Delineation of Dew Formation Zones in Iran Using Long-Term Model Simulations and Cluster Analysis

. Dew is a non-conventional source of water that has been gaining interest over the last two decades, especially in arid and semi-arid regions. In this study, we performed a long-term (1979–2018) energy balance model simulation to estimate 20 dew formation potential in Iran aiming to identify dew formation zones and to investigate the impacts of long-term variation in meteorological parameters on dew formation. The annual average of dew occurrence in Iran was ~ 102 days, with the lowest number of dewy days in summer (~ 7 days) and highest in winter (~ 45 days). The average daily dew yield was in the range of 0.03 – 0.14 L/m 2


Introduction
Scarcity and continuously increasing demand on freshwater is one of the socio-economic problems in many countries, especially in arid and semi-arid regions. It is anticipated that two-thirds of the world's population will suffer of freshwater shortage by the year 2025 (Human Development Report, 2006). In fact, the water crisis will not only be limited to freshwater resources but also will have an extreme impact on agriculture and livestock (Madani, 2005). 40 Scientists have also warned that water shortage will continue further in the coming decades in the Middle East, where water is one of the most valuable and vulnerable natural resources (Mehryar et al. 2015;Ashraf et al. 2019., Bozorg-Haddad et al. 2020). Iran is one of these countries suffering of freshwater shortage and climate change consequences (Karimi et al., 2018., Ashraf and Fahimi, 2019;Emami and Koch, 2019;Naderi, 2020). For instance, the annual average rainfall in Iran is about 250 mm (Alizadeh, 2011). Besides that, 65% of the country is arid, 20% is semi-arid, and only 15 % has a humid and semi-humid 45 climate. The Iranian Annual Renewable Water Resources is currently less than 2000 m 3 /capita and with the current population growth rate (~1.19 %;CIA, 2018), is expected to be reduced to be less than 1000 m 3 /capita in 2025 (Madani, 2005;Moradi, 2017). Therefore, looking for alternative resources of freshwater is a necessity in the arid and semi-arid regions in Iran.
The atmosphere can be considered a huge renewable reservoir of water (i.e., cloud, fog, and water vapor) and enough to meet the needs of every person on the planet (Tu et al., 2018). Dew is a non-conventional atmospheric resource of water, which 50 forms during phase transition from vapor to liquid (Tomaszkiewicz et al., 2015), or condensation of atmospheric water vapor on surfaces with temperature below dew point (Khalil et al., 2016). Although the amount of dew that can be harvested is relatively small, it can enhance water supply in certain climates/regions, particularly in the absence of precipitation (Tomaszkiewicz et al., 2015). Extracting dew water as a sustainable natural phenomenon by means of radiative (or passive) condensers has been gaining interest over the last two decades. Research on radiative condensers started in the early 1960s 55 with a study conducted in Negev Desert by Gindel (1965). Based on studies in different locations worldwide (Table 1), the highest amount of daily dew yield (typically in the range of 0.2-0.6 L/m 2 was observed in arid deserts and semi-arid areas (Kidron, 1999;Alnaser, 2000;Agam and Berliner, 2006;Lekouch et al., 2012;Tomaszkiewicz et al., 2017;Jia et al., 2019;Tuurre et al., 2019). Some regions with humid climates (e.g., coastal areas and Islands) showed lower yield (~ 0.2-0.4 L/m 2 ) (Sharan, 2005;Clus et al., 2008;Museli et al., 2002 andHanisch et al., 2015), and urban environments 60 with measured dew yield and could be also applicable elsewhere. For example, Tomaszkiewicz et al. (2016), applied a dew prediction model that was developed by Beysens (2016), to generate a dew yield atlas for the Mediterranean region (142 70 stations). The objective of this study is to identify the major dew formation zones in Iran using a long-term model simulation and to investigate the possible impacts of historic changes to the climate over the last 40 years on dew in Iran.

2.Methods
In order to estimate dew collection potential in Iran, we combined a computationally efficient dew formation model with 75 meteorological reanalysis data spanning 40 years. The model simulation results were used to investigate the spatial-temporal variation of dew yield in Iran. In this study, the term "dew yield " refers to the amount of water that can be harvested on a 1m 2 condenser.

