The study was conducted in a tropical montane evergreen dwarf cloud forest (‘elfin’ forest; tropical cloud forest) at Bawangling Area of Hainan Tropical Rainforest National Park (109°05′-109°25′E, 18°50′-19°05′N), located in Hainan Island, southern China (Long et al., 2011b). This region belongs to the tropical monsoon climate area with a mean annual rainfall of ~2500 mm, and a distinct wet season from May to October, which accounts for about 80% of the annual rainfall. In 2017, a local meteorological station was established near our experimental site and observations during the 2018 show that the mean monthly rainfall during the wet season was 306 mm, with July emerging as the wettest month with a rainfall of 375 mm ( Supplementary Figure S1 ). The lowest rainfall (58 mm) and frequency (5 days) of rain were observed for February. The dry season (monthly rainfall<100 mm), spanning a period of five months from November to March, is characterized by monthly average rainfall below 78 mm (see also Long et al., 2011b). April was relatively wetter with 138 mm of rain, after which the wet season begins in May. Based on these meteorological data, effective monthly rainfall (monthly rainfall – monthly potential evapotranspiration) was +235.7 mm for July and -16.7 mm for February, indicating only a slight rainfall deficit in the driest month.

The reserve is predominantly comprised of old-growth tropical cloud forest, on a substrate of lateritic soil developed primarily from sandstone bedrocks (Cheng et al., 2020). These forests typically occur as mountaintop islands over 1250 m above sea level, where terrain slope range from 3° to 65° (Long et al., 2011a). The sites typically have very shallow (30-70 cm) soil, high spatial extent (40%) of exposed rock, and very short tree root length (less than 30 cm) (Long et al., 2011c; Yang et al., 2021). The dominant plant species include Distylium racemosum, Syzygium buxifolium, Xanthophyllum hainanense, Camellia sinensis var. assamica and Cyclobalanopsis championii. The average tree height in these forests is rather low at 4.8 ± 2.8 m, but as is typical of cloud forests, average tree density is high at 9633 stems ha^-1 (Long et al., 2011b). A total of 89 tree species have been recorded in 41 plots of 100 m² each (Long et al., 2011a).

We carried out field sampling and measurements in the months of February and July of 2019, primarily to quantify the contributions of various water sources to leaf water in tree and epiphyte species in the wettest and driest months of the year ( Supplementary Figure S1 ). The sampling was carried out in 21 vegetation plots, each of size 20 × 20 m², which we had established in previous work (Long et al., 2011b). These plots are located within a narrow elevation range of 1313 m to 1395 m above mean sea level (Long et al., 2011b; Supplementary Figure S2 and Supplementary Table S1 ). The plots are separated from each other by about 100 m and do not show significant spatial autocorrelation in species abundances or soil properties (Long et al., 2015). The total area of the tropical cloud forest in the reserve is just about 3 hectares, and the 21 plots are scattered as widely as possible across this mountaintop patch ( Supplementary Figure S2 ).

In this study, we recorded all freestanding trees with a diameter of ≥ 1 cm at breast height (DBH) within each plot and identified them to species. The relatively low tree height in this cloud forest allowed us to accurately measure species abundance (total number of individuals) for all epiphyte species on the host trees in the 21 plots. We followed the method proposed by Sanford (1968) to record the species abundances for all vascular epiphytes in the plots. The specific details are given in Supplementary Data Sheet - Text S1 in the Supplementary Material .

To measure the contribution of different water sources to leaf water in the wettest and driest month of the year, we measured hydrogen and oxygen isotope ratios (δ²H and δ¹⁸O) in leaf water (for tree and epiphyte communities), rain water, soil water, and fog drip water, for both of these seasons. We also measured soil water content using standard gravimetric analysis. For isotope measurements of leaf water, we sampled 20 mature, healthy, sun-exposed canopy leaves from three to five individuals of each tree and epiphyte species present in the 21 plots. Due to the evaporative enrichment of the heavier isotope in leaves, stem xylem water isotope ratios may provide a better integrated signal of plant water (Zhao et al., 2016). However, given the challenges associated with sampling stem water, particularly for epiphytes, we focus our study on leaf water. Many epiphyte species in our tropical cloud forest are orchids with pseudobulb stems, so extracting their stem xylem water would have been difficult (Wang et al., 2019). We note that evaporative enrichment is lower in cloud forests because leaf wetting, fog, and high atmospheric humidity reduce transpiration rates (Alvarado-Barrientos et al., 2014). Finally, leaf water analysis may be a relatively non-intrusive method for quantifying the sources of plant water (Benettin et al., 2021).

