Rainwater and vapour samples were collected from the ground level at the Indian Institute of Tropical Meteorology (18.53° N, 73.85° E), Pune during the summer monsoon of 2019. This region receives > 90 % rainfall during the ISM and is situated at the lee side of the WGs (Fig. 1). Rainfall in Western India occurs from mid-tropospheric low-pressure systems in several episodes, each of which usually lasts for 2–3 d. These systems are locked in place during these periods and fed by moisture derived from the Arabian Sea (Ding and Sikka, 2006; Rao, 1976). The geographic location of the region, its altitude, rainfall variation across the WG mountains, and the topographic profile across Pune are shown in Fig. 1. There is a sharp variation of rainfall across the mountains from the coastal zone (30 mm d⁻¹) to the lee side (12 mm d⁻¹) which is a characteristic of orography-induced rainfall (Fig. 1). The surface air temperature in Pune varies from 20 to 30 °C during the ISM (Pattanaik et al., 2019).

The onset and withdrawal dates of ISM (based on wind direction, specific humidity, and Outgoing Longwave Radiation, OLR; IMD, 2019) at Pune in 2019 were 22 June and 4 October 2019, respectively. Rainwater samples were collected during this period following the guidelines of the International Atomic Energy Agency (see Sect. S1 and Fig. S1 in the Supplement). For vapour samples, an in-house fabricated glass condenser was used (see Sect. S2 and Fig. S2). Twenty-nine vapour samples were collected during the rainy days, along with the rain samples; additional 14 vapour samples were collected during non-rainy days. The vapour collection efficiency was estimated from the amount collected against the amount expected (see Table S1 in the Supplement). Due to logistical problems, vapour samples could not be collected before mid-July.

The samples (rain water and condensed vapour) were measured using a Liquid Water Isotope Analyser (Model Number TIWA-45-EP, Los Gatos Research). This instrument measures liquid samples using Off-Axis integrated cavity output spectroscopy (OA-ICOS) with a routine precision of 0.1 ‰ and 1 ‰ for δ18O and δD, respectively, relative to VSMOW (Rajaveni et al., 2024; see also Sect. S3). The d-excess values defined as: d-excess = δD - 8⋅δ18O (Dansgaard, 2012) have a precision of 1 ‰. The rain isotope value of a particular day is reported after being weighted by the rainfall amount on that day.

Daily average temperature, relative humidity and cumulative rainfall data were obtained from the Pune observatories of the IMD, available at the National Data Centre (https://dsp.imdpune.gov.in/, last access: 20 May 2025). The daily gridded data of zonal and meridional wind, specific humidity, air temperature, and cloud liquid water content were obtained from the European Centre for Medium-Range Weather Forecasts Reanalysis (ERA5) dataset with a resolution of 0.25° × 0.25° (Hersbach et al., 2020) and the Interpolated Outgoing Longwave Radiation (OLR) data (2.5° × 2.5°) were taken from NOAA (https://psl.noaa.gov/data/gridded/data.olrcdr.interp.html, last access: 3 December 2025).

The upper-air radiosonde data (relative humidity, RH; temperature, T) over Pune were obtained from the University of Wyoming repository (http://weather.uwyo.edu/upperair/sounding.html, last access: 29 January 2026). The values at every 50 mb interval (∼ 470 m in height) were available for two time periods: at 00:00 and 12 UTC, yielding two profiles for each day. The two profiles for each parameter (RH and T) were averaged to make representative daily profiles. Since the input for BCIM is required at every 1 m interval, a linear interpolation between two consecutive pressure levels in logarithmic scale (Ingleby et al., 2016) was carried out. However, the zone between the cloud base (Lifting Condensation Level, LCL) and the drop introduction height (taken as the height of Cloud Liquid Water Content, CLWC peak) poses a problem. As the BCIM requires 100 % RH for the formation of water droplets, the RH values above the LCL and up to the CLWC peak were considered as 100 %, disregarding the radiosonde data above the LCL (see Sect. 2.4.2). The typical uncertainties of T and RH are 0.3 °C (Sapucci et al., 2005; Jensen et al., 2016) and 8 % (Xu et al., 2023) respectively.

