The dropsonde measurements were conducted using two research aircraft: the High Altitude and Long Range Research Aircraft (HALO) and the NOAA WP-3D Orion (P-3) . The primary scientific motivation for these measurements was to characterize the area-averaged kinematic environment of trade-wind cumulus cloud fields and to quantify area-averaged vertical motion in the marine boundary layer. The cornerstone of the measurement strategy was the EUREC⁴A-circle (Fig. ), a standardized circular flight pattern designed to enable accurate estimates of properties including divergence, vorticity, and vertical velocity. This approach, adapted from and , exploits the assumption that atmospheric variations at the mesoscale can be approximated as linear in horizontal space over the scale of a single aircraft circle.

The EUREC⁴A-circle was centered at 13.30° N, 57.72° W with a diameter of approximately 222 km. Each circle flight pattern deployed 12 dropsondes at regular intervals around the circumference, with launches separated by approximately 5 min corresponding to the aircraft's 60 min circuit time. This systematic sampling enables estimation of horizontal gradients through regression analysis, from which mesoscale circulation properties can be derived.

A total of 85 EUREC⁴A-circles were flown during the campaign (71 by HALO and 14 by the P-3), providing unbiased sampling across meteorological conditions. All flights operated from Bridgetown, Barbados, with HALO operating at altitudes between 10.0–10.5 km, while the P-3 typically flew at lower altitudes near 7.5 km as part of the ATOMIC component of the campaign . A total of 1215 Vaisala RD-41 dropsondes were launched (895 from HALO and 320 from the P-3), each containing pressure, temperature, and humidity sensors sampling at 2 Hz, along with GPS receivers providing wind measurements at 4 Hz. The measurements are archived in the JOANNE dataset , from which we use the Level 4 circle-averaged products containing mesoscale diagnostics derived through regression analysis of the 12-sonde circle patterns. These products include horizontal gradients of atmospheric variables and derived quantities such as divergence, vorticity, horizontal divergence of wind velocity, and vertical velocity, with associated uncertainty estimates from the regression fitting.

The subcloud layer height h used here was first presented in and is determined using a temperature-based threshold method . Specifically, h is defined as the height where the virtual potential temperature θv first exceeds its density-weighted mean value (computed from 100 m up to h) by a fixed threshold ϵ=0.2 K . This approach follows established methodology for identifying the top of the well-mixed subcloud layer . The method accounts for the finite thickness of the transition layer separating the mixed layer from the cloud layer above . This transition layer, approximately 150 m thick, complicates the application of classical mixed-layer theory which assumes an infinitesimally thin inversion.

The entrainment rate E is computed using a modified flux-jump model that extends the classical approach of and . Within that diagnostic framework, E quantifies the exchange implied by small-scale mixing at the subcloud layer top and contributes to the tendency for h to deepen. This approach differs from the zero-order jump models that assume instantaneous transitions by accounting for the finite depth of the entrainment zone, providing a more realistic representation of the actual atmospheric structure observed during EUREC⁴A. Note that throughout this study, E refers only to entrainment from the cloudy layer (CL) into the subcloud layer (SCL). It does not include exchange between the CL and the free troposphere (FT). When E is used as an observational predictor or as a forcing coordinate in the synthetic mixed-layer model, it should be interpreted as a diagnosed cloud–subcloud exchange velocity rather than as a standalone prognostic equation for h. Negative diagnosed values indicate periods of interface descent or weak detrainment, not physically negative turbulent entrainment. None of the n=56 complete matched cases used in the regressions below has negative E; the minimum matched value is 6.9mms-1. Across the matched circles, the average entrainment rate is E=18.8±6.9mms-1, and the average subcloud layer height is h=0.69±0.10km.