We also collected water samples of fog drip and rainfall within each plot purely to measure isotope ratios (and not to quantify fog precipitation). We followed Liu et al. (2005) for fog drip collection, wherein a simple self-made fog drip collector made of plastic film was used to intercept and collect the fog droplets formed upon contact ( Supplementary Figure S3C ). Specifically, fog drip in each plot was collected by hanging a clean plastic film between the two trees, with clear exposure on the windward side ( Supplementary Figures S3B, C ). We set this from 19:00 to 21:00 h when heavy fog had set in and when there was no rain. The intercepted smaller fog droplets condense and gradually coalesce to form large water droplets, which are collected in a storage tank at the lower end of the plastic film. Finally, each condensed water sample was saved in a 20 ml tube. We collected fog drip water, soil water, and rain water separately. Rainfall samples were collected under the trees using a 20 ml tube at the beginning of a rainfall event. All the collected plant, soil, fog, and rainfall samples were stored in liquid nitrogen containers before measuring isotope ratios (δ²H, and δ¹⁸O). Due to the shallow soils and short tree root length, we sampled soils at 0-20 cm depth, collecting three samples (20 g each) in each plot, using a 5 cm diameter soil auger, in both the months. We collected soil water on a day when there was no rain. Before performing the isotope analysis, we needed to extract leaf water and soil water from all the samples, which we did using a fully automatic vacuum condensation extraction system (LI-2100, LICA United Technology Limited, Beijing, China). The extraction rate of water from the samples was assessed to be more than 98% on the basis of gravimetric analysis.

For plant trait measurements, undertaken in February and July, we severed a small branch from each tree being sampled and measured a set of plant traits including maximum photosynthesis rate (A_area ; μmol cm^-2 s^-1), transpiration rate (mol m^-2 s^-1), stomatal density (number of stomata mm^-2 of leaf area), stomatal conductance (mmol m^-2 s^-1), leaf turgor loss point (TLP; MPa), and leaf δ¹³C for tree and vascular epiphyte communities. To minimize intraspecific variation in trait measurements, we selected only individuals with DBH near the species mean value. We also ensured that leaves were collected from the same individuals for each species in both the wettest and driest months.

For the dry season measurements, we carried out the sampling and field measurements from February 5 to 10, 2019, during which period rainfall occurred only on February 10. We matched these measurements for the wettest month and started sampling on 5^th July and completed the field work by the 10^th of July. Although it rained almost every day in the month of July, we had only night rains from July 5 to 6, 2019. Thus, we collected soil water samples and leaf samples (for measuring leaf isotope) in the mornings of July 5 and 6 for the wet season. We collected fog drip from 19:00 to 21:00 h on the nights of July 8 and February 8 and rainfall samples on July 10 and February 10. This study design enabled us to quantify 1) the hydraulic responses of the tree and epiphyte communities in the wettest and driest parts of the year; and 2) the relative contributions of rainfall, fog, and soil water to the water use strategies of tree and epiphyte communities in wettest and driest conditions.

We measured δ¹³C using the conventional Pee Dee Belemnite standard (Farquhar et al., 1989). Then, we sampled 0.5-1.5 ml of leaf water, soil water, fog and rainfall, to measure δ¹⁸O and δ²H. The isotopic compositions were analyzed using a liquid water isotope analyzer (Model DLT-100, Los Gatos Research, USA) that employs cavity enhanced absorption spectroscopy. The precision of the isotope analyzer was typically better than ± 1.2‰ for δ²H and ± 0.3‰ for δ¹⁸O. To account for the possible presence of organic contaminants in cryogenically extracted water samples from plant tissues, the stable isotopic ratios of leaf water measured by the LGR system were corrected as described in previous studies (Schultz et al., 2011; Wu et al., 2016), and the specific details are given in Supplementary Data Sheet - Text S2 in the Supplementary Material .

