Contrasting roles of interception and transpiration in the hydrological cycle. Part 1: temporal characteristics over land

Moisture recycling, the contribution of terrestrial evaporation to precipitation, has important implications for both water and land management. Although terrestrial evaporation consists of different fluxes (i.e., transpiration, vegetation interception, floor interception, soil moisture evap5 oration, and open water evaporation), moisture recycling (terrestrial evaporation-precipitation feedback) studies have up to now only analysed their combined total. This paper constitutes the first of two companion papers that investigate the characteristics and roles of different evapora10 tion fluxes for land-atmosphere interactions. Here, we investigate the temporal characteristics of partitioned evaporation on land, and present STEAM (Simple Terrestrial Evaporation to Atmosphere Model) – a hydrological land surface model developed to provide inputs to moisture track15 ing. STEAM estimates a mean global terrestrial evaporation of 73 900 km year−1, of which 59 % is transpiration. Despite a relatively simple model structure, validation shows that STEAM produces realistic evaporative partitioning and hydrological fluxes that compare well with other global es20 timates over different locations, seasons and land-use types. Using STEAM output, we show that the terrestrial residence time scale of transpiration (days to months) has larger interseasonal variation and is substantially longer than that of interception (hours). Most transpiration occurs several hours 25 or days after a rain event, whereas interception is immediate. In agreement with previous research, our simulations suggest that the vegetation’s ability to transpire by retaining and accessing soil moisture at greater depth is critical for sustained evaporation during the dry season. We conclude that 30 the differences in temporal characteristics between evaporation fluxes are substantial and reasonably can cause differences in moisture recycling, which is investigated more in Part 2, the companion paper.

oration, and open water evaporation), moisture recycling (terrestrial evaporation-precipitation feedback) studies have up to now only analysed their combined total. This paper constitutes the first of two companion papers that investigate the characteristics and roles of different evapora-10 tion fluxes for land-atmosphere interactions. Here, we investigate the temporal characteristics of partitioned evaporation on land, and present STEAM (Simple Terrestrial Evaporation to Atmosphere Model) -a hydrological land surface model developed to provide inputs to moisture track-15 ing. STEAM estimates a mean global terrestrial evaporation of 73 900 km 3 year −1 , of which 59 % is transpiration. Despite a relatively simple model structure, validation shows that STEAM produces realistic evaporative partitioning and hydrological fluxes that compare well with other global es-20 timates over different locations, seasons and land-use types. Using STEAM output, we show that the terrestrial residence time scale of transpiration (days to months) has larger interseasonal variation and is substantially longer than that of interception (hours). Most transpiration occurs several hours 25 or days after a rain event, whereas interception is immediate. In agreement with previous research, our simulations suggest that the vegetation's ability to transpire by retaining and accessing soil moisture at greater depth is critical for sustained evaporation during the dry season. We conclude that 30 the differences in temporal characteristics between evaporation fluxes are substantial and reasonably can cause differences in moisture recycling, which is investigated more in Part 2, the companion paper.