Meteorological input data
The dew formation model (which is described in detail in section 2.2) requires meteorological data as input. In Iran, there are 80 very few stations with long-term observations of all the required meteorological variables. Therefore, instead of driving the dew model with observations, we use ERA-Interim (Berrisford et al., 2011;Dee et al., 2011), which is a meteorological global reanalysis produced by the European Centre for Medium-Range Weather Forecasts (ECMWF). Reanalysis combines a massive number of observations from a number of sources (satellite, radiosondes, aircraft, buoy data, stations, etc.) with a numerical weather prediction model to produce a coherent, long-term gridded data set of the atmospheric dynamic and thermodynamic 85 state over the whole globe (Tompkins, 2017).
ERA-Interim covers the time period from January 1979 until August 2019, has a native resolution of 0.75 degrees, which is approximately 80 km, and 60 model levels in the vertical profile. Here we considered the time period during 1979-2018 and used input data interpolated to a grid resolution of 0.25 degrees (~30km) over a domain covering all parts of Iran ( Figure 1). This interpolation was done during the download process using standard ECMWF procedures: continuous fields (e.g. 90 temperature, precipitation) were interpolated using bilinear interpolation and discrete fields (e.g. vegetation, soil type) were interpolated using a nearest neighbor approach.
Similar to all atmospheric reanalysis, ERA-Interim contains two distinct types of fields: analysis fields and forecast fields. The analysis fields were produced by combining a very short-range forecast and observations to produce the best fit for both. The forecast fields were produced by the numerical forecast model starting from an analysis. In ERA-Interim, the analysis fields 95 were available every 6 hours (00:00, 06:00, 12:00, and 18:00 UTC) and the forecast fields were available every 3 hours and hence can be used to fill in the gaps between the analysis. Furthermore, the forecast fields can be either instantaneous or accumulated over the forecast period.
The variables that are required for the dew formation model are: air temperature (Ta), dew point temperature (DP), wind speed (WS), short-wave (Rsw), and long-wave solar radiation (Rlw). From ERA-Interim we extracted the 2-m Ta and DP from both the 100 analysis and the instantaneous forecasts and obtain the short-wave and long-wave surface radiation as accumulated forecast fields. To obtain the mean value over each time interval, the difference of the accumulated values between two consecutive time steps was taken and then divided by the time difference in seconds. The wind speed at 2m was not directly available from ERA-Interim; therefore, we obtained the wind components (U and V) at 10 m and the surface roughness (z0an instantaneous forecast field) and assumed a logarithmic winds profile to obtain its values at 2-m according to 105 where z0 is the surface roughness and U10 and V10 are the horizontal wind speed components at 10 meters. It is important to understand that the logarithmic assumption is only strictly valid during neutral stability conditions. During stable conditions (such as during night time) it overestimates the 2-m wind speed whereas in unstable conditions it underestimates the 2-m wind speed (Riou, 1984;Holtslag, 1984;Petersen et al., 1998;Oke, 2002;Optis et al., 2016). 110

Dew formation model description and output
The global dew formation model used in this study was originally developed by Vuollekoski et al. (2015) to estimate dew potential. The approach is similar to Pedro and Gillespie (1982) and Nikolayev et al. (1996). The model reads all input data (described in section 2.1) for a given grid point and numerically solves the mass and heat balance equation by using a fourthorder Runge-Kutta algorithm with a 10s time step (i.e., The ERA-Interim data from 3 hourly resolution were linearly 115 interpolated to obtain 10-second resolution). The mass and heat energy balance model is written as: where dTc/dt is the rate of change of the condenser temperature. Cc, Cw, and Ci are the specific heat capacity of the condenser, water, and ice; respectively. Here, mc, mw, and mi are the mass of the condenser, water, and ice; respectively. The right-hand side of Eq.
(2) describes the heat exchange involved in the process: Prad is the net radiation, Pcond is the conductive heat 120 exchange between the condenser surface and the ground, Pconv is the convective heat exchange, and Plat is the latent heat released by the condensation or desublimation of water.
The model was setup so that it assumes similar conditions for the phase-change of pre-existing water or ice on the condenser sheet. For instance, if the water on the condenser is in the liquid phase (i.e., mw > 0) and the condenser temperature Tc < 0 ℃, then the sheet is losing energy (i.e., the right-hand side of equation (2) is negative). In that case, instead of solving Eq. (2), Tc 125 is assumed to be constant and the lost mass from the liquid phase of water is transferred to the cumulated mass of ice; i.e., the water is transformed from liquid phase to solid phase. Consequently, equation (2) is replaced by where Lwi is the latent heat of fusion. If the water on the condenser is in the solid phase (i.e., mi > 0) and the condenser temperature Tc > 0 ℃, a similar equation is assumed for the change rate of ice mass (mi). 130 Note that Eq. (3) is not related to the condensation of water; it only describes the phase change of the already condensed water or ice on the condenser. For the water condensation rate, which is assumed independent of Eq. (3), the mass-balance equation is then assumed as where m represents either the mass of ice (mi) or water (mw) depending on whether Tc is below or above 0 ℃. Psat(Td) is the 135 saturation pressure at the dew point temperature and Pc(Tc) is the vapor pressure over the condenser sheet. Here, Sc is the condenser surface area and k is the mass transfer coefficient, where Lvw is the specific latent heat of water vaporization,  is the psychrometric constant, Ca is the specific heat capacity of air, and p is the atmospheric air pressure. Here, h is the heat transfer coefficient, 140 h = 5.9 + 4.1 u (511 + 294) / (511 + Ta) where u and Ta are the prevailing horizontal wind speed and the ambient temperature 2 meters above the ground. This parameterization of the heat transfer coefficient is taken from Richards, (2009). However, the dew model is designed in such a manner that any functional form can be used for the heat transfer coefficient thus allowing the sensitivity of the modelled dew amounts to the formulation of the heat transfer coefficient to be assessed (see section 3.3 for such an analysis). 145 In practice, the wettability of the surface affects the vapor pressure Pc directly above it. In other words, Pc is lower over a wet surface; and thus, condensation may take place even if Tc > Td. It is also assumed (in equation 4) that there is no evaporation or sublimation during daytime even if Tc > Ta. Furthermore, the model simulation resets the cumulative values for water and ice condensation at noon (local time) and takes the preceding maximum value of mw + mi as the representative daily yield given in millimeter on a 1 m 2 condenser sheet (i.e., mm/ m 2 /day equals to L/ m 2 /day). 150 This way, the model simulation replicates the daily manual dew water collection of the condensed water around sunrise; i.e., after which Tc is often above the dew point temperature. All terms and nomenclature are described in more detail in Table 2 and Table 3.
It should be noted here that, similar to many numerical models, this model has some limitations that should be considered when interpreting the results. For instance, both heat and mass coefficients are semi-empirical parameters that depend on wind 155 speed (i.e., here we used the parameterization by Richards (2009), valid for u < 5 m/s). In addition, the 10s-time step in the model does not allow condensed water droplets to be eliminated on the condenser surface by evaporation. Moreover, the model predicts any dew condensation, regardless if it is collectible or not; therefore, it is expected to overestimate dew yield. The spatial data resolution is ~ 30 km, which limits the model's ability to resolve local microclimates, particularly in areas with complex topography where the topography can modify the large-scale winds and lead to large variations in local temperatures. 160 However, when considering cumulative dew yield over long time periods the model performs well. Therefore, as the model uses the meteorological gridded dataset (ERA-Interim), which is readily available for the whole globe, it can be applied anywhere in the world including other arid and semi-arid areas even if they lack observations.