Water vapor isotope measurements aboard the R/V Meteor were obtained using a Picarro L2130-i cavity ring-down spectrometer (CRDS) operating at 1 Hz resolution from 18 January to 22 February 2020. The analyzer was housed in the Air-Chemistry Laboratory at ∼ 20.3 m above sea level, sampling ambient air through a 5 m long, 4.6 mm ID PTFE inlet line heated to 45 °C and insulated with polyethylene foam. The inlet was housed in a downward-facing funnel to minimize contamination from rainwater and sea spray, and included a 0.2 µm PTFE aerosol filter. Flow was controlled by the CRDS system at approximately 0.03 slpm, resulting in a time delay of >2 min from intake to analyzer. Daily calibration checks were performed using four liquid water standards spanning δ18O values from -20.97 ‰ to -2.79 ‰ and δD values from -158.13 ‰ to -13.12 ‰, delivered in gas phase using a Picarro Standards Delivery Module. Prior to normalization, isotopic observations were corrected for small humidity-dependent biases of up to 0.24 ‰ in δ18O and 0.36 ‰ in δD. Total uncertainties were estimated at 0.29 ‰ for δ18O and 1.24 ‰ for δD by summing in quadrature the uncertainties associated with liquid standards, humidity-dependence correction, calibration measurement precision, and temporal drift over the campaign. Further details on the isotopic data collection program can be found in .

For the circle-based analysis, the shipboard Picarro observations were matched to the JOANNE Level 4 circle products by averaging quality-controlled δD and H₂O measurements within a 2 h window centered on each circle time and within the fitted circle footprint. Of the 85 JOANNE Level 4 circles, 57 had matched ship-based isotope and humidity observations; requiring valid matched entrainment (E) and vertical velocity (W) estimates removes one additional circle, yielding the n=56 complete cases used in the regressions below.

Near-surface relative humidity data were obtained from temperature and dew point measurements made at 29 m above the sea surface on the Meteor's mast. SST was measured at 2.3 m depth with matching port and starboard sensors.

The model advances two prognostic variables: the column water-vapor inventory Q=ρhq and the column heavy-isotopologue vapor inventory MD=ρhqR, where q is the specific humidity and R=[HDO]/[H2O] is the isotopic ratio. These inventories represent the mass stored in the well-mixed subcloud control volume per unit horizontal area. The budget equations are 4dQdt=Fsurf+Fent+Frain,5dMDdt=Fsurf,D+Fent,D+Frain,D, where the three flux terms on the right-hand side represent surface evaporation, entrainment from the cloudy layer, and rain evaporation, respectively. In the synthetic equilibrium experiments presented here, h is held fixed, so E is interpreted as the cross-interface exchange velocity across this prescribed control volume rather than as a prognostic deepening rate for h. The specific humidity and isotopic ratio are recovered diagnostically as q=Q/(ρh) and R=MD/Q. Isotopic ratios are reported in standard delta notation, δD=1000(R/RVSMOW-1), where RVSMOW=155.76×10-6.

By formulating the budgets in terms of Q and MD rather than q and R directly, the model avoids the nonlinear coupling and potential numerical inconsistencies that arise when isotopic ratio equations are solved independently of the mass budget. This conserved-variable approach ensures that the isotopic budget is strictly consistent with the mass budget at every timestep.

The surface moisture flux is parameterized using bulk aerodynamic theory: 6Fsurf=ρCEU(qsat-q), where CE is the bulk transfer coefficient, U is the surface wind speed, and qsat is the saturation specific humidity at sea surface temperature and surface pressure. The corresponding isotopic flux follows the Craig–Gordon formulation : 7Fsurf,D=ρCEUqsatRocαeqαk-qRαk, where Roc is the isotopic ratio of ocean surface water, αeq is the temperature-dependent equilibrium fractionation factor , and αk is the kinetic fractionation factor associated with diffusive transport through the interfacial sublayer .

This formulation deserves brief comment. The classical Craig–Gordon model expresses the isotopic ratio of the evaporating vapor as a function of relative humidity H=q/qsat, with a (1-H) factor in the denominator that produces a singularity as H→1. However, when the evaporative isotopic flux is written as the product of the mass flux Fsurf and the evaporating vapor's isotopic ratio, the (1-H) factors cancel exactly between the mass flux and the Craig–Gordon expression. Equation () implements this cancelled form directly, ensuring that the isotopic flux remains well-defined and proportional to Fsurf across all humidity conditions.

Entrainment from the cloudy layer into the subcloud layer is parameterized as 8Fent=ρE(qCL-q),9Fent,D=ρE(qCLRCL-qR), where E is the entrainment velocity (m s⁻¹), and qCL and RCL are the specific humidity and isotopic ratio of the cloudy layer, respectively. Because qCL<q in the trades, entrainment acts as a moisture sink and, because the cloudy layer is isotopically depleted relative to the subcloud layer (RCL<R), it simultaneously depletes δD. The isotopic flux is written in terms of the conserved quantity qR rather than R alone, maintaining strict consistency with the mass budget.