In 2017, an automatic weather station (YT-QXC4, Shandong, China) was installed in a central location, near the 21 sampling plots (also in forested area) and situated at an elevation of 1245 m a.s.l. ( Supplementary Figure S2 ). This installation enabled the continuous monitoring and collection of data - rainfall frequency (number of rainy days in a month), and monthly mean values for rainfall, total solar radiation, air temperature, humidity and wind speed. We also surveyed fog duration or timing from 6:00 to 23:00 h in July and February by observing for fog on each day. Specifically, nocturnal fog was observed by setting up a light in the plot. We used about 20 g soil for each of the 21 soil samples to measure gravimetric soil water (g/kg). All soil samples were oven-dried for 24 h at 105 °C for these measurements.

We used the Li-6800 portable photosynthesis system (Li-Cor, Lincoln, Nebraska, USA) to measure maximum photosynthesis rate, transpiration rate and stomatal conductance between 9:00 AM and 11:00 AM on sunny days. Five canopy leaf-bearing branches at similar heights (~20 mm in diameter) were harvested, and then photosynthetic measurements were taken within 1 h (Wyka et al., 2012). Based on preliminary trials, we set the photosynthetic photon flux at 1200 μmol m^-2 s^-1 to ensure all the species were measured for light-saturated photosynthetic rates (Zhang et al., 2018). We set chamber CO₂ and air temperature as 400 μmol mol^-1 and 28 °C, respectively. Before collecting the data, we first exposed the leaves to the above conditions for about 5 minutes to allow photosynthetic parameters to stabilize. We sampled five to six fully expanded and sun-exposed leaves from three to five mature individuals to measure maximum photosynthesis rate, transpiration rate and stomatal conductance, whose values are referred to as leaf area units.

Stomatal density was measured using the protocol in Carins Murphy et al. (2012). We first collect leaf surface film and then use an optical microscope (LEICA DM3000 LED) and Image J software to calculate stomatal density in the leaf cuticles (2 per leaf and 5 fields of view per cuticle).

Leaf turgor loss point was determined from leaf pressure-volume (P-V) curve (Sack et al., 2003) for each species in both seasons. For the measurement of P-V curve, we selected healthy leafy branches (or entire plants for several small epiphyte species) from five individuals in the early morning (05:00 and 07:00 h). The samples were packed in black plastic bags with the cut ends maintained underwater, and were immediately sent to the laboratory (within 1 h). The sampled leaves were water saturated because leaf water potential was higher than -0.3 MPa, and did not show a decrease during transportation. For each measured leaf, we determined saturated leaf mass and subsequently conducted a bench-drying procedure (dehydration on a lab bench at 25°C) to obtain a range of leaf water potentials. During leaf desiccation, we periodically measured leaf mass and leaf water potential (K _leaf) by using a precision scale (0.0001g) and a pressure chamber (PMS, Corvallis, OR, USA), respectively. Finally, the dry leaf mass was determined after drying in an oven at 70°C for about 72 h. We calculated relative water content (RWC) and then constructed P-V curve by plotting leaf RWC against K _leaf. Leaf P-V curves of the measured species showed distinct two-phase linear equation, and leaf water potential at turgor loss point (TLP) was estimated from the point of intersection of the two lines (Brodribb and Holbrook, 2003; Wang et al., 2021).

Leaf hydraulic capacitance (C) was determined from the slope of P-V curve for each species, with the help of following equation (Brodribb and Holbrook, 2003).

Here, DW is leaf dry weight (g), LA is leaf area (m²), WW is mass of leaf water at 100% RWC (g), and M is molar mass of water (g mol^-1). δRWC/δΨ_l could be attained from P-V curve.

Community weighted mean (CWM) of a functional trait is defined as the summation over all species in the community of the species mean trait values weighted by their respective relative abundances (Garnier et al., 2004). CWM is considered a good measure of ecosystem response to abiotic factors (Lavorel and Garnier, 2002; Suding et al., 2008; Laughlin, 2014). Therefore, for each of the 21 plots, community-level trait metrics or isotopes (CWM _jk ) for the tree and epiphyte community were quantified for each trait or isotopes (j) in each plot (k) following Garnier et al. (2004) and Buzzard et al. (2016) and using the following formula:

where x_ik is the relative abundance of species i in plot k and t_ik is the isotope or leaf functional trait value of species i in plot k. In reality, CWM _jk was calculated using function ‘dbFD’ in the FD package in R.