Tropospheric Emission Spectrometer (TES) Level 2 (Nadir-Lite-Version 6) retrievals of HDO and H₂O data for the available period (2005–2007) are used to construct mean vapour δD (δD_v) profiles (discussed later). The details of quality control criteria and biases associated with TES observations are discussed by Herman et al. (2014) and Worden et al. (2011). Grid point observations of δD_v have a precision of ∼ 10 ‰–15 ‰, which reduces to 1 ‰–2 ‰ when the data are averaged over a larger region (Lee et al., 2011; Pradhan et al., 2019).

To identify the moisture sources for vapour/rain at and around our study area, 48 h air mass back trajectory analysis was carried out at 850 mb pressure level using the NOAA Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model (Draxler and Hess, 1997). The model tracks the movement of air parcels backward from a given location for a desired period (see Sect. S4 and Fig. S3).

As mentioned earlier, to quantify the sub-cloud processes altering the rain isotope values, we used the BCIM (Graf et al., 2019). A brief description of this model, for application to the shallow cloud processes over Pune, is provided here. The model comprises a single vertical column that extends from the ground level to the point at which a single hydrometeor is introduced. Within this column, the hydrometeor descends under the influence of gravity, undergoes growth or evaporation (depending upon the ambient humidity and temperature), changes its isotopic composition through equilibrium and kinetic isotope exchange with surrounding vapour, and finally reaches the surface as a raindrop. The final isotopic composition of the drop is estimated by the following procedure: (1) setting up the initial condition involving the drop introduction height and its size, (2) estimation of the initial isotopic composition of the hydrometeor, (3) tracking the microphysical evolution of a falling hydrometeor, and (4) tracking the changes in isotopic composition of the hydrometeor along the descent. For these calculations, the model requires altitude profiles of T, RH and vapour isotopes for a given day as input parameters. The drop is assumed to form in equilibrium (at RH = 100 %) at a level which changes from day to day. The input parameters for the vapour can be introduced into the BCIM in two different ways: (1) the profiles can be calculated based on assumption of idealized (moist) adiabatic ascent of an air parcel from the surface to the top of the column; RH, T and vapour isotope values at various pressure levels are then estimated from the Rayleigh distillation equations starting from the measured surface values or (2) the height specific values of RH and T from radiosondes and vapour isotope values from satellite data and/or any model.

Since our aim is to understand the isotopic change and mass loss suffered by the drops due to rain vapour exchange, we introduce two parameters indicating the deviation of the final rain composition at the ground from the ambient surface vapour (Graf et al., 2019). This is most clearly expressed by the difference between the isotopic composition of vapour in equilibrium with the rain samples (rain eq. vapour) and the ambient surface vapour defined as: Δδ = δD (rain eq. vapour) - δD (surface vapour) and similarly for d-excess, Δd = d-excess (rain eq. vapour) - d-excess (surface vapour).

Measured rain and vapour isotope ratios (δ18O and d-excess) on a daily scale are plotted in Fig. 2a and b. The general pattern of variations in vapour δ18O (δ18O_v) and rain δ18O (δ18O_r) values is similar; both decrease significantly and consistently after mid-August. The vapour δ values are lower than the rain. In contrast, the d-excess values of vapour (d_v) are always much higher. The δ18O_r and d-excess (d_r) values of rainwater range from -10.8 ‰ to 1.5 ‰ and -2 ‰ to 12 ‰, respectively, while those of the vapour range from -19 ‰ to -9 ‰ and 10 ‰ to 30 ‰, respectively. The mean and 1σ standard deviation of δ18O_r and d_r values are -1.3 ± 2.6 ‰ and 4 ± 3 ‰, while those of the vapour are -12.5 ± 2.5 ‰ and 18 ± 5 ‰, respectively. The δ18O (Fig. 2a) and d-excess (Fig. 2b) time series show four interesting features: (1) for the four date ranges: 27–29 June, 24–27 July, 4–8 September and 19–27 September, significant and consistent decrease in isotope values are observed in both rain and vapour phases (marked 1, 2, 3 and 4 in Fig. 2a; no vapour data available for date range 1), (2) On 19 September, the vapour shows a sudden decrease (marked A in Fig. 2a), (3) there is a gradual decrease in vapour δ18O_v values and an increase in d-excess values with the progress of the monsoon, which is especially prominent in the later part, and (4) rain d-excess (d_r) values remain constant with time but δ18O of both rain and vapour start decreasing beginning from early September.

