertical winds carry moist air parcels aloft. In general, strong precipitating convection is expected in moist atmospheric conditions32,33. We have used the vertical wind speed (w-wind) data of the KNP tower site to examine whether higher wind speed, a proxy for convective activities34, is associated with the lower isotopic values. Figure 4 shows the variation of δ18O and w-wind observed at Tezpur during premonsoon to postmonsoon, for three years (2016 to 2018) of our study. The associated rainfall events obtained from the multi-component weather sensor at KNP are also shown (blue bars). In the case of 2016, the higher w-wind values are usually associated with the low δ18O values, though in one case (May 15, 2016), δ18O lagged high w-wind by about a week. But in most other cases, the low δ18O values occur near-synchronously with the high vertical wind. We observed similar behaviour in 2017 as well. Isotopic depletions follow three high w-wind events. Due to the non-functioning of the sonic anemometer, no wind data was available from the first week of July to the end of August. Hence, we could not examine this behaviour during this period in 2017. But again, in September, one such event was observed. In 2018, one large depletion in δ18O was associated with a high w-wind value. However, several high w-wind values were observed during March and April in 2018, but due to low moisture loading, sufficient convection could not take place32. Hence, isotope values did not suffer significant depletions.

Fig. 4: Co-variability of Tezpur precipitation δ18O and the w-wind.
figure 4

The daily time-series plots show the co-variability of δ18O and the w-wind parameters for years a 2016, b 2017, and c 2018 for the Tezpur site. The link between the isotopic lows and the vertical wind speed represents the convective processes. The diagonal arrows are marked where low isotopic values are associated with high w-wind. Blue bars show the daily rainfall amount.

The association between high w-wind speed and rainfall δ18O, operating within a spatial scale of ~50 km, shows that the local-scale processes significantly influence the isotopic depletions. To examine the effect of monsoon circulation, which operates on a much wider scale, we analysed the impact of moisture transport fluxes over the study region. The scaling of moisture transport in NEI is explained in Supplementary Note 3.

We calculated the vertically integrated moisture transport (VIMT) to examine the effect of moisture transport and regional scale convective activities on the precipitation isotopes, specifically for the monsoon season. This is depicted by the time-latitude Hovmöller diagrams of the average of National Oceanic and Atmospheric Administration (NOAA) interpolated outgoing longwave radiation (OLR) anomaly in the backdrop of the isotopic profiles. Figure 5 shows the results for 2015 and 2016, while the same for 2017 (Supplementary Fig. 4) and 2018 (Supplementary Fig. 5) are presented in the Supplementary Information section.

Fig. 5: Precipitation isotopes and convective activity.
figure 5

Relation between the precipitation isotopes (lines) and convective activities (represented by shading; the OLR anomaly) are shown. a The link between the vertically integrated moisture transport/convective events and the precipitation isotopes at four sites, Kolkata, Tezpur, Darjeeling, and Barisal. The Kakdweep data (white boxes) is also shown along with the Kolkata record. The upper right panel shows the rainfall anomaly observed during a cyclonic event ‘Komen’ during which high VIMT was found. But the anomaly had little effect either at Darjeeling or Tezpur. b The same as above but for the year 2016. Similar behaviour was observed concerning the control of the VIMT on the precipitation isotopes. A regional-scale convective activity (shown as an orange arrow at the bottom panel) happened soon after a deep depression (Roanu) around May 20 (lower right panel), which caused near synchronous precipitation isotopic depletions in all three sites.

The co-occurrence of the isotopic depletions with high VIMTs indicates the influence of the monsoon-induced moisture transport on the precipitation isotopes. In 2015, out of three intense VIMT events (see arrows), two events did not show any link between high VIMT and depleted isotopic values. A significant isotopic depletion took place in the last week of July. During this time, a deep depression formed in the north BoB (on 27 July, 2015), which intensified into a severe cyclonic storm Komen35. On July 30, very heavy rainfall was reported in Bangladesh and India. As a result, the coastal site of Sundarbans experienced a drop in δ18O by ~12‰ within a week.