The Bayesian stable isotope mixing model (Parnell et al., 2013) is widely used for tracing the proportional contributions of various sources to a stable isotope mixture. It is based on isotopic mass conservation and that the isotopic mass for the mixture should contain the signature of the proportional contributions of the isotopic masses from all its potential sources (Wang et al., 2019). As long as the multiple isotope sources have clearly different isotope ratio signatures, one can reliably apportion the contributions from these sources to a mixture. The Bayesian approach can incorporate priors such as rooting depth and other attributes that affect the relative contribution of different sources. In the cloud forest ecosystem, leaf water in trees can originate from soil water, fog, and rainfall, whereas the epiphytes we studied can access water only from the air via rainfall and fog and cannot tap into the host tree xylem water. Thus, by using Bayesian stable isotope mixing model and the isotope ratios (δ²H and δ¹⁸O) for leaf water, soil water, fog, and rainfall, we could trace the proportional contributions of soil water, rainfall, and fog to foliar/leaf water (Wu et al., 2018). This model was implemented using the ‘siarmcmcdirichletv4’ function in R (siar package), an algorithm which uses the distinctive δ²H and δ¹⁸O isotope ratios to quantify the relative contributions (%) of fog water, rainfall, and soil water to leaf water in tree and the epiphyte communities in the 21 plots for both the wettest and driest months.

We performed a series of comparisons to understand the differences in functional traits between life forms, differences in meteorological variables between seasons, and the impact of fog on plant physiology. We used a Wilcoxon Rank Sum test to compare differences in CWMs of maximum photosynthesis rate, transpiration rate and stomatal conductance and density for the tree and the epiphyte communities in the wettest and driest months. Further, we compared four meteorological variables (mean rainfall, total solar radiation, mean air temperature, and mean relative atmosphere humidity), and the soil water content between the two seasons. We also used Wilcoxon Rank Sum to compare the differences in CWMs of δ¹³C and leaf turgor loss point (TLP) between tree and epiphyte communities during the wettest and driest months. The objective of this analysis was to examine whether the tropical cloud forest ecosystems can maintaining a sufficient water supply in the driest month. Additionally, we determined whether there were any significant differences in CWMs of δ¹³C and leaf turgor loss point between the tree and epiphyte communities.

In July, the mean daily rainfall, total solar radiation, and mean atmospheric temperature were recorded 12.11 mm, 22.23 MJ m^-2, and 25.46°C. Conversely, in February, these values decrease to 2.07 mm, 8.64 MJ m^-2, and 17.49°C ( Figures 1A–C ; Supplementary Figure S1 ). both July and February exhibited mean relative atmospheric humidity levels exceeding 90%, with July registering a slightly higher value of 98.63% compared to 91.53% in February. However, this discrepancy did not reach statistical significance ( Figure 1D , P>0.05). Overall, soil water content was also not significantly different between July and February ( Figure 1E , P>0.05). Rainfall occurred every day in July, whereas there were just five days of light rain in February ( Figure 1F ). No discernible differences in fog occurrence were observed between July and February, as fog was consistently present during early morning hours (before 9:00 h) and night-time (after 19:00 h) daily in both months ( Figure 1G ). In other words, the driest month is marked by much lower rainfall, attenuated total solar radiation, and lower temperature, while some key attributes like fog frequency, air humidity, and soil water content were not significantly lower in this month compared to the values in the wettest (summer) month ( Figure 1 ).

We studied a total of 60 tree and 30 vascular epiphyte species belonging to a total of 38 families across the 21 plots ( Supplementary Table S2 ). The common tree species (relative abundance >5%) include Distylium racemosum, Psychotria rubra, Syzygium buxifolium, Ervatamia officinalis, and Symplocos poilanei, and the dominant vascular epiphyte species were Eria thao, Coelogyne fimbriata, Liparis delicatula, and Bulbophyllum retusiusculum ( Supplementary Table S2 ). Our sampled epiphyte community could be considered as totally dependent on atmospheric water (rain, fog, dew) for their leaf water, as all the epiphyte species in the 21 plots were anchored on the tree stems but could not tap into the tree xylem water ( Supplementary Table S2 ).

As evident from the scatter plot in Figure 2 , there were notable variations in the δ²H and δ¹⁸O isotopic values across various water sources (including soil water, rainfall, fog, and leaf water), between the tree and epiphyte communities, and between the months of July and February ( Figure 2 ). Notably, the leaf water isotope values were different between seasons for both trees and epiphytes ( Figure 2 and p<0.001, Wilcoxon signed-rank tests, Supplementary Table S3 ). An important exception being δ²H and δ¹⁸O values of leaf water for the tree community, which were similar to the values for soil water in the wettest month ( Figure 2 , p>0.05, Wilcoxon signed-rank tests, Supplementary Table S3 ). Another exception was for leaf δ²H and δ¹⁸O values for the epiphyte community, which were similar to the values for rainfall in July and fog in February ( Figure 2 , p>0.05, Wilcoxon signed-rank tests, Supplementary Table S3 ).