Isotopic depletions in rain and vapour samples in the tropics are often associated with deep convection (Lekshmy et al., 2014; Risi et al., 2008; Sengupta et al., 2020). Signature of such a phenomenon is possibly present here in the form of depleted-isotope events when isotope ratios of a group of samples fall below the overall mean (μ) minus half the standard deviation (σ) (Sengupta et al., 2020). To find the relation of these events to large convective episodes, a latitude-time Hovmoller plot of daily OLR anomaly (averaged over the longitude 70–75° E) is examined in Fig. 2c. OLR values are often used as a proxy for convection because OLR is determined by cloud top temperatures which indicate cloud height. Consequently, a negative OLR anomaly corresponds to colder cloud top temperatures and greater cloud thickness which are characteristics of mesoscale convection. A time synchronous association of low OLR and depleted-isotope events thus indicate mesoscale convection affecting isotope values. Figure 2c indicates four such isotope-depleting mesoscale events (marked as 1, 2, 3 and 4 in Fig. 2a). In addition, we also see one depleted-isotope event without such association (marked as A in Fig. 2a). Interestingly, prominent second CLWC peaks occurred on 3 d, 19, 25 and 27 September, at much higher levels (about 550 mbar or about 5.5 km altitude; see Fig. 3) corresponding to the event number 4 mentioned above.

Figure 2d shows that major rainfall occurred during the months of July and August. The relative humidity at the surface during the whole monsoon season varied from 71 % to 97 %, and the surface temperature varied from 25 to 30 °C (see Fig. S7). It is evident from Fig. 2d that deep convection is associated with high rainfall for the three events 1, 2, and 4. A recent study, based on a year-long continuous measurement of atmospheric vapour in Sri Lanka (a nearby tropical country with a similar monsoon system), also found such isotopic depletion during high rainfall events (Wu et al., 2025).

Figure 4a shows the Local Meteoric Water Line (LMWL) using rainwater samples and the Local Water Vapour Line (LWVL) using vapour samples from this study. The LMWL equation is δD_r = (7.3 ± 0.1) δ18O + (3.0 ± 0.3) and the LWVL, δD_v = (6.4 ± 0.2) δ18O - (1.9 ± 3.0), subscripts r and v denote rain and vapour. The slope and intercept of the LMWL values are lower than those of the Global Meteoric Water Line (GMWL), which are 8.0 and 10.0, respectively (Dansgaard, 2012; Gat, 1996). This difference, though small, suggests some amount of below-cloud evaporation of the rain. At Roorkee, a high-latitude Indian station, Saranya et al. (2018) found a LMWL with a lower slope (5.4) but a higher intercept (27). They attributed these changes to the contribution of evaporation from water bodies nearby and moisture recycling during the monsoon. Rahul et al. (2016) got a similar slope (7.4) but a lower intercept (1.5) in Bangalore (southern central India, at a high altitude of ∼ 1000 m). The lower slopes of meteoric water lines provide evidence of evaporation processes associated with kinetic fractionations occurring during rain evaporation.

The d-excess values of rain samples suffering evaporation generally bear a negative relationship with δ18O values (Bonne et al., 2014; Munksgaard et al., 2020). Such correlation is clearly seen in our data (Fig. 4b). In addition, the vapour d-excess values also show a statistically significant negative correlation with δ18O values (Fig. 4b; R2 = 0.61; p = 0.001), probably indicating contribution of vapour derived from rain evaporation (Kurita, 2013; Risi et al., 2021). Correlation studies can be indicative, but the causative factors behind the above variations can be explored only with the help of a process-based model like BCIM.

As discussed in Sect. 2.4, simulations of BCIM were carried out under three sets of initial conditions of RH, T and vapour isotope profiles designated as Run-1, Run-2 and Run-3. The aim was to achieve improvement in reproducing the measured rain isotope values by altering the input parameters step by step. The sources of input profiles of ambient temperature (T), relative humidity (RH), vapour δD (δD_v), and vapour d-excess (d_v) used for the three BCIM runs are given in Table 1.