The rain evaporation term is the additional process tested as a mechanism for the observed asymmetry. As raindrops formed by condensation in the cloud layer fall through the unsaturated subcloud layer, they partially evaporate. This process adds both mass and isotopic mass to the subcloud vapor, but the isotopic ratio of the evaporated vapor differs from that of the ambient vapor because of fractionation during partial evaporation.

The rain evaporation mass flux is 10Frain=f(W)⋅Pcb⋅χ, where Pcb is a reference cloud-base precipitation-rate scale (kg m⁻² s⁻¹), χ is the fraction of the falling rain mass that evaporates before reaching the surface, and f(W) is a rain activity function that modulates the effective precipitation input with mesoscale vertical velocity W. Equivalently, the effective mean cloud-base precipitation input is Pcb,eff(W)=f(W)Pcb. The corresponding isotopic flux is 11Frain,D=Frain⋅Rev, where Rev is the isotopic ratio of the vapor produced by rain evaporation.

The rain activity function f(W) represents the dependence of precipitation activity on mesoscale vertical velocity. Mesoscale ascent is associated with cloud organization and cloudier, moister mesoscale states , so we use W as a coordinate for the precipitation flux that can drive below-cloud evaporation. We parameterize this dependence as a logistic function, 12f(W)=11+exp⁡-W/Wwidth, where Wwidth controls the sharpness of the transition between suppressed and active precipitation regimes. This function varies smoothly from near zero for strong subsidence to near unity for strong ascent, with the midpoint at W=0. The specific functional form is a modeling choice rather than a universal precipitation law, but assigning larger effective precipitation input during mesoscale ascent is physically motivated by the observed cloud–circulation coupling during EUREC⁴A .

The isotopic ratio of the vapor released by partial rain evaporation, Rev, is computed using the below-cloud evaporation model of . This model tracks the evolution of a raindrop's isotopic composition as it falls through unsaturated air, accounting for both equilibrium fractionation at the drop surface and kinetic fractionation during diffusive transport of vapor away from the drop.

The raindrop is assumed to form at cloud base by equilibrium condensation from cloud-layer vapor, giving an initial liquid isotopic ratio Rr,cb=αeq(Tcb)⋅RCL, where αeq(Tcb) is the equilibrium fractionation factor at cloud-base temperature. Here and in the surface-flux formulation above, αeq denotes the liquid-to-vapor equilibrium fractionation factor, so αeq>1. Because liquid is enriched relative to vapor in equilibrium, Rr,cb>RCL: the raindrop starts with a higher isotopic ratio than the cloud-layer vapor from which it formed.

As the drop partially evaporates, the isotopic ratio of the remaining liquid evolves according to : 13Rr,sfc=γR+(Rr,cb-γR)fβ, where f=1-χ is the fraction of the original drop mass that survives to the surface, R is the ambient vapor isotopic ratio, and γ and β are functions of the equilibrium fractionation factor αeq, the molecular diffusivity ratio D/D′, the kinetic exponent n, and the ambient relative humidity H: 14κ=αeqDD′n(1-H),15β=1-κκ,16γ=αeqH1-κ. The isotopic ratio of the total vapor released by evaporation is then obtained by mass balance over the evaporated fraction: 17Rev=Rr,cb-fRr,sfc1-f.

This formulation captures the leading-order isotope mass balance associated with below-cloud evaporation. For the conditions typical of the EUREC⁴A subcloud layer (H≈0.7, Tcb≈18°C, χ≈0.8), the evaporated vapor is isotopically enriched relative to the ambient subcloud vapor. With the baseline parameters, the cloud-base liquid has δDr,cb≈+2.6 ‰ and the mass-weighted evaporated vapor has δDev≈-10 ‰, about 60 ‰ enriched relative to the ambient subcloud vapor. The enrichment arises because the raindrop starts with a high isotopic ratio (set by equilibrium condensation from cloud-layer vapor) and, during partial evaporation, preferentially loses light molecules. The vapor released is therefore enriched relative to both the cloud-layer vapor from which the drop formed and the ambient subcloud vapor into which it evaporates. This enrichment provides a mechanism for the asymmetric response: rain evaporation during mesoscale ascent adds vapor that is isotopically distinct from the ambient subcloud air, strongly modifying δD, while the mass of moisture added by rain evaporation is small relative to the total subcloud moisture budget, producing only a modest change in q.