Given the high and persistent rainfall in the wettest (summer) month, we found that soil water contributed nearly 100% of the leaf water for tree communities and rainfall contributed nearly all the leaf water for epiphyte communities ( Figure 3A ). The contribution of fog water to foliar water uptake was insignificant for the tree or the epiphyte community in the wettest summer month ( Figures 3A, B ). The contribution of soil water to leaf water declined to 46.4% for the tree community despite no decrease in soil water content in the dry/winter month. Fog contributed 52.3% of leaf water for the tree community in the driest/winter month ( Figures 3A, B ). Furthermore, the contribution of soil water to leaf water for the epiphyte community was extremely limited and infrequent (<2%; Figures 3A, B ). Thus, in summer month, leaf water in trees appears to be entirely due to soil water uptake and leaf water in epiphytes was obtained from rainfall. In the peak dry/winter season, nearly all (99.2%) the leaf water of the epiphyte community was sourced from fog, while the tree community used both soil water and fog water uptake almost equally to maintain leaf water supply ( Figure 3A ).

We computed correlations of isotope ratios of rain and fog water to soil water in the wet (July/summer) and dry (February/winter) seasons. Isotope ratios of rain and soil water showed a highly significant correlation during the wet season (R² = 0.61, P<0.005; Figures 3C–F ), when heavy rainfall occurs on a daily basis ( Figure 1 ), whereas no significant correlation was observed during February (R² = 0.21, p>0.05; Figures 3C–F ), when the rainfall was nearly 40 times lower ( Figure 1 ). On the other hand, isotope ratios of fog and soil water were significantly correlated in the dry/winter season (February; R² = 0.45, p<0.05; Figures 3D–F ). In comparison, there was no significant correlation between isotope ratios of fog water and soil water in the summer month (July; p>0.05; Figures 3D–F ). In other words, rainfall is the main water input for the soil water during wet summer months, while fog contributes significantly to soil water in the dry winter season.

We observed reduced photosynthesis and transpiration rates for both plant communities in the driest (winter) month compared to the wettest (summer) month: CWMs of photosynthesis rate in February (2.9 μmol cm^-2 s^-1 and 4.9 μmol cm^-2 s^-1 for the epiphytes and the trees, respectively) were approximately one-third to one-half of same values in July (8.2 μmol cm^-2 s^-1 and 10.7 μmol cm^-2 s^-1, for the epiphytes and the trees, respectively; p<0.001, Wilcoxon signed-rank tests, Figures 4A, B ). Similarly, CWMs of transpiration rates for both the plant communities in February (0.001 and 0.002 mol m^-2 s^-1 for the epiphytes and trees, respectively) were about one-third of those in July (0.003 and 0.0067 mol m^-2 s^-1, respectively; p<0.001, Wilcoxon signed-rank tests, Figures 4C, D ).

We also found a reduced stomatal density and stomatal conductance for both the plant communities in the driest (winter) month compared to the wettest (summer) month: CWMs of stomatal density for both the plant communities in February (62 stomata mm^-2 of leaf and 316 total stomata mm^-2 of leaf for the epiphytes and the trees, respectively) were about one-third to one-half of those in July (203 stomata mm^-2 of leaf and 676 stomata mm^-2 of leaf, for the epiphytes and the trees, respectively) (p<0.001, Wilcoxon signed-rank tests, Figures 4E, F ). Similarly, CWMs of stomatal conductance in February (0.06 mmol m^-2 s^-1 and 0.1 mmol m^-2 s^-1 for the epiphytes and the trees, respectively) were approximately half of corresponding values in July (0.11 mmol m^-2 s^-1and 0.2 mmol m^-2 s^-1) for epiphytes and trees (p<0.001, Wilcoxon signed-rank tests, Figures 4G, H ).