Cluster analysis
Cluster Analysis (CA) is an effective statistical tool and technique that groups similar data points such that the points in the 165 same group are more similar to each other than the points in the other groups. The group of similar data points is called a Cluster which can be used for various applications (Corporal-Lodangco and Leslie, 2016;Gungor and Ozmen, 2017). There are two main clustering methods: hierarchical and non-hierarchical cluster analysis. Hierarchical clustering (used in this study) combines cases into homogeneous clusters where objects at one level are combined with objects at another level and produce clusters that are not allowed to overlap (Bunkers and Miller, 1996;Yim and Ramdeen, 2015). Two different strategies for 170 hierarchical clustering exist: Agglomerative and divisive (Lior and Maimon, 2005). In this study, we used hierarchical agglomerative clustering (HAC, Nielsen, 2016) which starts with N clusters (i.e., here is the total number of grid points), each containing one object, and join those two objects that are most "similar". This process continues until only one cluster, containing all the data, remains (Bunkers and Miller, 1996). In order to decide which clusters should be combined (for agglomerative), a measure of dissimilarity between sets of observations is required. The similarity measurement is a critical 175 step in hierarchical clustering as it can influence the shape of the clusters (Nielsen, 2016). With metric data, the most commonly used distance measure (a measure of the distance between pairs of observations) is "Euclidean distance". The Euclidean distance (dij) between two objects i and j in a two-dimensional data matrix is simply the squared difference between two observations for each of p variables, summed over the variables and k is the number of observations (Fovell and Fovell, 1993;Dokmanic et al., 2015). This can be written as: 180 Here we applied this method to a two-dimensional matrix (2496×14610), where the number of rows represented the number of spatial grid points in the model simulation domain and the number of columns represented the time (i.e., cumulative daily dew yield).
After all distances were calculated, the next step is to merge the two closest entries to form a new cluster based on a linkage 185 criterion. The linkage criterion determines the distance between sets of observations (here is the spatial grid points) as a function of the pairwise distances between observations. There are some commonly used linkage criteria: single linkage, complete linkage, average distance, and Ward's minimum variance methods, which differ in a way how the distances between entries are calculated and how the two closest entries are defined (Stooksbury & Michaels, 1991;Murtagh and Legendre, 2014). In this study, Ward's minimum variance method (Ward, 1963) is used. This method is the most frequently clustering 190 technique used in climate research (Yokoi et al., 2011;Mimmack et al., 2000;Siraj-Ud-Doulah and Islam, 2019) and gives the most consistent clusters (Kalkstein et al. 1987). It calculates the means of all variables (the amount of dew) within each cluster, then calculates the Euclidean distance to the cluster mean of each case, and finally sums across all grid points (Unal et al., 2003).
In any CA, the optimal number of clusters is an important issue. There is no reliable and universally accepted method to 195 determine the optimal number of clusters. Kaufmann and Weber (1996) (see also Unal et al., 2003 andBurlando, 2009) suggested showing the total variance of subsequent merged clusters as a function of the number of remaining clusters. This information can be used as an indicator to decide the number of clusters but still, a visual check of the result can help to make the right decision. The suitable number of clusters has to be chosen somewhere in the transition between the distance values when a sudden decrease is observed as illustrated in Figure 2a. In our case, few steps at N=3, 4, 6, 7, and 10 are recommended 200 as optimal numbers of clusters. By visualizing all these steps, N=6 was found to be the best number of clusters for this study because fewer clusters (i.e. 3 and 4 clusters) were not able to capture the different climate and dew zones. Furthermore, choosing more clusters (i.e. 7 and 10 clusters) gives some groups that replicate each other. The results of hierarchical clustering are usually presented in a dendrogram (Nielsen, 2016). The dendrogram of our 6 clusters has shown in Figure 2b.