The model is integrated forward in time with a 20 s timestep from specified initial conditions until a quasi-steady state is reached (typically 60 h of integration, with results averaged over the final 24 h). For a given pair of forcing values (E,W), the model evolves toward an equilibrium in which the surface evaporation, entrainment, and rain evaporation fluxes balance. We compute the equilibrium state across a grid spanning the observed range of entrainment rates (E=10–35 mm s⁻¹) and vertical velocities (W=-10 to +10 mm s⁻¹).

We represent below-cloud rain evaporation as a steady drizzle source whose effective mean input is modulated by f(W). This mean-flux representation is appropriate for diagnosing the equilibrium response: the question is not the detailed timing of rain events, but whether a W-dependent source of fractionated rain evaporation can rotate the δD contours relative to the humidity contours. The model does not include a separate direct vertical-advection term proportional to W; instead, W is used as a forcing coordinate for the rain-activity pathway being tested. We therefore compare the drizzle case with a no-rain control in which Frain=Frain,D=0.

Figure  shows the resulting equilibrium structure. Using the same contour-angle convention as in Fig. , the drizzle case yields fitted δD contours with an angle of 31.0° in the E–W plane, whereas the fitted q contours have an angle of 66.5°, giving a contour separation of 35.4°. Equivalently, the ratio of the fitted W and E sensitivities is much larger for isotopic composition (|∂δD/∂W|/|∂δD/∂E|=1.66) than for humidity (|∂q/∂W|/|∂q/∂E|=0.44). Thus, rain evaporation produces a state in which δD is strongly modulated by mesoscale vertical velocity while q remains primarily controlled by entrainment. In the no-rain control, both variables have nearly vertical fitted contours and zero W sensitivity, so the contour separation collapses. The modeled separation is somewhat smaller than the observational separation of 41.5°, but the two are close enough to support the same qualitative interpretation. This comparison shows that below-cloud rain evaporation is a process that can generate the non-parallel q–δD structure.

The model parameters are summarized in Table . Environmental conditions are set to campaign-mean values: SST =27.3°C, surface pressure p=1013.25 hPa, subcloud layer depth h=700 m, and surface wind speed U=8 m s⁻¹. The bulk transfer coefficient CE=1.2×10-3 is a standard value for open-ocean conditions. Equilibrium fractionation follows . For the surface Craig–Gordon flux, kinetic fractionation uses αk=(D/D′)n with D/D′=1.0159. For the Stewart below-cloud evaporation calculation, we use the deuterium diffusivity ratio D/D′=1.024 and n=0.58 .

The cloudy-layer end member is specified as qCL=13.0 g kg⁻¹ and δDCL=-78 ‰, consistent with the mean dropsonde profiles. The cloud-base temperature is set to Tcb=18°C, and the ocean isotopic ratio to δDoc=0 ‰.

The rain evaporation parameters – reference cloud-base precipitation-rate scale Pcb=5×10-5 kg m⁻² s⁻¹, evaporated fraction χ=0.8, and logistic width Wwidth=4 mm s⁻¹ – are free parameters, not independently fitted observations. We choose them to represent weak trade-cumulus drizzle: the precipitation-rate scale is consistent with observed trade-cumulus rain rates , and the high evaporated fraction reflects the shallow cloud base and warm subcloud conditions, in which much of the falling precipitation can evaporate before reaching the surface. The fixed evaporated fraction is a bulk simplification; the model does not resolve drop-size distributions, ventilation, or vertical variations in relative humidity and temperature along the fall path.

Figure  shows a targeted two-parameter sensitivity sweep spanning (Pcb,χ) at fixed Wwidth=4 mm s⁻¹. The baseline configuration used throughout the paper is marked on each panel. The upper-left panel shows the modeled contour separation between δD and q, the upper-right and lower-left panels show the corresponding W-to-E sensitivity ratios for humidity and isotopic composition, and the lower-right panel shows the absolute mismatch between the modeled and observed contour separations. The baseline point lies in the regime where δD is clearly W-sensitive (|∂δD/∂W|/|∂δD/∂E|>1) while humidity remains E-dominated (|∂q/∂W|/|∂q/∂E|<1), and it yields a modeled contour separation close to the observed value. Because Wwidth controls the shape of the prescribed W–rain transition rather than the rain-flux amplitude, we also ran a baseline-only sensitivity check over Wwidth=2–8 mm s⁻¹ (not shown). For Wwidth=2–6 mm s⁻¹, the model retains the qualitative target of W-sensitive δD and E-dominated humidity, with contour separations of approximately 33–36°; a broader transition of 8 mm s⁻¹ weakens the isotope W sensitivity to near unity and reduces the separation to about 30°.

