3.1.1 Relative humidity, radiative cooling and temperature

The 11 grasslands and 19 forests can be characterized at night by their mean surface temperature and mean relative humidity (from 23:00 to 05:00 the the next day) over the growing season. A general positive trend is observed in Fig. 2a and is mostly explained by the effect of relative humidity on radiative cooling (Fig. 3a). Indeed, a higher water vapor content will make the air more efficient at absorbing and re-emitting longwave radiation, therefore the nighttime boundary layer will be more opaque to radiative loss, and this mechanism will keep the surface temperature warm at night. Additionally, sites with a high elevation like MOAB (1779 m), ONAQ (1662 m) and RMNP (2742 m) are associated with cold and dry air due to the low pressure. The relative humidity within the canopy of the forests is usually higher than the grasslands at night (86±9 % vs. 76±15 %, 1 SD) not only because of a denser canopy structure but also because forests occur in wetter regions than grasslands (Woodward et al., 2004). The two exceptions here are the grasslands LAJA and DSNY, which present tropical characteristics ($\mathrm{RH}=\mathrm{95}\pm \mathrm{3}\mathrm{\%}$, 1 SD).

Figure 3Mean values from 23:00 to 05:00 the next day for each site over the growing season for the following parameters: R_n versus RH (a), difference in T_s between the canopy within and the top of the canopy versus the canopy height (b), and T_a−T_s (within canopy for the forests, top of the canopy for the grasslands) versus R_n (c). Four sites are selected for discussion.

Download

Measurements performed at different heights provide the opportunity to study the vertical variability in the surface temperature and relative humidity. The difference in surface temperature between the bottom and the top of the canopy (Fig. 2b) is similar at night for the forests ($\mathrm{0.3}\pm \mathrm{0.5}{}^{\circ }\mathrm{C}$, 1 SD) and the grasslands ($\mathrm{0.7}\pm \mathrm{0.6}{}^{\circ }\mathrm{C}$,  SD) based on a t-test ($p=\mathrm{0.07}>\mathrm{0.05}$). One could consider that the vertical variability in the surface temperature at night is larger for tall-canopy ecosystems than low-canopy ecosystems because radiative fluxes are exchanged in a bigger volume. Although forests have a much higher canopy height than the grasslands (5–40 m of difference; Fig. 1), the vertical variability in their surface temperature is not larger at night. One possible explanation is that forests are wetter and have more biomass than grasslands. Their canopy will therefore cool down slower, as will the ground, because of a higher heat capacity and inefficient radiative cooling (Fig. 3a). The positive relationship observed among the forested sites in Fig. 3b seems not to be primarily driven by the canopy height. The ground flux or cloud cover data (absent from this analysis) might play an important role in the vertical gradient of surface temperature at night, as well as the vegetation density and the site wetness. Looking now at the relative humidity, it reveals to be always higher within the canopy than above the canopy for both ecosystem types (Fig. 2c). Indeed, air parcels within the canopy are not effectively mixed with drier air higher in the atmosphere (low wind speed within canopy), and the combination of soil evaporation and transpiration provides continual moisture to the canopy air space.

The difference in temperature between the air and surface (T_a−T_s) is mainly driven by radiative cooling for most sites (Fig. 3c). Interestingly, three grasslands (LAJA, DSNY and HEAL) and one forest (DEJU) with the shortest canopy height (6 m) show a large T_a−T_s despite minimal radiative cooling (Fig. 3c). These four sites do not belong to the same population: HEAL and DEJU are both located in Alaska (cold and dry air), while LAJA and DSNY are tropical (warm and wet air). Moreover, HEAL and DEJU are the only two sites that are not part of the trend in the relationship between R_n and RH (Fig. 3a). It is not clear at this point what drives these sites away from the otherwise strong relationship. The ground flux could act as a sink of energy at night for these four sites and could decrease the temperature of the canopy, which would consequently reduce the radiative cooling. Another explanation could be that nocturnal atmospheric dynamics such as the presence of warm convection above the canopy (increasing the difference in temperature) associated with an opaque cloud cover (drastically reducing the radiative cooling, even if the relative humidity is below 85 % for HEAL and DEJU; see Fig. 3a). Our current data are not sufficient to verify these hypotheses, and further investigations are required to understand what is driving T_a−T_s when the radiative cooling is weak. In a study using RDCs in tropical areas (Tahiti and Tikehau, French Polynesia), Clus et al. (2009) observe that the relative humidity impacts the efficiency of the radiative cooling, however these sites cannot be compared to LAJA or DSNY because of the presence of strong winds, a much lower relative humidity (∼80 % at night compared to ∼95 %) and the absence of surface temperature data except for one night.