Spatial-temporal variation of dew occurrence and yield
According to the model simulation results (cumulative daily dew yield in the form of dew and hoarfrost), dew formation occurred almost everywhere in Iran as illustrated in Figure 3, which shows the seasonal occurrence of dew as a fraction of days with any dew yield. The frequency of dew occurrence was more than 80% (~75 days) in most areas of Iran in wintertime (December-February, Fig. 3a). The mean occurrence of dew was rather similar during spring (March-May, ~ 50 days, Figure  210 4b) and autumn (September-November, ~40 days, Fig. 3d) with the highest number of dew days (more than 90% (~80 days)) in the mountainous and coastal areas and the lowest (less than 40% (~35 days)) mostly in the dry interior and eastern areas.
The lowest frequency of dew occurrence (i.e., less than 10 days) was in summer (June-August, Fig. 3c) when dew formation was limited to a narrow part along the Caspian Sea and the northern domains of Alborz mountain.
Limiting the dew occurrence analysis to days with dew yield > 0.1 L/ m 2 day also confirmed the seasonal characteristics of the 215 temporal-spatial occurrence of dew. However, in this case, the frequency of dew occurrence days was less (in the range of 6-45 days for summer and winter, respectively (Fig. S1), and the spatial scale of dew formation shrank to include only a few parts of the coastal and high mountain regions during spring, summer, and autumn. This notable difference between the two maps (i.e., Fig. 3 and Fig. S1) is associated with the model setup. The model tends to forecast any dew event, regardless if it can be collectible or not. In practice, very small dew quantities are generally not harvestable as droplets remain pinned to the 220 condenser surface and gravity cannot lead them to the collection tank.
We subsequently calculated the seasonal daily means of the cumulative dew yield (Fig. 4) which show a clear seasonal cycle with high dew yields during the winter and low yields during the summer in most parts of Iran. The monthly means of the cumulative dew yield are shown in Figure S2. Both seasonal and monthly maps show that the mountain regions had dew occurrence throughout the year with mean cumulative daily dew in the range 0.11-0.18 L/m 2 /day. In winter, dew occurred 225 almost everywhere in Iran with the highest yields in the southern part of the Persian Gulf and Oman Sea coastline (mean cumulative daily dew in the range 0.15-0.23 L/m 2 /day). In spring (i.e., April, May), a spatial pattern was observed which indicated the formation of dew was mainly parallel to the mountain range (Alborz (East-West) and Zagros (north-west and south-east)). The reason could be related to the temperature, which increases, and relative humidity which decreases during these spring months. Therefore, in spring, in most areas conditions for dew formation were not present except in high elevation 230 areas where the condition still favor dew formation. During summer and until the middle of autumn (i.e., July-Oct) a unique spatial pattern was evident which shows the distribution of dew formation was only limited to a narrow belt in coastal areas in the north along the Caspian Sea. In all other areas, the monthly amount of dew yield was almost zero.

Dew zonesa general overview 235
According to our Cluster Analysis (CA) summarized in Section 2.2, we identified 6 dew formation zones in Iran (Fig. 5). The amount of daily dew yield in Iran and related climatological parameters (e.g., temperature, relative humidity) for dew formation as well as the percentiles (i.e. 25%, median, 75% and 99%) of daily dew yields as averages for each cluster are listed in Table   4. As will be shown in this section, the dew formation zones in Iran are clearly aligned with topography, sources of moisture, and climate zones. Furthermore, the mountains and seas played major roles in the spatial distribution of dew formation zones. 240 Note that the maximum daily dew yield in this section is presented as the 99th percentile of daily dew. In order to have an insight into the climatological condition in each dew zone, we selected one synoptic station in each dew zone and investigated some related meteorological parameters (e.g., temperature, humidity, wind speed, …) in nighttime hourse (i.e. 18:00, 21:00, 00:00, 03:00), when dew formation occurs, for the time period 1980-2010 (30 years) which is shown in Figure 7 and Figure 8.

Dew zone A -Caspian Sea region 245
We identified the first dew formation zone as the "Caspian Sea region", which covered the southern shores of the Caspian Sea and the northern domains of the Alborz mountain range This dew zone includes about 7% of the total land area of Iran ( Fig.   5), which also includes the largest forest area in Iran. The overall mean daily dew yield in this region was ~0.14 L/m 2 which was the highest among all of the dew zones and the maximum dew yield was 0.30 L/m 2 /day (Table 4). Interestingly, this dew zone is different compared to the other dew zones concerning the annual cycle of dew formation; in this dew zone, dew 250 formation occurred throughout the year whereas all other zones exhibit a strong annual cycle (Fig. 6). The mean frequency of dew occurrence in this zone was more than 330 days/year. Even in summer, when dew almost vanished in other dew zones, this zone had a significant amount of dew yield (Fig. 4). The mean yearly dew yield in this region is estimated at about 53 L/m 2 and the maximum yield is more than 100 L/m 2 . The high potential of dew formation in this zone during the year is due to very suitable climatological and geographical conditions. The synoptic station Ramsar, located in this dew zone, shows the 255 climate of dew zone A to be characterised by, low temperature, high humidity, and the smallest dewpoint depression (i.e., the smallest difference between the temperature and dew point) along with little variation in the relative humidity and dewpoint depression throughout the year (Fig 7a). Moreover, due to being a forest area, the wind speed is relatively low (Fig, 8a), which favors dew condensation 260