The figure also shows why the parameter choice cannot be increased arbitrarily. Moving toward larger Pcb and χ strengthens the isotopic response further, but it also increases the humidity response to W and reduces the separation between the humidity and isotope contours. In other words, sufficiently strong rain evaporation rotates both fields and degrades the observed q–δD asymmetry. The qualitative model result is therefore robust to moderate parameter perturbations, but it is reproduced only within a moderate-rain regime rather than for arbitrary rain forcing.

Figure  provides the mechanistic closure for the observational regressions in Fig. . In the drizzle case, the model produces fitted δD contours that are much more W-sensitive than the fitted humidity contours: |∂δD/∂W|/|∂δD/∂E|=1.66, compared with |∂q/∂W|/|∂q/∂E|=0.44 for humidity. This nearly fourfold contrast is the key quantitative result of the model. It shows that mesoscale ascent can strongly modify isotopic composition through W-dependent rain evaporation while leaving the total water vapor field primarily controlled by entrainment.

The no-rain control is equally important, but it should be interpreted within the structure of this process test. When Frain=Frain,D=0, both q and δD have nearly vertical fitted contours and zero W sensitivity because the model does not include direct W-driven vertical advection of the subcloud-layer budget. Surface evaporation plus cloudy-layer entrainment therefore does not produce the observed non-parallel q–δD structure by itself in this model. Adding a third source term associated with below-cloud rain evaporation provides one physically plausible route to this structure. This does not mean that rain evaporation dominates the moisture budget; rather, the successful drizzle case occupies the regime in which rain evaporation is large enough to rotate the isotope contours but small enough that humidity remains E-dominated.

This balance also constrains the interpretation. In sensitivity tests with much stronger precipitation forcing, the rain evaporation mass flux becomes large enough that q also rotates toward W sensitivity, and the q–δD contour separation collapses. The observed asymmetry is therefore not reproduced by arbitrary rain addition; it requires moderate below-cloud evaporation whose isotope effect is large relative to its net moisture effect. The model should consequently be viewed as an equilibrium process test rather than a unique retrieval of rain rate, rain intermittency, or microphysical parameters.

These findings provide observational and mechanistic context for the shallow mesoscale overturning circulations (SMOCs) identified by and the counteraction mechanism proposed by . demonstrated that mesoscale vertical motions counteract entrainment-driven drying, refuting the mixing–desiccation hypothesis of by showing that stronger mixing does not necessarily lead to cloud-base desiccation. provided direct observational evidence of SMOCs, documenting robust divergence dipoles between the subcloud and cloud layers and showing that convergence branches reduce entrainment efficiency and amplify moisture variance near cloud base.

Our analysis extends these studies by localizing the coupling mechanisms in the vertical. The strongest correlations between vertical velocity and both δD and humidity occur just below the subcloud layer top, identifying where mesoscale circulations most directly influence near-surface composition. This height range is consistent with the layer where mesoscale vertical motion, cloud-base exchange, and evaporation of shallow precipitation are expected to interact most strongly. One plausible interpretation is that the ∼ 100 m offset of the δD correlation maximum below the SCL top marks the layer where mesoscale vertical motion samples condensate-derived vapor recently released by shallow precipitation evaporation near cloud base. We cannot separate this effect from the vertical structure of W alone with the present observations, but the combined observational and model evidence points to a three-flux balance among surface evaporation, cloudy-layer entrainment, and below-cloud rain evaporation. The resulting isotopic contrast between ascending and descending branches exceeds the corresponding humidity contrast.

The pronounced isotopic sensitivity arises because δD responds to moisture source and phase-change history, whereas humidity is constrained by the total mass balance. Entrainment brings relatively dry, isotopically depleted cloudy-layer air into the subcloud layer. Surface evaporation supplies comparatively enriched vapor from the ocean. Below-cloud rain evaporation supplies a third contribution: condensate-derived vapor whose isotopic ratio reflects cloud-layer condensation followed by partial evaporation in unsaturated subcloud air. Mesoscale ascent can change the relative importance of these three fluxes in a way that strongly affects isotopic composition but only weakly alters total water vapor.