In our data, the d_v is not significantly correlated with rainfall amount, relative humidity, specific humidity and temperature (details are provided in Sect. S12 and Fig. S13). However, we do see synchronous low OLR values and low isotope values (in both vapour and rain) as the convective cloud bands move to the sampling location in Pune from the southwest (Fig. 2).

Rain isotopes often vary with rainfall, humidity and temperature (Dansgaard, 2012; Lee and Fung, 2008). But the rain isotopes in Pune do not have a simple relation with the local rainfall (Fig. 2). The absence of rain amount-isotope correlation in the tropics has also been found in several other regions (Chakraborty et al., 2016; Moerman et al., 2013; Vimeux et al., 2011). Despite the absence of isotopes with local rainfall, a correlation is often found with the regional convective activities (Kurita, 2013; Lekshmy et al., 2018). Risi et al. (2023) have noted that in the tropics, most of the precipitation falls under deep convective systems (see Sect. 3.1 and Fig. 2) which are controlled by various microphysical processes (like rain evaporation, diffusive liquid-vapour exchange) connected through mesoscale circulations. These processes probably add on to the effect of surface meteorological parameters in this region to offset a simple dependence of rain isotopes on rainfall. However, as noted above, movement of large-scale convective bands reflected by low OLR registers its signature in both low isotope events and high rainfall (Fig. 2).

The observed increasing trend (13 ‰ to 30 ‰) in the vapour d-excess values associated with a decrease in the δ18O values with the progress of the monsoon (Fig. 2b) is an intriguing feature and could be ascribed to a significant contribution from some evaporative sources. Changes in moisture sources can also cause concomitant changes in isotope values in rain and vapour (Deshpande et al., 2010; Midhun et al., 2018). We investigated this possibility by 48 h of air-parcel back trajectory analysis (Fig. S4-1) which shows that moisture was derived mainly from the Arabian Sea throughout the season.

The process of evaporative exchange during the fall of raindrops causes isotopic enrichment in the rain, which cannot be easily quantified. Evaporation is reflected in higher δ values and lower d-excess values of the rain samples. Froehlich et al. (2008) used d-excess values of precipitation in the Alpine region to derive the magnitude of evaporation using assumed end-member values of the regional vapours. Any isotope exchange between the rain and ambient vapour would result in correlated changes. A strong correlation between rain and vapour δ18O values is indeed found (Fig. 6a; R2 = 0.7, p < 0.01, n = 29). Sinha and Chakraborty (2020) also found significant positive relations (R2 > 0.8) between rain and vapour δ18O values over the Andaman Islands. However, they did not find any anticorrelation between δ18O_r and d_r, as found here (Fig. 4b). The current study exhibits an anticorrelation between the differences in d-excess (Δd-excess_r − v) and δ18O (Δδ18O_r−v) of rain and vapour (the subscript r - v indicates rain isotope minus vapour isotope; Fig. 6b).

As raindrops evaporate, part of the newly formed vapour may get down-drafted to lower levels; the rain and vapour phases at the ground level would then exhibit opposite changes because the generated vapour is lower in δ18O but higher in d-excess compared to the rain. This happens when the evaporative contribution is large. However, in the case of tropical precipitation, the fractional addition from rain evaporation is small because the ambient vapour is a large reservoir. It has been shown in several earlier studies that the total rain constitutes only a few percent of the overhead vapour mass (Pathak et al., 2014; Rahul et al., 2016). Earlier studies have also shown that vapour d-excess values do not exhibit any systematic change in central or southern WG stations, but their rain δ18O values exhibit gradual depletion (1 ‰ to -10 ‰) in the latter part of the monsoon (Lekshmy et al., 2018; Rahul et al., 2016). The negative correlation found in this study, albeit minor, suggests that the ground-level vapour gets a significant contribution from drop evaporation. How can moisture generated by drop evaporation during raindrop descent contribute to the ground-level vapour? This is possible when there is a strong downdraft associated with intense monsoon rains (Risi et al., 2023). In a modelling study, Mandke et al. (1999) pointed out that deep convective cloud systems contain both upward and downward components. The downward motion is driven by evaporation of the falling drops and dragging of the ambient air and vapour by big droplets. This downdraft brings moisture down from above and increases the vapour d-excess at the ground (Risi et al., 2010; Kurita, 2013; Aemisegger et al., 2015).