Dew zone B -Zagros Mounain region
Dew zone B included the Zagros mountain region (i.e., northern and central parts) and the eastern part of the Alborz mountains.
This dew zone covered about 15% of Iran (Fig. 5) and represented a mountain climate with very cold and dry weather in winter and mild weather in summer ( Fig. 7b; Zanjan station). Furthermore, due to the high elevation, the diurnal variation of temperature within this dew zone is large. These areas receive high levels of solar radiation during the daytime and reflect it 265 back quickly to space in the form of long-wave radiation during night-time. Therefore, the temperature drops rapidly during night-time. Enough moisture in the atmosphere, in addition to this strong nocturnal cooling, favored dew formation. The overall mean daily dew yield and variation in this region was 0.08±0.05 L/m 2 /day and the highest dew yield was 0.23 L on a 1m 2 condenser sheet. The highest amount of dew yield in this dew zone was observed during spring when typically, the prevailing winds in this region are westerlies and which are accompanied by moderate to high relative humidity (Fig. 7b) and low wind 270 speed (Fig 8b). The amount of dew yield decreased rapidly after May and was almost absent during summertime (Fig. 6). This is a result of higher temperature (i.e., due to atmosphere transparency and receiving high solar radiation) and lower relative humidity (Fig. 7b) and also the lack of efficient moisture sources in this dew zone In general, the mean frequency of dew occurrence in this zone was 63% (~245 days). The mean yearly dew yield in this zone was about 30 L/m 2 and the maximum was more than 70 L/m 2 (Table 4). 275

Dew zone C -Central Iran
The third dew zone is the Central Iran region. Central Iran consists of the southern slopes of the Alborz Mountains in the north, the Zagros Mountains in the south and the central Iranian ranges. These areas are mostly hot and very dry. Alborz and Zagros mountains prevent moisture penetration from the Caspian Sea and westerlies so that the amount of water vapor pressure is 280 very low (~7 hPa, Masudian, 2011). This zone covered about 20% of Iran and included the Kavir desert basin, Salt Lake, and some parts in the north-east (Fig. 5). The overall mean daily dew yield in this region is estimated to be about 0.05 L/m 2 and the maximum yield (99 th percentile) was about 0.21 L/m 2 /day. The average yearly dew yield in this region was about 18 L/m 2 and the maximum yield was less than 50 L/m 2 /year. The dew period in this zone starts in autumn and continues until midspring (i.e., October-April), however, the frequency of dew occurrence (> 0.1 L/m 2 /day) is about 80 days. Isfahan station is 285 located in this dew zone and is representative of the climate of this dew zone. The dew point temperatures are very low (mainly around or less than zero) all year round and have little annual cycle (Fig 7c). The relative humidity is low in spring but increases in Autumn (Fig. 7c) when temperatures start to decreases and dew formation also starts. Furthermore, the wind speed is quite weak (less than 5 m/sec, 2.5 m/sec in average, Fig. 8c) and does not have a pronounced annual cycle. Therefore, most likely humidity and temperature are the key factor in formation of dew in this station/zone. More specificly, once relative humidity 290 start increasing and temperature decreases (in autumn and winter), dew can also form.

Dew zone D -Lut desert
We identified the fourth dew zone (i.e., Dew zone D) that included the Lut desert (175,000 km 2 ; Alizadeh et al. 2014), which 295 is an arid and hyper-arid desert (Fig. 5). This zone, with 35% of all grid points in the land areas of Iran, is the largest dew zone; however, it has the least dew occurrence (~15 days per year with dew yield > 0.1 L/m 2 /day) and a mean yield of 0.03 L/m 2 /day. Indeed, this part of the country includes the driest areas (i.e., water vapor pressure is < 5 hPa. Based on a survey conducted by scientists at NASA's Earth Observatory during the summer of 2003-2009 (see: Temperature of Earth: https://www.universetoday.com/14367/planet-earth), the Lut Desert was the hottest (~71 ℃) land surface on Earth, see also 300 Khandan et al. (2018). The synoptic station Tabas is located in this dew zone and has a climate characterized by high temperatures (higher than the synoptic stations we considered in dew zones A, B and C) and low relative humidity in summer ( Fig. 7d) In addition to the dryness, these areas have high diurnal variations in temperature, mostly clear sky, extremely sparse vegetation, and frequent high wind speed. In wintertime, the temperature decreases and the moisture increases (Fig. 7d), as a result of the westerly prevailing wind and thus, this dew zone experienced its highest amount of dew yield in winter. In contrast, 305 in the warm season (i.e., May-September) dew was almost completely absent (Fig. 6). The reason is due to high temperature, longer day time duration, and a strong north-south pressure gradient between the thermal low-pressure system over the desert lands and a cold high-pressure over the Hindu Kush mountains in northern Afghanistan (Alizadeh et al., 2014) that generates the strong summer wind called "the Sistan wind of 120 days". It was called so since it occurs during late May through late September (about 4 months) in the east and southeast of the Iranian's Plateau, particularly the Sistan Basin. The typical wind 310 speed of the Sistan is 30-40 km/h, but it could occasionally exceed 100-110 km/h, which impedes dew formation during the summer season. Thus, the key factors for dew condensation (high humidity, low wind speeds) are not present for most of the year in this dew zone. Consequently, the average yearly dew yield in this zone was low -about 12 L/m 2 and the maximum yield was about 40 L/m 2 /year.