3.1.2 Dew frequency, duration and yield across sites

The most frequent location where dew forms is on the top of the canopy of grasslands, which is much cooler than the ground (Fig. 2b), and within the canopy of the forests, where the relative humidity is much higher than above the canopy (Fig. 2c). Indeed, the dew frequency within the canopy of the grasslands is only 7 % compared to 29 % at the top of the canopy (DSNY and LAJA are excluded because of their tropical characteristics). For forested sites, dew forms only 11 % of the nights at the top of the canopy of the forests compared to 25 % within canopy, although the position of the RH sensors above the canopy for the forests is far away from the top of the canopy (difference of ∼15 m), which is why dew frequency at the top of the canopy might be underestimated for the forested sites. This study will therefore focus on the dew formation at the top of the canopy of the grasslands and within the canopy of the forests only.

The grassland sites show a large gap in temperature between the air and the surface at night (Fig. 3c). Consequently, the main limitation of dew formation at the top of the canopy of the grasslands will be the relative humidity. A threshold of RH=75 % separates two patterns for the dew frequency in the grasslands (Fig. 4a): (i) a low and constant dew frequency of ∼15 % of nights for RH<75 % and (ii) a linear increase in dew frequency with rising relative humidity (slope of 3.5, r²=0.92, p value <0.01) for RH>75 %. We have isolated three populations of grasslands based on their relative humidity: dry (six sites with RH<80 %), temperate (three sites with $\mathrm{80}\mathrm{\%}<\mathrm{RH}<\mathrm{90}\mathrm{\%}$) and tropical (two sites with RH>90 %). The dry, temperate and tropical sites have a dew frequency of 15±6 % (1 SD), 59±4 % (1 SD) and 95±4 % (1 SD), respectively. This dew frequency versus relative humidity relationship revealed by several grasslands with different climatic conditions has not yet been reported in the literature for the following reasons: the former dew studies were limited to one or two sites; there is a lack of standardized measurements between them; there is an absence of continuous humidity measurements or a decision to not report them (Xiao et al., 2009); and finally the sites are mostly chosen for their high nocturnal RH, especially in arid areas close to a coast (Zangvil, 1996; Muselli et al., 2009; Moro et al., 2009; Uclés et al., 2014; Maestre-Valero et al., 2015). The in situ radiometric network from NEON has not been specifically designed to work in locations with high dew yield, and thus the data provide information on sites that span a continuum from exceptionally low dew frequencies to an almost nightly occurrence. However, former dew studies focusing on a single site have reported a similar RH threshold value below which dew has a low probability of forming: RH=78 % in Wang et al. (2017c) (204 nights in a semi-arid site in China using leaf wetness sensors) and RH=75 % in Guo et al. (2016) (166 nights in a cold desert in China using eddy-covariance data). Even if the linear relationship in Fig. 4a is strong, the relative humidity alone is not sufficient for predicting dew frequency with a good accuracy (the residuals have a standard deviation of 10 %), which highlights the role of canopy structure and wind speed in dew formation (Sect. 3.2). See Beysens (2018) for a detailed explanation of the relationship between dew formation and relative humidity.