We also observed a near synchronous reduction in δ18O in Kolkata and Barisal. However, the δ18O values at Darjeeling indicated that the Komen isotopic signature did not propagate toward the north, as no clear signature of this cyclone was observed (in Fig. 5a). The δ18O profile steadily decreased from the high values in premonsoon to the low values in the late monsoon with a LEMPIG of 12.5‰. The Tezpur site also did not show any Komen signature in its precipitation isotopes. Hence, the isotopic anomaly created by this cyclonic event was confined within its spatial extent (~100–200 km). We also examined the propagation of the rainfall anomaly further north. The right panel in Fig. 5a shows a high positive rainfall anomaly over the coastal region, but the anomaly hardly reached farther north. As a result, no high isotopic depletion was observed in Darjeeling or Tezpur. However, the regional-scale convective activities could occasionally cause concurrent low values in the precipitation δ18O. For example, a near synchronous negative δ18O anomaly occurred in all the sites during the fourth week of June. The convective activities spreading across this region caused the depletion, as revealed in the Hovmöller diagram of the interpolated OLR (Fig. 5). The VIMT showed a slightly high value during this time. Another low-intensity convective activity took place in the second week of July.

We observed smaller δ18O depletions at some sites but did not find a concurrent high VIMT. Similar behaviour is observed in 2016 as well, in which the isotopic profiles at Kolkata (including Kakdweep), Tezpur, and Darjeeling are examined (Fig. 5b). It is evident from this plot that the high VIMT values did not always associate with the low precipitation δ18O. However, one distinct low δ18O was observed in all three sites in the third week of May. On 17 May 2016, the VIMT showed a high value. In the subsequent days, all three sites experienced low δ18O. For example, the Kolkata, Tezpur, and Darjeeling sites had low δ18O values on 20, 23, and 24 May. The negative excursion of δ18O at Kolkata was caused by a deep depression Roanu. However, its associated rainfall was not significantly observed at far distant sites in Darjeeling and Tezpur. We calculated the rainfall anomaly during this event and observed its spatial distribution across the region to further check this assumption. As shown in Fig. 5b (right panel), a high positive rainfall anomaly existed in the coastal area of north BoB. The rainfall anomaly weakened considerably towards Tezpur, and practically no such effect was observed in Darjeeling. But it is interesting to note that a high w-wind at the KNP site was observed just before this event (Fig. 4a), which acted as a precursor of a regional scale convective activity. Around the same time, the VIMT also showed an increased value (see the orange arrow in Fig. 5b lower panel), and OLR reduced considerably, as evident in the OLR anomaly plot. As a result, the precipitation δ18O registered a near-synchronous depletion across the region.

In 2017, the Tezpur site recorded six or seven low δ18O events (Supplementary Fig. 4). Still, those are not caused by the regional scale convective activities, as revealed by the OLR anomaly plotted on a Hovmöller diagram. Instead, some of the low δ18O values seem to be associated with high w-wind. For example, −7‰ in mid-June 2016, −12‰ in mid-August 2016, −9‰ in May 2017, −14‰ in June 2017, and −17‰ in June 2018 (Supplementary Fig. 5) typically correspond to the high w-wind values as depicted in Fig. 4. An increase in the w-wind values shows the influence of the local-scale convective activities (as described in the previous section). A regional-scale convective event was observed in the middle of May. But as depicted in the δ18O plots, no significant drop in δ18O was observed at any of these three sites. By the end of May, a severe cyclone Mora developed in the north-central BoB, making landfall near Chittagong in Bangladesh (approximately 500 km east of Kolkata). During this event, the rainfall anomaly (Supplementary Fig. 4 right panel) and the rainfall isotopic records of Kolkata and Tezpur (Supplementary Fig. 4 left panel) show Tezpur to experience moderate rainfall. As a result, the isotopic values were reduced to about 6‰ relative to the preceding days, but the Kolkata site suffered from no such isotopic depletion.

In 2018, the isotopic profiles at Tezpur and Darjeeling (Supplementary Fig. 5) showed a concurrent low δ18O event around the middle of June. Though the Kolkata site also experienced depleted isotopic values around the same time. The extent of depletion was not significant, implying that the convective activity was confined to the limited areas in the northern part of NEI. During the third week of July, heavy rainfall in Kolkata caused high isotopic depletion, the effect of which was observed at Darjeeling. Still, a moderate effect was observed on the precipitation isotopes in Tezpur, implying the absence of widespread convective activity across the region.