Dew zone E -Persian Gulf region
The Persian Gulf dew zone included the coastal line of the Persian Gulf and some parts of the western half of the land areas in Iran (~9% of all grid points; Fig. 5). The overall mean daily dew yield is about 0.06 L/m 2 , which is lower than the other coastal zones in the north (i.e., Caspian Sea; dew zone A) and south of Iran (i.e., Oman Sea; dew zone F). However, the maximum daily dew yield in winter (i.e., December-February) was higher than that in the Caspian Sea zone. Indeed, this dew zone 320 benefits from two huge sources of moisture (i.e., Persian Gulf and Karon river), although high temperatures (e.g. as observed at the synoptic station Ahvaz , Fig 7e) , thermal high pressure, and dry winds, especially during the warm season (i.e. May-September), do not favor the formation of dew. Therefore, , the period of dew formation was about 7 months starting in October and ending in April. However, the frequency of dew occurrence > 0.1 L/m 2 /day is about 117 days during November-February ( Fig. 6), when relative humidity is at its highest level and temperature and wind speed are relatively low compare to the rest 325 of the year (Fig. 7e and Fig. 8e). The average yearly dew yield in this zone was about 24 L/m 2 and the maximum was > 70 L/m 2 /year (Table 4).

Dew zone F-Oman Sea region
The coastline along the Oman Sea and the strait of Hormuz formed the sixth dew zone, which is also the smallest dew zone in 330 Iran covering only 5% of the grid points (Fig. 5). The overall mean daily dew yield in this zone was about 0.09 L/m 2 and the maximum dew yield was about 0.23 L/m 2 /day (Table 4), which was the highest among all dew zones. This is not surprising because this region benefits from a generous source of moisture (i.e., Oman Sea, Persian Gulf, Arabian Sea, Indian Ocean through the summer monsoon). Observations from the station Bandarabas confirm the presence of a moisture source as the difference between the temperature and dew point temperature is quite small (about 5℃) and constant throughout the year. 335 However, despite these conditions, the formation of dew was mostly limited to the cold season (i.e., starting in September and ending by March, Fig. 6) and during the warm season (i.e., April-August), dew occurrence was rare (Fig. 6). The reason is likely due to the increase in wind speed (as shown by obseravations at the synoptic station Bandarabas, Fig. 8e) and the pemperature during summer. In particular, in the warm season, high temperatures leads to the formation of low-pressure systems (i.e., Gang and Persian Gulf) over the seas, which intensified the hot and humid conditions in the southern coastal 340 region. High humidity results in amplified long-wave radiation downwards, and therefore less radiative cooling. In addition, due to the strong gradient between the low pressure over the Persian Gulf and the high pressure over Saudi Arabia, an intense airflow is stimulated, so that condensation does not occur despite high humidity. Lastly, although this zone had the highest daily dew yield, it does not have the highest yearly yield (i.e., > 80 L/m 2 ) since the frequency of dewy days (~ 150 days) in this zone is lower than in dew zone A (the Caspian Sea region, Fig. 7  345

Long-term temporal variation in dew formation zones
In order to investigate the long-term (1979-2018) variation of dew formation, we applied the Mann-Kendal trend test (Pohlert, 2016) to the yearly means of dew yield with a confidence level of 95%. Figure 9 shows the statistical significance (p < 0.05) of the overall changes in the mean yearly dew yield. The result of this trend analysis showed that in more than 60% of the land areas in Iran (i.e., mostly dew zones C and F and the northern half of dew zones B and D), dew formation has decreased during 350 the past 40 years. The remaining parts of Iran did not show any significant trend (α = ) however, their negative slope (82% of the remained grid points) might be a sign of a future decrease in dew formation for these regions. Such negative trends in dew yield over a wide geographical region could be due to different reasons that control the condensation process. To identify potential causes for the detected decrease in dew formation, we first calculate correlations between the dew formation and meteorological parameters (temperature, dewpoint temperature, dewpoint depression, relative humidity, wind speed, and cloud 355 cover, obtained from ERA-Interim) for each dew formation zone (Table 5). Subsequently, we calculate the trend for each of the 6 meteorological variables.
The correlation analysis (i.e., Pearson's correlation) revealed that dew formation in almost all dew zones (i.e., B-F) has a very strong negative correlation (values of -0.93 to -0.95) with temperature, a strong positive correlation with relative humidity (values of 0.88 to 0.98), and a negative correlation with the dewpoint depression (-0.69 to -0.88). In contrast, dew zone A (the 360 Caspian Sea) has weak correlations between dew formation and temperature, relative humidity, and dewpoint temperature indicating dew formation in this region is controlled by different processes. In addition, zone A is the only zone to have a weak and negative correlation between dew formation and cloud cover. These huge differences between dew zone A and other zones are likely due to differences in topography as dew zone A is mainly covered by forests and the behaviour of some climatological variables can be different than the rest areas. A moderate negative correlation between dew formation and wind speed (-0.62) 365 does exist in zone A which may indicate that wind speed is the meteorological parameter with the most influence on dew formation in Zone A.
When the long-term trends are considered, air temperature, which has a negative effect on dew formation, showed a significant positive trend (p < 0.05) in all dew zones over the 40 years. The magnitude of these changes for zones A-F was 0.6, 0.6, 0.7, 0.4, 0.3, and 0.3 ℃ per decade; respectively. Relative humidity (RH) and cloudiness had a positive effect on dew formation 370 (except in Zone A), however, they both had a negative trend over 40 years. The average decrease in relative humidity for dew zones (i.e., A-E) was about 1.5% / decade (Table 5). Therefore, the increase in temperature and decrease in RH and cloudiness can largely explain the decreasing trend in dew yield during the last 4 decades .