As mentioned in the previous sections, forested sites have a higher relative humidity within canopy than the dry and temperate grasslands at night (Fig. 2a) but a lower air–surface temperature gap (Fig. 3c). The relationship between dew frequency and relative humidity within the canopy can still be considered to be linear for RH>80 % (Fig. 4a; slope of 2.3), but with a much lower coefficient of determination (r²=0.31) because the inefficient radiative cooling or very low wind speed within canopy are also limiting dew formation in these ecosystems. Forests have a mean dew frequency of 25±14 % (1 SD), which is between the dry and temperate grasslands. Interestingly, the forest DEJU is consistent with the linear relationship calculated from the population of grasslands (Fig. 4a), which is certainly because it has the shortest canopy height (6 m) among the forested sites. Dew formation is less well documented in forested sites than arid areas or croplands and/or grasslands, possibly because small inputs from dew are not likely to have a significant impact on these ecosystems or because their tall canopies are considered as a natural radiative barrier to dew formation. However, Hao et al. (2012) reported a dew frequency of 73 % over 142 nights using eddy-covariance data in a hyper arid forest (canopy height 8–10 m, forest coverage of 50 %, and annual rainfall of 35 mm). The much higher dew frequency might be explained by the short canopy height, the low vegetation density, the presence of a river close to the area of study or simply the poor performances of the eddy-covariance method at night (de Roode et al., 2010). Unfortunately, nighttime relative humidity measurements are not available and cannot be compared with our study. In a sparse elm wood site (unknown canopy height, forest coverage of 40 % and annual rainfall of 385 mm), Wang et al. (2017a) measured a dew frequency of 54 % over 80 nights using weighing methods, a value which is much more consistent with our data. Again, relative humidity measurements are not available for the period of study.

Figure 4Dew frequency (a) and duration (b) versus RH (from 23:00 to 05:00 the next day) for each site over the growing season. The slope of 3.5 (r²=0.92) in (a) has been calculated based on the grasslands with RH>70 % only. Dew formation has been calculated at the top of the canopy of the grasslands and within canopy of the forests only.

Download

We define dew duration as the time during which the dew point temperature is above the surface temperature at night. The dry and temperate grasslands have a dew duration of 3.4±1.1 h (1 SD), and the forests have a duration of 4.5±1.1 h (1 SD). A comparison with former studies is difficult because of the varying range of dew duration in the literature: from 1–4 h (Pan et al., 2010; Hao et al., 2012) to 4–6 h (Wang et al., 2017a) or above 6 h (Moro et al., 2009; Uclés et al., 2014; Guo et al., 2016) depending on the climatic conditions, canopy structure and measurement protocol. Dew duration is positively affected by the relative humidity (Fig. 4b), but it is also sensitive to a change in wind speed (see Sect. 3.2 for a detailed analysis of the sensitivity of dew duration to these parameters). The two tropical grasslands (LAJA and DSNY) show a much higher mean dew duration (8.8±0.1 h, 1 SD) compared to the rest of the sites (Fig. 4b). Both have high nocturnal relative humidity ($\mathrm{RH}=\mathrm{95}\pm \mathrm{3}\mathrm{\%}$, 1 SD) with a surprisingly high difference in temperature between the air and the surface of the canopy at night ($T_{\mathrm{a}}-T_{\mathrm{s}}=\mathrm{2.1}\pm \mathrm{0.5}{}^{\circ }\mathrm{C}$, 1 SD), despite having the most inefficient radiative cooling due to the saturated air (see Fig. 3c). As previously mentioned, this result needs further investigation to be fully understood, but it does not look unrealistic, as former studies have reported longer dew durations, for example 9.6±3.2 h (1 SD) in a coastal steppe over a 4-year period (Uclés et al., 2014). However, measurements of relative humidity in extremely wet conditions are notoriously difficult, and these data could be associated with a sensor failure.