1 Introduction
Terrestrial evaporation is mediated by land-surface properties, rainfall characteristics, and evaporative demand -conditions that humans are modifying at an unprecedented scale (e.g., Crutzen, 2002;Dore, 2005;Gordon et al., 2005;Rock-40 ström et al., 2009b;Trenberth, 2011). Understanding evaporation interaction with land and climate is essential, because evaporation holds a key role in regulating hydrological flows as well as atmospheric feedback. One important land-atmosphere mechanism is the contribution of terrestrial 45 evaporation to precipitation through the process of moisture recycling, which has implications for both water and land management. For example, studies have shown that changes in land-use may potentially reduce crop yields through reductions in moisture recycling (Bagley et al., 2012), that ir-50 rigation may increase moisture recycling (e.g., Tuinenburg, 2013; Wei et al., 2013), and that livelihoods in some semiarid regions are particularly vulnerable to changes in upwind moisture source regions (Keys et al., 2012).
Up to now, moisture recycling studies have only anal-55 ysed total evaporation. However, the partitioning between transpiration, vegetation interception, floor interception, soil moisture evaporation, and open water evaporation depend on land-use and meteorological conditions. For example, interception and soil moisture evaporation are ephemeral (Gerrits 60 et al., 2009), whereas transpiration continues long into the dry season depending on infiltration rates and the capacity of the soil in the root zone to retain moisture. Vegetation that can access deeper soil moisture can therefore maintain evaporation through transpiration beyond what can be sustained timescale of evaporation response through convolution representation of precipitation history and applied it on interception, soil evaporation and transpiration globally. Lohmann and Wood (2003) employed a similar approach to compare 16 land surface models and found significant differences in 90 response between models. Nevertheless, the role of evaporation partitioning and evaporation time scales specifically for moisture recycling has not been studied. Although there have been much efforts in estimating global land evaporation and evaporation partitioning, the ac-95 tual magnitudes of the different evaporative fluxes remain disputed. Methods to estimate spatially distributed global land evaporation can broadly be grouped into land surface models, remote sensing, reanalysis, and data-upscaling methods. While the latter two generally do not provide evapora-100 tion partitioning, the first two methods are highly reliant on the assumed parameters, algorithms, and terminology definitions in order to assess the partitioning. Thus, it is not surprising that the range of reported evaporation partitioning is large. Model-based global mean transpiration ratio estimates 105 range from 38 to 80 % (see Sect. 5 and Table 3).
Thus, there remain many difficulties and uncertainties in 130 estimating evaporation partitioning. In particular, the lack of evaporation partitioning data available at the spatial and temporal scale required for moisture tracking might be a reason for the omission of moisture recycling research in the potentially contrasting effects of separated evaporation fluxes.

135
The research presented here is divided into two separate research papers. The general aim is to investigate the characteristics and roles of different evaporation fluxes to the atmosphere with respect to moisture recycling. This paper (Part 1) analyses the temporal characteristics of partitioned evap-140 oration on land, and presents and evaluates STEAM (Simple Terrestrial Evaporation to Atmosphere Model) -a hydrological land surface model developed and used for the analyses. van der Ent et al. (2014), (hereafter, Part 2), tracks interception and transpiration fluxes in the atmosphere using 145 the WAM-2layers (Water Accounting Model 2-layers) and investigates the resulting moisture recycling patterns.
Specific research questions investigated in this paper relate to the temporal characteristics important for understanding the reasons for evaporation fluxes to produce different mois-150 ture recycling patterns: 1) what are the terrestrial residence time scales of evaporation fluxes? 2) how does the timing of precipitation matter for evaporation partitioning? 3) how robust are the temporal characteristics to uncertainties in storage capacities? We use STEAM to model these fluxes. As a 155 relatively simple evaporation model for analysing the relationship between land-use and moisture recycling, STEAM aims to 1) be tailored for coupling with the atmospheric moisture recycling model WAM-2layers, 2) be flexible for land-use change by land-use parametrisation and by includ-160 ing representation of features particularly important for evaporation (e.g., phenology and irrigation), 3) remain simple, transparent, and computationally efficient, and 4) simulate evaporation and evaporation partitioning in line with current knowledge. 165 2 Model description STEAM (Simple Terrestrial Evaporation to Atmosphere Model) is a process-based model assuming water balance at grid cell level. Because of our need to properly quantify partitioned evaporation and its seasonal variations, STEAM 170 includes an irrigation module and calculates dynamic seasonal vegetation parameters based on meteorological condi-tions. For our current research purposes, we have considered it acceptable to disregard groundwater interactions and lateral flows.
STEAM estimates five evaporative fluxes, and is represented by five stocks, see Fig. 1. First, the vegetation interception stock S v represents canopy and vegetation surface (such as leafs, branches, and stems) that are the first to be wetted by rainfall (P − P sf ). The evaporation from this stock is vegetation interception E v , and the water exceeding the storage capacity S v, max is throughfall P tf . Second, the floor interception stock S f represents the ground and litter surface which intercepts the throughfall. The evaporation from this stock is floor interception E f . The remainder is effec-185 tive precipitation P eff , which is generated when the storage S f, max is exceeded. Third, water that subsequently reaches the unsaturated root zone stock S uz can be evaporated either as soil moisture evaporation E sm , or be taken up by plant roots and transpire as transpiration E t . Fourth, the water 190 stock S w represents open water in the land-use classes water (01:WAT) and wetlands (12:WET), and water below vegetation in the land-use classes wetlands (12:WET) and rice paddies (19:RIC). The water stock is replenished by adding water J add that prevents dry-out in the absence of lateral flow 195 routines. Water below vegetation also receives P tf from vegetation. Excess water comprises Q uz (exceeding S uz, max ) from the unsaturated zone and Q w from the water stock (exceeding S w, max ). The last and fifth stock S snow does not have a limit, and allows snowfall P sf to accumulate until melting occurs.