Uncertainties in the dew model simulation results 375
A detailed investigation of the model setup (i.e. input parameter (e.g. emissivity and albedo), wind profile assumption, heat transfer assumptions, etc.) revealed that the dew yield estimation is very sensitive to the heat transfer coefficient. In order to obtain an estimation of the final uncertainties in the model simulation results (i.e. daily dew yield) caused by the heat transfer coefficient, we ran the model with eight different parameterizations of the heat transfer coefficients for four grid points (Table   6) for one year (2000). We selected 4 stations / grid points (red stars in Fig. 1) in different dew zones: Ramsar station in dew 380 zone A, Zanjan in dew zone B, Tabas in dew zone D, and Bandarabas in dew zone F. Figure 10 shows the daily dew yields estimated using the parameterization by Richards (2009, this study) against the daily dew yields obtained from the seven other parameterizations listed in Table 6. for all 4 grid points considered, the parameterizationss of Beysens (2005) and Watmuff (1997) give the largest estimates of daily dew yields whereas the parmeterizations of Kumar (1997) and Maestre-Valero et al. (2011) give the lowest estimates (Fig 10, Table 6). Figure 10 also 385 demonstrates that the largest differences occur for the very low values of daily dew yield.
The absolute differences in daily dew yields between the parameterizations are calculated in Table 6. At Ramsar, (located in dew zone A, in a forested region), the daily mean dew estimates range from 0.10 to 0.19 L /m 2 which is the largest absolute range of 4 stations. However, the daily dew yield at Ramsar, from all parameterizations, is the largest of all 4 stations (Table  390 6) and thus this station has the smallest relative difference: the smallest estimate of 0.10 L /m 2 is 83% of the value obtained from the Richardson parameterization and the largest estimate of 0.19 L /m 2 is 158%. Bandarabas (located in dew zone F, coastal region) also has a large range of daily dew estimates (0.01 to 0.14 L /m 2 ) but combined with the much lower daily dew yield (0.04 L /m 2 from the Richards parameterization) means this grid point has a much larger relative variation (25 % to 350%) in estimated daily dew yields in comparison to Ramsar. The heat capacity parameterization has a strong impact on the 395 modelled daily dew yields, however the standard deviations of the daily means are also large (Table 6). However, we conclude that the parameterization used in this studythe Richards (2009) formulation -gives estimates that are similar to the average of all methods and are neither much lower nor much higher than the majority of other parameterizations.
In an ideal situation we would compare our model results to observations, however, unfortunately observational data of dew formation in Iran is not available. Therefore, the accuracy of the modelled dew yields in comparison to observational data 400 cannot be performed for this study. However, Vuollekoski et al. (2015, Section 2.3 and Fig. 2) and Atashi et al. (2021) presented detailed comparisons between results from this dew model and observations in other locations, where experimental dew data was available. The results of these studies revealed that in most cases the model overestimates the dew yield due to some limitations that discussed in section 2.2; however, the cumulative sum of observed and simulated dew yield were found to agree well after with smoothing down the daily variations. 405