The existence of drop evaporation is further supported by a relation between Δd-excess_r − v and surface RH (R2 = 0.31; Fig. 6c). The difference between rain and vapour isotopes is higher (more negative) in lower RH and less in higher RH (Stewart, 1975). A similar analysis (Xing et al., 2023) in China also found that the change in isotopic composition is large when RH is less than 60 %.

Falling raindrops and the vapour in the atmospheric column constitute an interacting two-phase system. Below the cloud base, the water molecules are constantly exchanged between these two phases depending on the ambient RH and T. The difference between isotopes (δD_v and d-excess_v) of vapour in equilibrium with raindrops and the observed surface vapour (Δδ and Δd, respectively) is a useful signature of departure from equilibrium exchange. Graf et al. (2019) demonstrated how the Δδ - Δd plot can be used to decipher the effect of sub-cloud processes, such as evaporation and equilibration, which influence the rain isotopes. The time series of Δδ values (Fig. 7a) for the Pune rain samples shows that the values varied between -15 ‰ and 21 ‰. For Δd, the time series shows negative values in all cases (ranging from -2 ‰ to -24 ‰). The close-to-equilibrium samples correspond mostly to the high-humidity periods in July (Fig. 7b). Fifteen samples show the influence of below-cloud evaporation with positive Δδ values and associated strongly negative Δd values (up to -20 ‰).

A Δδ - Δd scatter plot (Fig. 8) shows that none of the rain samples is in equilibrium with the corresponding ground-level vapour. If the equilibrium pertained, the corresponding point would plot at the origin. Fifteen points fall in the lower right quadrant of the diagram (positive Δδ and negative Δd values), where the raindrop evaporation is relatively more significant. The rainfall amount for these samples was low (less than 5 mm d⁻¹), consistent with a substantial evaporation effect. Fourteen samples have both negative Δδ and Δd values (located in the lower left quadrant), indicating incomplete equilibration with near-surface vapour. The crucial driving factors for below-cloud processes seem to be the size of raindrops related to the intensity of precipitation. This is because raindrops with larger diameters correspond to increased intensity and have shorter residence times in the atmospheric column. As a result, they experience reduced evaporation during descent. It is to be noted that the drop diameter in this study was not measured by a disdrometer directly. They were estimated from the rain rate using the Marshall–Palmer relationship. Therefore, larger rain rates will always translate to larger droplet sizes. This physical relationship is thought to be a result of increased collision-coalescence during higher rainfall intensity (Law et al., 2021).

Distribution of points in the Δδ - Δd plot (Fig. 8) shows that 10 samples with < 5 mm d⁻¹ rain rate fall in the lower right quadrant compared to 5 samples in the left quadrant. This suggests that drop evaporation is a dominant process in low rainfall events (where smaller drop diameters dominate). In Fig. 8, the size and colour of the points denote drop size and rainfall, respectively. It is clear that larger drop size points with higher rainfall plot in the lower left quadrant. This suggests that for larger sizes, the memory of the isotopes is partly retained despite the sub-cloud evaporation. Twenty-nine samples are nearly equally distributed in the two quadrants, suggesting an equal number of equilibration-dominant and evaporation-dominant rain events. It is to be noted that deep convective rains during the monsoon exhibit significantly higher mass-weighted diameter compared to shallow convective rains or stratiform rains (Kumar et al., 2025). The five big diameter points in the lower left quadrant correspond to cases of deep convection.

The regression line in the Δδ - Δd cross plot (Fig. 8) has a slope of -0.45 based on Bootstrap analysis (see Sect. S13). In contrast, Graf et al. (2019) obtained a smaller value of -0.30 for their study area, Zurich, Switzerland. The difference is intriguing and merits a discussion. Their study was based on short-time intra-event samples (covering about 16 h and each rain sample being collected for 10 to 15 min) in a mid-latitude region, whereas Pune samples were 29 daily rain collections in a tropical region (covering a few months and each collected for 24 h). A set of complex processes operates to dictate the value of the slope, and Graf et al. (2019) pointed out that the slope could represent a balance between below-cloud evaporation and equilibration. They suggested that it would be insightful to explore the slope for other climatic regions, hinting that the slope will help assess the evaporation. A quantitative estimate of the evaporation fraction can be obtained from BCIM by using the mass loss parameter in the output (see Sect. 4.3).