So far, two robust results (dew frequency and dew duration) have been presented and calculated based on the sign of T_d−T_s, without needing to consider atmospheric stability, which is essential for estimating the dew yield (expressed mm night⁻¹). The dew yield estimation for the grasslands is based on the calculation of the aerodynamic resistance to vapor transport, which depends on many variables and is therefore prone to more uncertainties. Figure 5a shows the dew yield versus dew duration relationship using the nightly average data of the grasslands, and only one site (DSNY) failed to produce a realistic dew yield (usually below 1 mm night⁻¹). DSNY is a tropical site with frequent rain and fog events that might largely contribute to its nocturnal water balance and reduce the performance of the MOST, even if the dew duration and frequency of DSNY are consistent with the other tropical site (LAJA). A quadratic curve ($y=a\times x^{\mathrm{2}}$, $a=\mathrm{0.0056}\mathrm{mm}{\mathrm{h}}^{-\mathrm{2}}$) fits remarkably well for the dew yield versus dew duration (r²=0.94, p value <0.01) for the other grassland sites (Fig. 5b). The fit has been calculated on 22 median values (30 min resolution on a dew duration below 11 h), and the mean value of the residuals (nightly data) is $\mathrm{0.02}\pm \mathrm{0.16}\mathrm{mm}{\mathrm{night}}^{-\mathrm{1}}$ (1 SD, 575 nights). A similar quadratic pattern can be observed in the relationship of dew yield versus dew duration for two croplands (Meng and Wen, 2016) using a flux-profile method and the MOST, but the authors did not perform a regression analysis for comparison. The quadratic nature of the relationship can be qualitatively explained by the fact that the vapor pressure deficit (q(T_a)−q^sat(T_s) in Eq. 1) is positively related to the dew duration. This can be seen in Fig. 7b, knowing that $q\left(T_{\mathrm{a}}\right)-q^{\mathrm{sat}}\left(T_{\mathrm{s}}\right)=q^{\mathrm{sat}}\left(T_{\mathrm{d}}\right)-q^{\mathrm{sat}}\left(T_{\mathrm{s}}\right)$. Per definition, the dew yield is the integral of the dew flux over the night or is more visually the area below the dew flux curve. This area is roughly equal to the dew duration multiplied by the mean dew flux, which is driven by q(T_a)−q^sat(T_s) and also depends on the dew duration as explained above. For example, if the dew duration is divided by 2, the mean dew flux will approximatively be divided by 2 as well, and the area will therefore be divided by 4, which explains the quadratic nature of the relationship.

Figure 5Dew yield versus dew duration for the grasslands (top of the canopy, nightly timescale). (a) shows the failure of the Monin–Obukhov Similarity Theory for DSNY, which is why the quadratic fit ($y=a\times x^{\mathrm{2}},a=\mathrm{0.0056}\mathrm{mm}{\mathrm{h}}^{-\mathrm{2}},r^{\mathrm{2}}=\mathrm{0.94}$) has been calculated for the other grasslands based on the 30 min median values for dew durations between 1 and 11 h (b).

Download

Figure 6Median value and [25,75] % quantiles (error bars) of the net radiation above canopy (R_n), and the sensible (H) and latent heat fluxes (LE) averaged during a dew event for each dewy night (indicated in parenthesis). The sign convention is positive when the grassland receives energy, and the residuals are $R_{\mathrm{n}}+\mathrm{LE}+H$, with R_n measured and LE and H calculated with the Monin–Obukhov similarity theory.

Download

The mean dew yield of the dry and temperate grasslands is $\mathrm{0.14}\pm \mathrm{0.12}\mathrm{mm}{\mathrm{night}}^{-\mathrm{1}}$ (1 SD), which is consistent with the 25 studies collecting dew on RDCs ($\mathrm{0.13}\pm \mathrm{0.10}\mathrm{mm}{\mathrm{night}}^{-\mathrm{1}}$, 1 SD; see Tomaszkiewicz et al., 2015), and this provides strong support for the fact that the MOST produces an accurate estimation of the dew yield based on radiometric surface temperatures. The mean dew yield of the tropical site LAJA is 0.52 mm night⁻¹, which is high, but frequent fogs might be included in this budget because of the exceptionally low vapor pressure deficit at LAJA (saturated air and large difference in temperature between the surface and the air). In a study based on weighing methods, Pan et al. (2010) showed a significant change between dew yields forming in foggy conditions ($\sim \mathrm{0.3}\mathrm{mm}{\mathrm{night}}^{-\mathrm{1}}$) and non-foggy conditions ($<\mathrm{0.1}\mathrm{mm}{\mathrm{night}}^{-\mathrm{1}}$) that supports our result. Additionally, the maximum dew yield measured by high precision lysimeters in two temperate grasslands reached 0.5–0.6 mm night⁻¹ (Xiao et al., 2009) for six nights. These events were associated with meteorological conditions that might occur frequently in LAJA.