Discussion
Iran is a country located in arid and semi-arid regions, which has a growing population and has suffered from water scarcity over the last decades. Therefore, finding renewable sources of water is rapidly becoming a necessity. Dew is one of these atmospheric resources of water which can be vital especially in more dry conditions. 410 The average daily dew yield in Iran was in the range of 0.03 -0.14 L/m 2 and the maximum was in the range of 0.29 -0.52 L/m 2 /day. Our modelled-based results are largely in agreement with previous observational dew measurement studies conducted in similar climates (i.e., arid and semi-arid, coastal desert, Mediterranean) using planer dew condensers. However, the quantitative estimates of dew formation can differ between stations located within the same climatic zone.  Sharan et al., 2011). The average frequency of dew occurrence in Iran was 102 days, while the average number of rainy days in Iran is 38 days (Kashki and Dadashi Roudbari, 2017), suggesting that dew is more frequent than rain. Furthermore, a comparison between the total amount of harvestable dew water with rainfall in seven different stations in different climate zones in Iran performed by Atashi et al (2019) revealed that in the arid coastal areas in the south and in central desert areas, dew formation could be about 25% of 425 rainfall which is significant ( see Atashi et al., 2019, Section 3.3, Fig. 4 for further details).
Water scarcity is becoming even more serious with global warming and the impacts of climate change on water resources. As such, the dew formation yields calculated in this study showed a significant decreasing trend in the majority of Iran over the last 4 decades. Similar decreases in dew have also been reported in different areas of the world. Xu et al (2015) investigated the effects of global warming on dew variation in a paddy ecosystem in China (The Sanjiang Plain of Heilongjiang Province) 430 over the last 50 years. Their findings showed that with the current rate of change in T and RH, the average daily dew intensity would decline by 0.036 mm/year. They suggested that a warmer and drier climate would lead to a reduction in dew amount because water cannot condense when RH falls below 71%. In another study, Tomaszkiewicz et al., (2016) used the forecast trends in temperature and relative humidity to estimate dew yields under future climatic scenarios for 142 stations in the Mediterranean region during the critical summer months at the end of the century (2080). Their study predicted that dew 435 harvesting may decline (up to 27%) by the end of the century during the dry season.
In closure, it should be noted that a reliable prediction of dew is still a challenge and the model used in this study has also some limitations (i.e., heat (h) and mass (k) transfer coefficient are semi-empirical parameters, spatial and time resolution of gridded data, the "dew collecting" method in the model might be different than the measurement studies, etc.) that tends to overestimate the daily dew yield. However, uncertainty in the results caused by model assumptions is very unlikely to affect 440 the main conclusions of this study. Namely, these uncertainties do not affect the spatial (dew zones) and temporal (seasonal variation) patterns, nor the obtained results for the historical climate change impact on dew yield. Lastly, to obtain more accurate estimates of future dew formation, and thus a robust scientific basis for future water resource plans to be built upon, our dew formation model should be calibrated with actual dew experimental observations in multiple different climates; this is a topic on ongoing work. Finer spatial and temporal data resolution would also help to resolve local variations in 445 microclimates.

Conclusion
Iran is a relatively dry country with a limited source of water. Water scarcity has been a serious problem over decades, so that, considering renewable resources of water is imperative. Dew is a non-conventual atmospheric source of water that can be vital, particularly in arid and semi-arid climates, where other water resources are rare. Therefore, in this study, we estimated the 450 potential of dew water yield, identified the main dew zones in Iran and investigated the impacts of already detected climate change on dew formation. In order to estimate dew potential, we used an analytical model based on mass and heat balance between a condenser sheet and the atmosphere. Long-term (1979Long-term ( -2018 model simulation results revealed that dew can form almost everywhere in Iran, even in hyper dry deserts. The average of dew events was ~ 102 days, with the lowest number of dewy days in summer (~7 days) and the highest in winter (~ 45 days). The average daily dew yield was also in the range of 455 0.01-0.14 L/m 2 with the maximum yields in winter (0.23 mm/day). In both dew occurrence and yield, the coastal and mountain parts of Iran had the highest values and interior and eastern areas had the lowest values. Conserning the uncertainty in the model simulation results, the uncertainty analysis with eight different parameterizations of the heat transfer coefficients (dew yield estimation is very sensitive to the heat transfer coefficient) for four grid points (in different dew zones) for one year (2000). The results revealed that the parameterization used in this studythe Richards (2009) formulation -gives estimates 460 that are similar to the average of all methods and are neither much lower nor much higher than the majority of other parameterizations and the largest differences occur for the very low values of daily dew yield.
In order to identify the dew formation zones in Iran, we used a hierarchical agglomerative clustering method which identified 6 distinct dew zones. The geographical variation of the dew formation zones closely matched with the topography and the sources of moisture (e.g., nearby sea areas) in Iran. Zone A (i.e., Caspian Sea) had the highest overall mean daily dew 465 occurrence (~ 330 days) and yield (0.14 L/m 2 ), and Zone D (i.e., Lut desert zone), had the lowest dew events (~ 15 days) and yields (0.03 L/m 2 ).
The Mann-Kendal trend test revealed a significance (p < 0.05) negative trend in the yearly dew yield in the majority of Iran during the last 4 decades . This reduction in dew was mainly the result of increases in air temperature and decreases in relative humidity which are key factors in dew formation. 470

Data availability
The model and data used in this study are publicly available and can be accessed as follows: • The program source code, written in Python and Cython is available at https://github.com/vuolleko/dew_collection/.
• The meteorological input data using The European Centre for Medium-Range Weather Forecasts (ECMWF) 475 reanalysis and forecast fields (ERA-Interim): https://apps.ecmwf.int/datasets/data/interim-full-daily/levtype=sfc/.       Table 6. A selection of 8 various parameterizations for the heat transfer coefficient, the mean±std daily dew yield and [the mean differences between daily cumulative] (L/m 2 /day; negative values indicate the underestimated dew yield relative to the Richard's parameterization and positive values indicate the overestimated dew yeild) dew yield caused by each parameterization in 4 selected stations compare to the used coefficient in this study (i.e. Richards (2009)). The first three are studies on dew formation. Here, u and Ta are the horizontal wind speed and air temperature at 2m height, and L is the 765 characteristic length of the condenser (e.g. 1 m).