Northeast India is characterised by a humid climate and harbours rich forests. In contrast to NEI, the Pune city (18.5°N, 73.8°E) is a semi-arid site in western India. In this section, we have compared the isotopic records of these two regions. We have evaluated the variation of the δ18O record23,36 (Supplementary Fig. 6) of Pune from 2014 to 2017. We examined its pattern of intra-seasonal variation. Until the middle of the monsoon, the moisture comes primarily from the Arabian Sea. From September, the marine component decreases, but the ET-derived moisture increases. The climatological variation of evaporation and transpiration for the Pune region (Supplementary Fig. 6, lower panel) shows low ET-generated moisture until the middle of May. After that, its contribution steadily increases, driven by the premonsoon showers, resulting in the fluctuation in the isotopic value (grey shaded ellipse). A steady flow of moisture from the Arabian Sea causes rainfall in early June, reducing the isotopic values. But from the latter part of August, the fluctuation in δ18O increases due to the increased proportion of ET-derived vapour (Supplementary Fig. 6, lower panel). We estimated the seasonal mean RR for this region to be about 30% during the postmonsoon season (Fig. 6), which caused large variability in its isotopic values. Unlike NEI, the δ18O profile did not show any declining trend in the case of the Pune site. But raindrop evaporation caused an increase in the isotopic values and lowered the slope of the LMWL compared to the GMWL. For example, the slope of Pune LMWL (7.18; see inset in the upper panel of Supplementary Fig. 6) is considerably lower than that of the NEI-LMWL (8.02; Fig. 2).

Fig. 6: Recycling ratio.
figure 6

The seasonal mean recycling ratio (recycled to total rainfall in %) estimated using ERA5 data (2014–2018) by employing the recycling model of Eltahir and Bras43.

Figure 6 shows the spatial variation of RR across India, generated using the ERA5 reanalysis data. This map shows that the RR during the MAM season is minimal in western peninsular India, including the Pune site (1–4%), but reasonably high (24–32%) for the NEI. This means that the evaporated and transpired moisture sources are small in western peninsular India. Hence, this region receives scanty rainfall during the premonsoon time, which is in sharp contrast to NEI (Fig. 1c).

In NEI, the isotopic variability of rainfall shows a decreasing trend superimposed with the periodic low values throughout the monsoon season. During the premonsoon season, the precipitation isotopic values remain reasonably high, driven mainly by the land-surface processes mediated by enhanced plant productivity. High radiation level, adequate rainfall, optimum soil water content trigger the photosynthetic activity, enhancing the transpiration process that produces moisture rich in heavier isotopes.

During the second phase associated with the monsoon onset, the BoB generated water vapours depleted in heavier isotopes sweep across this region. The extent of ET-derived water vapour is reduced. Local to regional-scale convective activities produce anomalous depletions in isotopic values. The convective events produce a significant water vapour depleted in the heavier isotopes (2H and 18O). Further, the secondary effects, such as the recycling of water vapour within the convective cells, reduce the rainwater’s heavy isotopic composition during the convective events.

Our study shows that the isotopic values of precipitation in this region get modulated mostly by the land-surface processes; external factors play a secondary role in this regard. The study reveals that the hydrological processes which control the precipitation isotopic composition on an intra-seasonal timescale are relatively insulated from the external influences over the NEI region. During the monsoon season, the rainfall in NEI has been rapidly decreasing for the last few decades. The extent of recycled rain is also experiencing a reducing trend in this season17. Though a similar analysis for the premonsoon season is not yet available, the long-term rainfall pattern during the MAM season shows a declining trend. This is especially significant for March because this is the time when an enhanced hydrological cycle triggers the primary productivity. Analysis of the rainfall data (1871–2016) for Assam and Meghalaya reveals that rainfall during March decreased approximately at a rate of 1.66 mm y−1 (see Supplementary Note 4). The RR for the MAM season in NEI is about 25% (see Fig. 6), the highest in India. A reducing trend in rainfall may have severe consequences for ecosystem productivity. This study shows a strong coupling between the hydrological cycle and the plant physiological processes during the premonsoon season. So, a declining trend in the RR may indicate the reduction of primary productivity, which may alter the carbon balance of the fragile ecosystems of northeast India.


