NOTE DE CLASSE 21 MAI

additional strategy (besides reducing the wire size) for maximizing droplet shedding. Experimental
tests of superhydrophobic fog harps were outside the scope of this initial study, which focused on
the more practical case of the uncoated, hydrophilic wires. Given that our untreated fog harps do not
exhibit any appreciable clogging, it is unclear whether this hypothetical 8-fold reduction in shedding
volume for superhydrophobic wires would even increase the waterharvesting efficiency (η) by a
significant amount, as the aerodynamics of our harp H1 are already near-optimal. Modeling the
hysteresis of droplets sliding down the mesh structures is significantly more complex due to the
presence of the intersecting horizontal wires45 and the tendency of the water to partially or fully clog
the holes. When observing fog collecting on the surfaces M2 or M3, it was seen that the droplets only
partially clog individual holes in the mesh before shedding. Further, the shedding events themselves
followed a tortuous path along the mesh and often involved chain reactions between water in
adjacent holes. Therefore, any analytical model for M2 or M3 is beyond the scope of this present
work, and we simply refer to the experimental values of Vc,e in Table 2. For the fine-scale mesh M1,
the holes were uniformly clogged and the shedding dynamics of a droplet were more simple. It was
observed that each shedding droplet was much larger than any individual hole size, such that the
droplet’s receding contact line extended horizontally along a single wire (Figure 5c). It follows that the
critical droplet volume scales as ρ gV w ≈ − γθ θ (cos cos ) water c,t r a (7) where w is the width of the
receding contact line along a horizontal wire and had to be measured to predict Vc,t ≈ 6 ± 3 μL. By
measuring both the width (w) and height (h) of five droplets sliding down the clogged mesh, the
experimental critical volume was found by assuming a cuboid shape: Vc,e ≈ hwD ≈ 4.1 ± 1.5 μL.
Looking at Table 2, it is apparent that sliding droplets on the mesh structure do not seem to get any
smaller with decreasing wire size (unlike harps), which causes the clogging issue. Note that the sliding
droplets on harp H1 were smaller than with mesh M1 by a factor of 5. Finally, we demonstrated the
scalability of our fog harp by creating a large, 1 m2 model. This full-scale prototype was made with
common wood 2 × 4s, threaded rods, and the same 254 μm diameter steel wire used for harp H1.
The harp frame was 101.6 cm long with four 91 cm long threaded rods spanning between the
wooden frame (Figure 7). One set of holes drilled into the wood were routed into slots, so one rod
could be moved to tighten the wires if they became loose. This square frame structure enabled the
wires to be wrapped around the threaded rods to create the parallel wire geometry analogous to the
smaller fog harps tested in the laboratory. Threaded rods were used because they provided grooves
into which the wires would remain in place at a consistent pitch. After initial tests of manually wiring
the harp by wrapping the stainless steel wire around the rods in loops, we devised a more automated
system to expedite the wiring process. We calculated Figure 6. Scaling models that predict the critical
departure volume for droplets sliding down (a) single wires (eq 5) and (b) between two wires (eq 6).
Each data series represents a different surface wettability, including aluminum (θr/θa = 31/50°), steel
(θr/θa = 33/51°), and superhydrophobic (θr/θa = 160/165°). Data points represent the experimental
measurements from the three scale-model fog harps. Figure 7. (a) Large-scale fog harp wired on a 1
m2 frame. The frame rotates about a central axle for efficient wire winding, see Movie S3. (b) The
pitch of the wire is determined by the pitch of the threaded rods. ACS Applied Materials & Interfaces
Research Article DOI: 10.1021/acsami.7b17488 ACS Appl. Mater. Interfaces XXXX, XXX, XXX−XX

Monthly and daily dew amounts are in line with or higher than observations in the Israeli desert
(Heusinkveld, 2008; Jacobs et al., 2000; Zangvil, 1996), and daily maximum values correspond to
those observed in some parts of India (0.45 mm, Raman et al., 1973), Morocco (0.50 mm, Lekouch et
al., 2012) and Croatia (0.60 mm, Muselli et al., 2009). Maximum dew amounts that can be physically
expected are in the order of 0.8 mm per night (Muselli et al., 2009; Sharan et al., 2007), and dew
amounts observed in crop canopies have been found to reach 0.8 mm (Kabela et al., 2009). Recorded
dew represented 90 and 41% of total rainfall during the 2013 dry season observation period in the
littoral and inland site, respectively, while total dew amounts reached 58.4 mm corresponding to
19% of total rainfall over the whole observation period at the littoral site. Ecologically, dew duration
may play a more important role for plants than actual dew amounts (Munne-Bosch et al., 1999; 
Sharma, 1976). We found that, across 140 observation nights during the dry season, dewfall started
soon after sunset, continued until 7 am and all dew had evaporated before 10 am (Fig. 3). Negative
values after 10 am may be attributed to the intense heating of the metal collection plate by the sun
leading to distorted weight measurements. Average dew duration of 15 h is much higher than
observed by Zangvil (1996) in Israel (5e10 h night1 ) but comparable to estimated and observed
values by Pedro and Gillespie (1981), Malek et al. (1999) and Kabela et al. (2009) in crop fields at
northern American sites. It has been suggested elsewhere that dew may play a crucial role in wild and
crop plants' water balance, particularly during periods of drought (Goldsmith et al., 2013; Malek et
al., 1999; Munne-Bosch  and Alegre, 1999; Went and Babu, 1978). While dew is unlikely to affect soil
moisture reserves given its quick evaporation after sunrise, it may reduce transpiration during the
night and thereby conserve soil water (Richards, 2004; Ucles et al., 2013  ), as well as enable higher