At higher humidity, the evaporation is lower, the change of Δd is smaller, and this leads to a lower slope. Conversely, when the temperature is higher, the slope is higher due to higher evaporation. The value of the slope is determined by a differential effect in evaporative fractionation. Evaporation decreases d_r, and increases δD_r, but the magnitudes of these changes (negative for d_r and positive for δD_r) are not the same. Fractionation values (involving equilibrium and kinetic effects) show that the change in δD_r is larger compared to that in d-excess (about 30 % of the δD_r change, considering absolute values). This is because in evaporation, the kinetic fractionation occurs along with equilibrium fractionation, and the former has more influence on δD compared to δ18O. If evaporation is higher (due to higher T and lower RH), the deviation of the predicted rain isotope values from those predicted by the equilibrium fractionation alone will be more, and the slope will be higher. In the frontal systems of Switzerland, the T was about 12 °C and RH about 80 % (Graf et al., 2019), whereas in Pune, the T was about 25 °C and RH about 85 %. Even though the RH was nearly similar, the T was much higher in Pune, which resulted in a higher slope for Pune (see next section).

As per the simulation of BCIM, the mass of the drop reduces as it falls. The ratio of final mass to the initial mass (m/mo) can then be used to estimate the fractional mass loss suffered by the drop on its way down. The difference (1 - m/mo) represents the effective rain evaporation. With this definition, a time series of evaporation values (Fig. 9a) shows variation from 4 % to 61 %. The average evaporation is 23 % if we consider all 29 values. But there are four large values 45 %, 47 %, 58 % and 61 %, all of which correspond to low rainfall or small drop diameter (Fig. 9d). If we exclude them, the average is 18 % (n = 25). As expected, drop evaporation is inversely related to relative humidity (Fig. 9b) and drop diameter (Fig. 9d) but directly proportional to the temperature (Fig. 9c). The relative importance of the three determining factors, namely, RH, T and drop diameter, is seen through the multiple regression equation of evaporation fraction. Here, we should use normalized multiple regression because simple (or unnormalized) multiple regression uses variables in their original units, while normalized multiple regression transforms all features onto a similar scale, allowing for direct comparison of feature importance. The normalized evaporation fraction as a function of normalized values of RH, T and D shows that (1) D is the major determinant and (2) T plays an important role, nearly as much as RH, for evaporation. 2Normalized Evaporation Fraction=-0.329*RH+0.370*T-0.665*Diameter

For the larger values of temperatures, we expect more evaporation; this leads to the higher slope value of -0.45 (in the Δδ - Δd cross plot in Fig. 8) for Pune compared to -0.30 for Zurich, as noted in the previous section.

The evaporation was particularly high (61 % and 59 %) on 22 July and 21 August due to combined effect of low RH (78 % and 74 %), and high T (30 and 27 °C) and small D (Fig. 9a). On average, the evaporation fractions are moderately high (23 ± 16) % which is consistent with the observed anticorrelation between d_r and δ18O_r of rain samples (Fig. 4b).

Among the controlling factors for evaporation, the temperature does not vary much (26.8 ± 1.0 °C) or about 4 %, while for RH, the variation is slightly larger (85 ± 6 %) or 7 %. The diameter variation, on the contrary, is much higher, about 30 % (1.0 ± 0.3 mm) and has a higher impact on the evaporation. The net uncertainty in the evaporation fraction due to combined uncertainties in RH, T and D can be determined by a simple multiple regression equation (using unnormalized variables) as given below. The Evaporation Fraction (EF) was regressed with RH (in %), T (in °C) and D (in mm) and yields: 3Evap. Fraction=-36.68-0.71*RH+5.72*T-34.08*D

Equation (3) can be used to estimate the error in drop evaporation, knowing the uncertainties in RH, T and D and using the partial regression coefficients as partial derivatives. Using the standard quadratic formula for error (Farrance and Frenkel, 2012), we write: 4σ(EF)2=∂EF/∂RH2*(RH)2+∂EF/∂T2*(T)2+∂EF/∂D2*(D)2 where σ denotes the uncertainties in EF, RH, T and D, and the quantities in brackets express the partial derivative of EF with respect to the variable (see Eq. 3). The uncertainties associated with RH (see Sect. S14; Fig. S15), T and D are discussed in Sect. S10. Adding these three errors by the above quadratic formula, we obtain the error in the evaporation fraction for 29 d, which varies from 7.4 % to 13.8 % (for EF values from about 4 % to 61 %). The average for 29 d is ±8.9 %, which is taken as the overall error in the evaporation estimate.