200
Snowmelt P melt is allowed only if there is snow in S snow . If the daily mean temperature T mean is above 273 K, P melt goes directly to the floor interception stock, otherwise it only adds to Q uz . In case of irrigation, some water is assumed to be spilled to the vegetation I v , the floor I f and the water bodies 205 I w . All notations are listed in Appendix A.

Potential evaporation
Total evaporation, the sum of vegetation interception E v , floor interception E f , transpiration E t , soil moisture evaporation E sm , and open water evaporation E w , is driven by the 210 daily potential evaporation, and restricted by resistances and water availability. The Penman-Monteith equation (Monteith, 1965) is used to estimate the daily potential evaporation E p, day [m d −1 ], which is formulated as follows: ture evaporation. Thus, we introduce k (used in Eq. 8, 10, and 11), which is expressed as a function of a surface resistance r s and an aerodynamic resistance r a : (2)

235
The surface stomatal resistance r s, st is calculated based on the Jarvis-Stewart stress function and optimal temperature based on latitude and altitude, see Appendix B2 for details. The soil moisture resistance r s, sm is applied to soil moisture evaporation and estimated based on the soil moisture content 240 of the top soil layer (Bastiaanssen et al., 2012): where r s, sm, min is the minimum surface soil moisture resistance assumed as 3.5 × 10 −4 d m −1 , and Θ top [-] is the ef-245 fective saturation expressed as: Since there is no explicit top soil storage in STEAM, top soil moisture at the present time θ top, n [-] is derived 250 daily, based on the inflow to the unsaturated storage and top soil moisture from the previous day θ top, n−1 (Pellarin et al., 2013): where ∆n is the time step of 24 h, θ top, res is the volumetric 255 residual soil moisture content assumed as 0.01, y top is the top soil depth, and χ is the dry out parameter which varies with clay content of the top soil. The assumed y top is 0.03 m. In Pellarin et al. (2013), the values used for y top were 0.05 m and 0.1 m, but we considered that a shallower depth is more 260 relevant for estimating soil moisture evaporation stress. The dry out parameter χ is estimated using the following semiempirical equation: where η clay is the clay content [%] and χ min is the minimum of χ taken as 60 h. This set of equations (Eq. 5 and 6) was tested in semi-arid West Africa, in the type of regions where soil moisture evaporation is most important. Factors not taken into account include solar radiation, the presence 270 of vegetation and the wind velocity (Pellarin et al., 2013).