photosynthesis rates in early morning hours due to increased plant water status from dew absorption
through the leaves (Grammatikopoulos and Manetas, 1994; Munne-Bosch and  Alegre, 1999;
Munne-Bosch et al., 1999  ). A study in Israel showed that annual variability of dewfall was lower
than of rainfall suggesting that even small amounts of dew may be a relatively reliable water source
for micro-organisms such as lichen (Zangvil, 1996). 3.2. Relation between dew measurements and
weather data At Efoetsy nights with dewfall were characterized by significantly lower mean
temperatures, higher relative humidity, higher dew point temperature, lower mean differences
between air and dew point temperature (Tair  Tdp) and lower wind speed than nights without
dewfall (Table 2). These results were similar at Andremba (data not shown). In nights with highest
dew deposition, mean T corresponded to mean dew point T and mean relative humidity reached
100% (Fig. 4). The high relative humidity even during nights without detected dewfall, which
continued to increase through the night, may be due to the generally high supply of moisture from
the sea, even under low wind speed conditions. Maximum wind speed during dew nights reached 5.7
m s1 , which is the same upper limit for dew formation found by Lekouch et al. (2012) in Morocco,
and similar to values obtained by Nilsson (1996) and Muselli et al. (2009). The upper limit of wind
speed in our observations can be expected to be overestimated as the dew collection surface was
slightly protected from wind (Clus et al., 2008; Lekouch et al., 2012). In regression analysis the best
fitting linear model for daily accumulated dew was obtained with mean wind speed, maximum Tair 
Tdp values and air temperature drop (Tmax  Tmin) as predictors (Table 3). While the model only
explains 46% of observed dew data variation, it corroborates findings that Tair  Tdp, which is strongly
correlated with relative humidity, and wind speed are the main parameters influencing dew yield
(Lekouch et al., 2012; Luo and Goudriaan, 2004; Muselli et al., 2009; Sharan et al., 2007). In our study
dewfall was most strongly correlated with wind speed (r ¼ 0.576, p < 0.001). Cloud cover data,
which is often included when modeling dewfall as it indicates night time cooling (Muselli et al., 2009),
was not available in our study, but the temperature drop during the night (Tmax  Tmin) represents
the same phenomenon, and slightly improved the model. Weather data was not able to predict dew
rates at 30 min time intervals as R2 values only reached 0.082, while at 1 h time steps R2 only
improved to 0.131. Uncertainties can be attributed to the fact Fig. 2. Monthly dew amounts during
the observation period at the sites Efoetsy (coastal) and Andremba (inland), and monthly rainfall at
Efoetsy above bars in the study area of south-western Madagascar. S. Hanisch et al. / Journal of Arid
Environments 121 (2015) 24e31 27 that dewfall is determined by the temperature of the
condensation surface for which air temperature used in our regression is just an approximation, as
well as wetting conditions and heat transfer parameters of the material (Beysens et al., 2005; Clus et
al., 2008; Pedro and Gillespie, 1981). These factors play a more important role in smaller time steps,
hence regression with weather data alone cannot sufficiently explain observed dew rates. 3.3.
Balance evaluation and consequences for natural dewfall Correlation coefficients between total dew
yield during the night from the different measuring methods were 0.71 (balance e plastic sheet on
the ground), 0.84 (balance e 1 m high plastic sheet on wooden surface), 0.82 (balance e 1 m high
plastic sheet on metal surface) and 0.99 (1 m high plastic sheet on wooden surface e 1 m high plastic
sheet on metal surface; Fig. 5). Average daily dew values in the evaluation period amounted to 0.064
mm (plastic sheet on ground), 0.136 mm (plastic sheet at 1 m of height on wood), 0.123 mm (plastic
sheet at 1 m of height on metal) and 0.184 mm (dew balance; Table 4). Recorded dew amounts were
significantly different between methods (P < 0.001) except between the plastic sheet on wood and on
metal at 1 m height (P ¼ 0.878). Hence, recorded dew values on the ground tended to be lower than
the ones obtained at 1 m height with the plastic sheet method which reflects more efficient
radiational cooling given enhanced heat flux from the ground to the sheet surface (Fig. 5). Differences
in dew deposition with height have also been reported in other studies that found dew precipitation
to be increasing with height up to 100 cm above the soil (Jacobs et al., 1994; Raman et al., 1973;
Subramaniam and Rao, 1983). On the other hand, Ninari and Berliner (2002) showed that water
vapor adsorption adds to moisture in the soil profile as measured with micro-lysimeters. Water vapor
adsorption is a different source of atmospheric water, but unlike dew not detectable on artificial
condensation surfaces. Since water vapor adsorption may play a role in our study area, balance
values may sometimes be lower than actual moisture deposition on the soil (Agam and Berliner,
2006; Ninari and Berliner, 2002; Ucles et al., 2013  ). The specific shortcomings of the Hiltner
balance, using a much smaller metal surface connected to an analog balance and a writing device for
assessing dewfall, are similar to those of our method (Zhang et al., 2011). Our evaluation results
suggest that measurements with the balance are Fig. 3. Dew accumulation and evaporation during
the night and morning hours (n ¼ 140) in the study area of south-western Madagascar. Error bars
represent 95% confidence intervals. Reference lines indicate that mean dew formation continues until
7:00 am and dew is evaporated by 10:00 am. Table 2 Weather data statistics of nights with and
without dewfall and significance-values of ManneWhitney-tests in the study area of south-western
Madagascar. Dew nights (n ¼ 343, wind speed n ¼ 192) Nights without dew (n ¼ 56, wind speed n ¼
32) Significance Mean T (C) Min 12.9 16.3 0.001 Max 27.4 26.4 Mean 19.4 20.7 Mean RH (%) Min
55.5 56.8