Evaporation fraction plotted as a function of rainfall (Fig. 10) shows a power law. For a smaller rainfall range (less than 5 mm d⁻¹), evaporation affects the rainfall significantly. The reason is that smaller rainfall is usually associated with smaller drops, which are very sensitive to evaporation, resulting in a power law. If the EF increases by 10 % (say from 15 % to 25 %), the rainfall reduces by about 4 mm d⁻¹ (from 5 to 1 mm d⁻¹).

It is instructive to compare our results to the evaporation estimates obtained in similar studies carried out in other climatic regimes. Using a steady state one-dimensional model of rain in the North Atlantic Trade Wind region (Barbados), Sarkar et al. (2023) found a high value of 63 % (63 ± 23 %) for raindrop evaporation (using radar reflectivity data of rain evaporation flux), which is three times more than our average value of 23 % (23 ± 16 %). The reason for the large difference in Pune evaporation from Barbados is possibly due to a large difference in drop diameter and RH. A comparison reveals that in Barbados, the drop diameter was much smaller (from 0.1 to 0.6 mm) in comparison to ours (from 0.61 to 1.80 mm). The Barbados drops were so small that in some cases (for example, smaller than 0.3 mm on 4 February 2020) they evaporated completely (evaporation ∼ 100 %) during descent. In addition, in their sampling region, the RH was lower, ranging from 65 % to 80 %, compared to ours (74 % to 97 %). A smaller drop diameter and lower RH lead to higher raindrop evaporation. In contrast, only four high evaporation days (more than 45 %) occurred in Pune out of 29 sampling days. We also note that their drop diameters varied over a wider range, leading to a larger variability compared to our study.

In a similar study as here, rain and vapour isotopes were measured in a cold-front passage over Zurich during 19–25 July 2011, and the data were interpreted by an isotope-enabled regional weather prediction model COSMOiso (Aemisegger et al., 2015). The authors showed that by switching off the raindrop evaporation, the rainfall increased by about 75 % because the cooling induced by evaporation causes diminished convective activity. The estimated average evaporation in their study was about 40 %. This value is nearly twice our average value. The reason is probably a smaller drop diameter and lower RH and the authors write: “weak rainfall intensities (small droplets and thus lower falling velocities), and possibly lower relative humidity in the air column could have contributed to the evaporative enrichment of precipitation”.

The present study using the BCIM applied to the Pune region is associated with the following limitations: 1.

We used TES satellite data averaged over 2005–2009 to guide our choice of vapour isotope profiles, but the year of analysis was 2019. In this matter, there is no way to ascertain the degree of deviation of the true profiles from the adopted ones in Run-3.

Evaporation of raindrops during the ISM has been postulated earlier in several theoretical models, but this study provides, for the first time, a quantitative estimate of rain evaporation on a day-to-day basis in the monsoon season using combined rain and vapour isotope data in a BCIM. We found that about 23 % raindrop evaporation occurred in the 2019 monsoon season in the highly humid Pune region. The average seasonal rainfall in Pune is about 55 cm (during monsoon), and if ∼ 23 % of this is evaporated, it would mean considerable cooling of the boundary layer, leading to localized downdrafts, formation of cold pools, and changes in atmospheric stability. Cooling can also hinder efficient formation of convection (Hwong and Muller, 2024) and can have a large effect on the precipitation patterns in the tropics (Bacmeister et al., 2006; Sarkar et al., 2023). Given the large share of precipitation recycling found in this study for Pune, the question arises as to how large the precipitation recycling is at regional or continental scales. We need to have a comprehensive program for carrying out such analysis, aided with appropriate BCIM input parameters, to understand the evaporation of raindrops over various climatic subdivisions in India. Moreover, high-frequency observations of vapour and rain isotopes would be useful to quantify this fraction during various convective events associated with low-pressure systems during the monsoon. A quantitative estimate of raindrop evaporation would be highly useful for modelling the energy and moisture budgets during the monsoon season.