3.1.3 Estimation of three energy components during dew formation in grasslands

The importance of dew formation in the nocturnal energy budget is poorly documented because of the technical challenge in measuring fluxes from eddy covariance during periods of low turbulence. Opinions differ, from dew being a negligible component of the energy budget that cannot be distinguished from the noise (de Roode et al., 2010) to dew being a process that significantly participates in the energy budget and helps to better constrain the energy closure (Jacobs et al., 2008). In this section, we compare the value of the sensible heat flux (H; W m⁻²), latent heat flux (LE; W m⁻²) and net radiation above the canopy (R_n; W m⁻²) associated with a dew event. See the Method section to understand how H and LE were calculated with the MOST.

Our system is a grassland (biomass above ground, roots excluded), and the energy budget is described as a volume energy balance:

with S as the energy storage within the system (W m⁻²) and G as the ground flux (W m⁻²); here the term of advection has been neglected. The volume energy balance is calculated only during a dew event and the sign convention is for positively considering an incoming flux to the system. In this case, H and LE are always positive because a dew event is associated with the condensation of water vapor on a surface (LE >0), which can only happen when the air is warmer than the surface (H>0). Figure 6 shows the median value (error bars: 25 % and 75 % quantiles) of each energy budget component for each site (number of dewy nights indicated in parenthesis), with the residuals $H+\mathrm{LE}+R_{\mathrm{n}}$ being equivalent to S−G.

The uncertainties associated with the MOST are high at night and sensitive to not only the gradient of temperature between the air and the leaf but also to the wind speed and stability of the atmosphere. For this reason, the discussion here will be only based on the order of magnitude of the components, and DSNY will be excluded from the analysis (see Fig. 5a). In the following, we use the 50 %[25 %,75 %] quantile notation. A typical latent heat flux released by dew formation is $\mathrm{11}[\mathrm{6},\mathrm{20}]{\mathrm{W}\mathrm{m}}^{-\mathrm{2}}$ for the grasslands (excluding LAJA and DCFS). The sensible heat flux for the same eight sites during dew formation is 3 times larger than LE and is associated with a higher variability ($H=\mathrm{32}[\mathrm{18},\mathrm{55}]\mathrm{W}{\mathrm{m}}^{-\mathrm{2}}$), while the net radiation is twice as important as the sensible heat flux ($R_{\mathrm{n}}=-\mathrm{64}[-\mathrm{71},-\mathrm{54}]\mathrm{W}{\mathrm{m}}^{-\mathrm{2}}$). For the tropical site LAJA, the latent heat flux can reach $\mathrm{36}[\mathrm{28},\mathrm{43}]\mathrm{W}{\mathrm{m}}^{-\mathrm{2}}$ and stays above the sensible heat flux ($H=\mathrm{23}[\mathrm{18},\mathrm{29}]\mathrm{W}{\mathrm{m}}^{-\mathrm{2}}$). The Bowen ratio (defined as H∕LE) might therefore decline below 1 during dew events in exceptionally wet sites (nighttime RH>90 %).

As we will see in Sect. 3.3, the leaf temperature is rarely increasing during dew formation, despite the latent heat and sensible heat fluxes bringing energy to the system. The continuous decrease in leaf temperature means that the storage term S is negative during the night. The residuals (S−G) shown in Fig. 6 are positive for three sites: OAES, DCFS and LAJA. This result suggests one of the following: (i) the ground flux G is an important sink of energy (at least $\sim -\mathrm{30}\mathrm{W}{\mathrm{m}}^{-\mathrm{2}}$) for these sites at night and/or (ii) H and LE are overestimated and the MOST fails to partition the energy budget for these sites at night. This study will not be able to validate or invalidate these two hypotheses. However, we can confirm the preliminary work of Jacobs et al. (2008) and affirm that the latent heat flux released by dew formation is a non-negligible component in the nocturnal energy budget, with a typical Bowen ratio of ∼3 during a dew event, similar to the results reported by Meng and Wen (2016) for two croplands using a flux-profile method (69 and 128 dewy nights, Bowen ratio of 1–3 in presence of dew).