Sampling and isotope analysis

We collected rainwater daily at four sites; Tezpur (26.63°N, 92.80°E; 48 m above mean sea level (AMSL)) in the state of Assam, Darjeeling (27.04° N, 88.26° E; 2042 m AMSL) in West Bengal, Kolkata (22.57°N, 88.36°E; 9.14 m AMSL) in the southern part of West Bengal and Kakdweep (21.87°N, 88.18°E; 4 m AMSL) in the coastal region of the Sundarbans in West Bengal. Figure 1 shows a map depicting NEI and Bangladesh with the abovementioned sites marked as red dots. Blue dots in this figure represent the study sites of other investigators. Rainwater samples were collected using ordinary rain gauges, providing the rainfall amount. This data, however, was not available for Kakdweep, Darjeeling 2017, and part of Darjeeling 2013 and 2016. We also collected rainwater for isotope analysis near the tower site (~1 km apart) in 2016. However, due to logistic difficulties, we could not continue the sampling in the subsequent years.

The sample collection time was typically 8:30 LT (local time). After collection, we transferred samples to leak-proof polypropylene vials of 15 ml size and then shipped them to the Indian Institute of Tropical Meteorology, Pune, for isotopic analysis. We used a liquid water isotope analyser (Model: TIWA-45-EP of the Los Gatos Research, USA) to measure the isotopic ratios with a precision of 0.12‰ and 0.8‰ for δ18O and δ2Η, respectively. While the uncertainty for d-excess, defined as δ2Η − 8 × δ18O was 1.0‰.

Study area

The climate of NEI is quite distinct compared to the rest of India. This region receives very heavy rainfall, with several places being among the wettest in the world19. For example, the mean annual rainfall in NEI is 151.3 cm; the premonsoon time (March–May) witnesses a significant amount of rainfall driven by the thunderstorm activities19, which is scanty in most other parts of India. A large portion of NEI is forest-covered (64%), much higher than the national average of India (Indian Forest Service;; see Fig. 1b; adapted from ref. 37). According to the Köppen classification, the climate in this region is humid sub-tropical (CWa). The climatological (1871–2016) distribution of rainfall (the Indian Meteorological Department or IMD’s rain gauge data available at were used) on a monthly scale for the two states of NEI (Assam and Meghalaya; a meteorological sub-division defined by the IMD) has been shown in Fig. 1 (lower panel). The same for the Madhya Maharashtra meteorological sub-division covering the Pune region is also shown for comparison purposes. The temperature and relative humidity of these two regions are shown in Fig. 1d. These figures show that the NEI region receives much higher rainfall, maintains a lower temperature, and has higher humidity than the Pune region.

EC data

Under the MetFlux India Project38, we installed a flux tower at the Kaziranga forest (26.58°N, 93.10°E) to quantify the biosphere-atmosphere carbon exchange process, located approximately 50 km east of Tezpur. A micro-meteorological tower consisting of an EC system was set up in this location that measured CO2 and water vapour concentrations using an enclosed path infrared gas analyser or IRGA sensor (Model: 7200, Li-COR Biosciences, Lincoln, USA).

This observational programme’s instrumentation and analysis details are presented in16,17. This EC system’s LE data (2016–2018) have been used16 to interpret the precipitation isotope data.

Reanalysis data and GLEAM model output

To examine the behaviour of the LE over a broader scale, we have prepared the monthly climatology of the LE for 2014–2018 using the ERA5 reanalysis data for an area of 23°N–28°N, and 89°E–95°E. Then the CO2 flux has been superposed on the LE values (Supplementary Fig. 2) to visualise their combined behaviour. The basic technique of calculating the CO2 fluxes is presented in Sarma et al.16.

We have used the ERA5 reanalysis data39 to study the ET characteristics at the regional scale. Apart from the EC-derived LE data, we used the reanalysis data to determine the same on a regional scale. We used surface-level air temperature and relative humidity data to calculate the respective climatology for the study region. Evaporation and transpiration fractions of the moisture fluxes and the soil water content were obtained from satellite-data-driven reanalysis data (GLEAM25). We estimated the soil water content at two levels (surface [0–10 cm] level and root zone level; 10–800 cm) using the GLEAM data (Supplementary Fig. 3b).

Meteorological data

The amount of rainfall at Tezpur and Darjeeling sites was obtained from ordinary rain gauges available since 2016. To get the regional scale rainfall information, we have used the TRMM (Tropical Rainfall Measuring Mission) satellite-derived rainfall data40. To calculate the rainfall anomaly during the cyclonic events, the rainfall climatology was first prepared from 21-year rainfall data (1998–2018). Then the rainfall climatology was subtracted from the daily rainfall data averaged on a 3-day scale over the study region, including the coastal areas of India and Bangladesh.