For the tree community, CWM values of δ¹³C were -31.32 ‰ and -31.97 ‰ for July and February respectively, and CWMs for TLP were -1.07 MPa and -1.06 MPa for July and February, respectively. The CWM values for the tree community did not differ between July and February for δ¹³C and TLP (p>0.05, Wilcoxon signed-rank tests, Figures 5A, B ). In contrast, for the epiphyte community, the CWM values of δ¹³C (-28.95 ‰ and -31.97 ‰ in July and February, respectively) and TLP (-1.06 MPa and -1.61 MPa in July and February, respectively) were all significantly more negative (p<0.001, Wilcoxon signed-rank tests, Figures 5A, B ), in February compared to July ( Figures 5A, B ).

A significantly increased leaf hydraulic capacitance (C) was observed for both the plant communities in the driest (winter) month as compared to the wettest (summer) month: CWMs of C in February (4.77 mol m^-2 MPa^-1 and 4.19 mol m^-2 MPa^-1 for the epiphytes and the trees, respectively) were approximately 2.2-2.5 times the corresponding values in July (1.88 mol m^-2 MPa^-1 and 1.72 mol m^-2 MPa^-1, for the epiphytes and the trees, respectively; p<0.001, Wilcoxon signed-rank tests, Figure 5C ).

Consistent with the previous observations (Scholl et al., 2011; Brinkmann et al., 2018; Schihada et al., 2018; Zhan et al., 2020), our results clearly show that soil water, rainfall, and fog have very different isotope (δ²H, and δ¹⁸O) concentrations with the changing intensity in rainfall between seasons. Therefore, the Bayesian stable isotope mixing model could apportion the relative contributions of fog, soil water, and rainfall to leaf water for tree and epiphyte communities. We found a clear contrast between the source of leaf water in July (wet summer) and February (peak of a dry winter season), with nearly all the leaf water in trees in the wettest summer month being taken up from soils. We also found that transpiration rates were significantly high in July (>3 times that in the winter month), and the soil water being continuously replenished by the daily rainfall (indeed, the isotope signals for soil water were highly related to those for rainfall). In this scenario, it is most likely that the trees rely on the SPAC pathway (instead of foliar uptake of rainfall). On the contrary, during dry winter weather, the temperatures are low, transpiration rates are much lower, and photosynthesis is also strongly reduced. In such scenario, we infer that soil water requirement (and thus its uptake) is reduced, and a FWU (of fog) could be facilitated by trees so that the leaf turgor remains maintained.

Pathways/strategies for FWU vary across different plant species (as reviewed extensively by Berry et al., 2019, and by Schreel and Steppe, 2020), and it is plausible that the net cumulative effect of FWU at the community/ecosystem levels are far reaching than initially thought, a hypothesis that our data indicates. Further research is warranted in this direction.

By examining our results in totality, we could infer that rain is a major contributor of soil water during summers and this water is taken up by the trees by adapting a traditional SPAC flow. But during dry winters, the trees might additionally adapt a strategy involving FWU of fog to replenish leaf water, while soil water uptake may still remain the primary source of transpired water. Further, there could be potentially confounded contributions of fog to soil water to some extent. It is possible that FWU might decouple leaf-gas exchange from soil water availability (Schreel and Steppe, 2020), thus further complicating the quantification of the different sources to leaf water. It is also conceivable that trees might also tap deep underground water in the driest month, but some scholars have already reported that this tropical cloud forest has very shallow soil (less than 30 cm) and short root length (less than 30 cm) (Long et al., 2011c; Yang et al., 2021), therefore, this is very unlikely (Hildebrandt and Eltahir, 2007).

Surprisingly, we found that TLP and δ¹³C do not differ between the peak wet and dry seasons even though TLP and δ¹³C are good indicators of general water availability or water stress (Farrell et al., 2017; Acosta-Rangel et al., 2018). Also, these are highly associated with photosynthesis rates, as they influence the ability of plants to maintain cell turgor under drought conditions and reflect long-term water-use efficiency (Nogueira et al., 2004; Bartlett et al., 2012). The stable TLP and δ¹³C values for the tree community and comparable soil water content between July and February indicate adequate water availability to the trees in both seasons. Any difference in water availability between seasons may in part be compensated by differences in leaf hydraulic capacitance (Goldsmith et al., 2013). The observation that leaf hydraulic capacitance in the dry season was 2.5 times than in the wet season indicates modification of leaf traits for better water retention in the dry season. All things considered, the low rainfall input in the driest month could have been compensated with increased leaf hydraulic capacitance, suppressed photosynthesis rate and transpiration, low temperature, high humidity, and possibly a fog input. These conditions appear sufficient to maintain adequate water availability for this cloud forest, at least for the tree and epiphyte communities (discussed below).