This asymmetry is evident in the joint entrainment–vertical velocity (E–W) space, where δD exhibits a strong diagonal organization analogous to the SMOC moisture anomaly structure described by , but with larger amplitude. While both δD and humidity respond to entrainment, mesoscale ascent produces large isotopic enrichment with only modest moistening. Isotopic tracers therefore expose compensation mechanisms that remain largely hidden in humidity alone.

Our results extend existing theoretical and observational frameworks for marine boundary layer isotopic composition. The MBL-mix framework of successfully interprets isotopic variability through surface evaporation and mixing with isotopically depleted air aloft. Our analysis indicates that active mesoscale periods include an additional rain-evaporation pathway, so the subcloud-layer isotope budget cannot always be reduced to a two-endmember balance between ocean evaporation and dry air aloft.

Similarly, the asymmetric isotopic response we document is consistent with the processes outlined by , who showed that water vapor isotopologues provide sensitivity to boundary layer mixing and decoupling processes in stratocumulus regimes that is not evident from humidity alone. Our EUREC⁴A observations demonstrate that this sensitivity extends to the trade cumulus regime and is expressed through mesoscale overturning circulations rather than solely through boundary layer decoupling. The vertical localization observed here provides a mechanistic link between surface isotopic signatures and mesoscale circulation patterns that transport locally evaporated but differentially processed vapor within the boundary layer.

Our findings also complement the analysis of , who emphasized the role of local mixing processes in controlling tropical surface-vapor isotopic variability. The EUREC⁴A results indicate that active mesoscale periods cannot be represented as a simple two-endmember mixture between ocean evaporation and dry air aloft. A third, rain-evaporation source provides a physically plausible contribution because it changes isotopic composition without producing a proportional humidity anomaly.

The strong isotopic sensitivity to mesoscale vertical motion provides an observational constraint on boundary layer coupling that is not available from humidity alone. Large-eddy simulations have shown that shallow-cumulus organization can be coupled to mesoscale aggregation of moisture and convection , and EUREC⁴A observations show that observed moisture variance is strongly expressed near cloud base . Our isotopic observations demonstrate that mesoscale dynamics also leave a clear imprint on near-surface composition, indicating that δD is a sensitive tracer of mesoscale–surface coupling even when humidity anomalies are muted.

While our analysis focuses on mesoscale circulations, cold pools and precipitation also influence boundary layer structure during EUREC⁴A . The strongest isotopic–vertical velocity correlations occur near 500–600 m, generally above the shallowest layers most strongly affected by cold-pool outflows. This suggests that the isotopic signal primarily reflects deeper mesoscale overturning and cloud-base exchange rather than only near-surface cold-pool dynamics, though future work explicitly incorporating cold-pool diagnostics and rain-rate observations would help further separate these influences. The synthetic equilibrium model is evaluated over E=10–35 mm s⁻¹, so it is intended to diagnose the positive-entrainment regime that dominates the observational comparison, not the negative-E cases interpreted above as interface lowering or weak detrainment.

The W–rain relationship used in the model should not be interpreted as universal. It is likely to depend on cloud organization, aerosol conditions, precipitation efficiency, subcloud relative humidity, and the depth over which falling drops evaporate. These dependencies could change the strength of the isotope response in other trade-wind regimes or other forms of shallow convection. Likewise, the 60 h model integrations diagnose the equilibrium response to slowly varying forcing rather than the transient response to any individual 2 h observational sample. Because W enters the model only through the rain-activity function, the model cannot separate W-dependent precipitation effects from any direct vertical-advection effect of mesoscale motion on the SCL moisture budget. The more general result is that water isotopologues provide a sensitive constraint on whether mesoscale circulation changes moisture pathways mainly through entrainment, surface evaporation, rain evaporation, direct vertical advection, or some combination of these processes.

Overall, our results show that the marine boundary layer isotopic composition emerges from the combined effects of surface evaporation, entrainment mixing, below-cloud rain evaporation, and mesoscale circulation modulation. These processes are not independent: mesoscale vertical motion changes the relative strength of each moisture pathway without producing proportional changes in humidity. This asymmetric response highlights the value of isotopic measurements as probes of mesoscale dynamics and provides a framework for incorporating isotope constraints into representations of shallow convection and cloud feedbacks.