To determine the rainfall trend for Assam and Meghalaya of NEI, observed rainfall data for March (1871–2016; available at were plotted, and an ordinary regression line was fitted (see Supplementary Note 4). The slope yielded the rate of change of March rainfall for this region.

OLR analysis

NOAA interpolated OLR data were used to characterise the convective activities. First, 10-year OLR climatology (2009–2018) was prepared over the NEI region. Then OLR anomaly was calculated for each year by subtracting the climatology from the respective year which enabled us to compare the extent of convective activities for a different year. We plotted a time-latitude Hovmoller diagram for daily OLR anomaly. Afterwards, the isotopic time profiles were superposed on the Hovmoller diagram to illustrate the synchronisation between the low OLR (i.e., high convection) and the isotopic depletions.

To study the regional-scale moisture transport characteristics, the vertically integrated moisture transport (VIMT) fluxes41 were calculated for the study region (5°lat × 6°lon) using the following formula42.

$${\rm{VIMT}} = \frac{1}{g}\mathop {\sum}\limits_{j = 1}^J {q_j} \left( {u_j,v_{j,}} \right)\Delta p_j$$

where g is the acceleration due to gravity, qj is the specific humidity (kg kg−1), uj is the zonal wind speed, vj is the meridional wind speed at jth pressure level, and Δpj is the thickness of the jth pressure level. The total number of pressure levels in the consideration for accumulation is J.

The ratio of recycled to the total rainfall (RR)

A fraction of moisture from local ET contributing to local rainfall is called recycling. The ratio between recycled and total rainfall is called the recycling ratio (i.e., recycled rain/total rain). The recycling ratio is estimated by employing the diagnostic model of Eltahir and Bras43. The model relies on the basic atmospheric water vapour budget equation and atmospheric parameters (zonal and meridional winds, specific humidity at surface to 200 hPa levels, and ET) as input from ERA5 reanalysis. The observed rainfall is used to calculate the amount of recycled rainfall based on the estimated recycling ratio. The diagnostic model uses two basic assumptions: (i) the rate of change in the storage term is negligible and (ii) the atmospheric water vapour is well mixed. The recycling ratio is calculated at each land grid point of a selected domain considering species of water vapour molecules, which evaporate within the region (i.e., local) and outside the region (i.e., advection part). The second assumption (i.e., well-mixed) implies that water vapour molecules from advected and local origins have equal probabilities of falling back as precipitation. In contrast, the first assumption holds reasonably well for long-time average cases (i.e., monthly/seasonal). A detailed description of the methodology and simulations is given in Eltahir and Bras43 and Sujith et al.44.

Seasonal mean (December–February, March–May, June–August, and September–November) values were calculated for 2014 through 2018, and their climatological pattern was also estimated. It may be noted that the estimation of recycled rainfall in a short time scale introduces significant error, which arises from the basic assumptions of the Eltahir and Bras43 model. It was found that over the Indian subcontinent, the uncertainty level is negligible if the calculation is done on a seasonal time scale19,44.

Canopy–atmosphere decoupling coefficient

The canopy–atmosphere decoupling coefficient (Ω) is defined as follows15:

$$\Omega = \frac{{\lambda + 1}}{{\lambda + 1 + \frac{{G_{\rm{a}}}}{{G_{\rm{s}}}}}}$$

with λ = δ/ψ, where δ is the slope of the saturation vapour pressure–air temperature curve (in Pa K−1), ψ is the Psychrometric constant (in Pa K−1), Ga is the aerodynamic conductance of heat (in m s−1), and Gs is the surface conductance to water vapour transport (in m s−1).

There are several methods of calculating Ga and Gs. We have, however, computed them from friction velocity (u* in m s−1), canopy roughness parameter, displacement height, and Businger stability function. Computations of these parameters are quite involved and require various other parameters. These steps are listed in Deb Burman et al.26, and we do not repeat those explicitly here as that remains well out of the scope of the present study. Ω is usually considered as a relative measure of environmental control to canopy control on the biosphere–atmosphere water vapour exchange in any ecosystem. Its value ranges within [0,1], with larger values signifying the environmental control dominating the canopy control and vice versa.