There remain some gaps in our understanding of water cycling for the tree community. We need further investigation to understand the mechanism of FWU for tree community in the dry season. We also need specific data on leaf temperature, leaf water potential, soil water potential, and sap flux measurements to understand the direction and flow rates of water movement. The effects of leaf phenology should also be considered in interpreting the effects of variation in solar radiation, temperature, and rainfall on this tropical cloud forest ecosystem. To assess fog inputs, we need data on fog intensity and duration, above-canopy observations of climatic variables, wind-driven rain and continuous soil water content measurements alongside sap-flow observations of transpiration (Holwerda et al., 2006; Schmid et al., 2011).

Given that the epiphyte community on Hainan Island derives its water fully relies on atmospheric sources, such as fog and rainfall, which could be directly absorbed by their leaves and roots, and could serve as leaf water resources in both the seasons (Gotsch et al., 2015, 2018; Wu et al., 2018). Although, fog and rainfall are equally frequent in July, rainfall can provide much higher water input than fog in the wet season. Moreover, photosynthesis rate and transpiration for epiphyte community are relatively high in July, which should favor root water absorption. In a peak dry winter period, fog still occurs every day, and the resultant leaf wetting should favor its uptake.

In February the epiphyte community has lower TLP than that in July. Further, the low temperature, and limited and infrequent rainfall in the driest month may cause epiphytes to lower their osmotic potential and to increase the uptake of fog water (Bartlett et al., 2014; Gotsch et al., 2015, 2018). Indeed, we observed that fog acted as a main source for leaf water for the epiphyte community in the driest month. We also found δ¹³C in February was higher than that in July, indicating higher water use efficiency under lower water availability (Acosta-Rangel et al., 2018). However, water use efficiency could also be affected by the changes in the photosynthesis rates (Nogueira et al., 2004), and the suppressed photosynthetic rates in February could have counteracted the increase in water use efficiency to a certain extent.

Previous studies have documented reduced photosynthesis and transpiration rates during the arid winter season, but limited to a select few species (Burgess and Dawson, 2004; Goldsmith et al., 2013; Alvarado-Barrientos et al., 2014). We show that community level ecophysiological responses (such as reduced photosynthesis rate and transpiration, and fog utilization) vary with the quantity of rainfall and life-forms (trees vs. epiphytes). These findings therefore expand prior knowledge on the varied hydraulic responses as a function of water availability and life-forms from the species level (Oliveira et al., 2014; Gotsch et al., 2015; Wu et al., 2018; Berry et al., 2019) to the community level.

We also found that fog had different effects on tree and epiphyte communities. Given that the epiphytes here cannot tap into the host tree xylem water, foliar uptake of fog water could be critical to ensure leaf water supply for the epiphyte community when rainfall is critically low in the driest month. However, reduction in temperature and suppression of photosynthesis might result in lower leaf water use efficiency and higher leaf turgor loss point for the epiphyte community. Fog contributed to leaf water supply for tree community in the cold dry month, when rainfall was limited, and was therefore important for the tree community as well.

Detailed knowledge of the hydrological feedbacks of tropical forest ecosystems is only emerging (Staal et al., 2020), but there are predictions of dramatic losses (>50%) of tropical cloud forest cover due to climate warming and increase in cloud-base heights [predictions for 2052 by IPCC 5th assessment reports; as well as studies by (Los et al., 2019)]. Another independent simulation study predicts significant loss of tropical cloud forests in Mexico by 2080 if temperatures were to rise by >4°C and surrounding lowland forests were lost (Ponce-Reyes et al., 2012). We found that fog water made little contribution to tree and epiphyte communities in the wettest month, reasons of which need further investigations, which are out of scope of this report. However, in the peak of the dry season, fog appears to be important for water safety, hence the importance of maintaining cloud immersion for cloud forest persistence (Wu et al., 2018).

Forest ecosystems provide the most sustainable and highest quality freshwater (Vose, 2019) and soil water content appears to be the key determinant of freshwater supply from forest ecosystems (Neary et al., 2009). Sustaining this important ecohydrological process involves maintaining the structural and functional integrity of the cloud forest ecosystem and avoiding serious climatic changes (Zhan et al., 2020).