Actual evaporation
To simulate actual evaporation at 3 hour time steps (∆t), we first downscale the daily potential evaporation E p, day using the diurnal distribution of ERA-I 3 h evaporation. The down-275 scaled potential evaporation is subsequently used to evaporate moisture in the following logical sequence -vegetation interception, transpiration, floor interception, and soil moisture evaporation: E sm, lu, vs = min S uz, lu ∆t , a (10) where the first subscript ( v , t , f , sm or w ) denotes an individual evaporative flux, the second subscript ( lu ) the land-use type ID (see Table C1), and the third subscript ( vs , vw or ow ) the type of vegetation-water occupancy (see Table C2). Thus, for the fraction of vegetation in water φ vw in wetlands and For the water land-use type and the fraction of open water 300 φ ow in wetlands, evaporation is expressed as: The total of an evaporation flux from wetland (12:WET) or rice paddy (19:RIC) is determined by the weighted sum 305 based on the fractions of vegetation covered soil φ vs , vegetation covered water φ vw , and open water φ ow (see also Table C2): E j, lu = φ lu, vs E j, lu, vs + φ lu, vw E j, lu, vw + φ lu, ow E w, lu, ow (14) 310 where E j, lu is an evaporation flux ( j denotes v , t , f , sm , or w ) of the land-use type lu .
Subsequently, the total of an evaporation flux from a grid cell is determined by the weighted sum of the land-use types: where φ lu is the land-use occupancy fraction of the land-use type lu .

Phenology
The growing season index i GS (Jolly et al., 2005) varies be-320 tween 0 and 1, and is used to determine the seasonal variations of leaf area i LA . We formulate i GS in STEAM as follows:

325
where f (T min ) is the stress function of minimum temperature, f (N ) is the stress function of day length, and f (θ uz ) is the stress function of soil moisture. Note that f (θ uz ) is a modification of the original expression for i GS , where vapour pressure deficit D a was used as a proxy for soil moisture 330 (Jolly et al., 2005). However, since soil moisture is calculated in STEAM, it makes sense to use the soil moisture stress function to replace the original vapour pressure stress function. The stress functions are expressed as: Tmin−Tmin, low Tmin,high−Tmin,low T min,high > T min > T min,low ,

1
T min ≥ T min,high where the lower sub-optimal minimum temperature T min, low is 271.15 K, and the higher T min, high is 278.15 K. The lower 340 sub-optimal threshold day length N low is assumed to be 36 000 s, and the higher N high is 39 600 s (Jolly et al., 2005). T min is taken from the coldest 3 h ERA-I temperature of the day. Calculation of day length N follows the approach of Glarner (2006). The soil moisture stress parameter c uz is 345 fixed at 0.07 (Matsumoto et al., 2008). The soil moisture content θ uz is S uz /y uz , where y uz [m] is the depth of the unsaturated root zone. The soil moisture contents at wilting point θ uz,wp and at field capacity θ uz,fc depend on soil type. To prevent unrealistically unstable fluctuations in leaf area, the 350 mean i GS,21 of the past 21 days is used to scale i LA between the land-use type dependent i LA,max and i LA,min (Jolly et al., 2005): 355

Storage capacities
The storage capacity determines the maximum water availability for the evaporation flux of concern. We derived vegetation interception storage capacity S v, max [m] from the monthly i LA based on the storage capacity factor c sc of 360 roughly 0.2 reported by, for example, de Jong and Jetten (2007) and used in van den Hoof et al. (2013): where c AR is the area reduction factor introduced to com-365 pensate for rainfall heterogeneity in space and time. The relationship between i LA and vegetation interception storage varies with vegetation type and a strong relationship has not yet been established. In fact, Breuer et al. (2003) even suggests that no general relationship can be established across 370 vegetation types due to the inherent differences in vegetation structures. Nevertheless, vegetation stock linked to i LA has proven to be useful in many cases where there is a lack of detailed vegetation information.
We assume c AR to be 0.4 for STEAM running with a 3 h 375 time step at the 1.5 • scale. Area reduction factors have been developed to establish a relationship between average precipitation and extreme precipitation of a region, but can be analogously used to reduce interception storage capacity. In an example diagram obtained from catchment analyses (Shut-380 tleworth, 2012), areas larger than 10 000 km 2 have an area reduction factor up to approximately 0.6. In STEAM, grid cell areas with 1.5 • resolution are 10 000 km 2 already at 68 • N, and grow to almost 28 000 km 2 at the equator. Ideally, c AR should vary with the area considered and rainfall duration, 385 but due to a lack of well-established functions, we consider c AR = 0.4 to be acceptable. The floor interception storage capacity S f, max [m] is modelled as a function of the leaf area and a certain base value:

390
The floor storage capacity increases in areas with vegetation, due to litter formation from fallen leafs. A base value is considered, because wetting of the surface always occurs irrespective of the land cover. However, litter is assumed 395 to have been removed in croplands (i.e., 13:CRP, 15:MOS, 18:IRR, and 19:RIC). Thus, S f, max [m] for crops corresponds to that of the litter-free floor:

400
As a result of the large grid scale (reflected in the area reduction factor), interception storage in STEAM is smaller than normally found in point scale field studies. For example, the vegetation interception storage capacity at the maximum i LA of 5.5 is 0.44 mm, which is about a third of the 1.2 mm 405 reported in a summer temperate forest (Gerrits et al., 2010) and a fraction of the 2.2 to 8.3 mm per unit of crown projected area in a tropical rainforest site (Herwitz, 1985).
The storage capacity of the unsaturated root zone S uz, max is assumed to reach field capacity when: The S uz,max is modelled as a function of soil texture and land-use based rooting depth. This is a simplification as many other factors govern root water uptake, including topography 415 (Gao et al., 2013), soil properties, hydraulic redistribution of soil water by roots (Lee et al., 2005), groundwater table (Miguez-Macho and Fan, 2012), and climate (Feddes et al., 2001). In addition, variations of rooting distribution (e.g., Jackson et al., 1996) and the existence of deep roots (e.g.,

420
Canadell Kleidon and Heimann, 2000) may conflict with the assumption of one rooting depth parameter per land-use type. 425 constitute an important moisture source to the atmosphere (e.g., Gordon et al., 2005;Lo and Famiglietti, 2013;Tuinenburg, 2013;Wei et al., 2013). Irrigation water supplied is assumed to meet the irrigation requirement and is not restricted by water availability. Net irrigation enters the unsaturated 430 zone and is estimated as a function of soil moisture. In rice paddies (19:RIC), irrigation water simply upholds a 10 cm water level. For non-rice crops (18:IRR), irrigation requirement I req is the amount of water needed to reach field capacity in the unsaturated root zone:

STEAM includes irrigation because it has been shown to
However, because a certain amount of irrigation water applied is always lost due to inefficiencies in the system, an irrigation efficiency should be applied in order to correctly 440 estimate runoff and water withdrawal. In STEAM, we assume the gross irrigation I g to be twice the I req . Although irrigation efficiency in practice varies greatly with irrigation technique, crop type and country (Rohwer et al., 2007), we consider our simplification acceptable since the gross irriga-445 tion assumption affects evaporation (our major concern) less than, e.g., runoff and water withdrawal. Of gross irrigation applied to irrigated non-rice crops (18:IRR), 15 % is directed to the vegetation interception stock S v , and 85 % to the floor interception stock S f . Of the gross irrigation applied to rice 450 paddies (19:RIC), 5 % is directed to vegetation interception stock S v , 5 % to the floor interception stock S f (assuming inter-paddy pathways), and 90 % to the water stock S w .

Data
Meteorological data were taken from the ERA-Interim 455 reanalysis (ERA-I) produced by the European Centre for Medium-Range Weather Forecasts (ECMWF) (Dee et al., 2011). We used evaporation, precipitation, snowfall, snowmelt, temperature at 2 m height, dew point temperature at 2 m height, wind speed in two directions at 10 m height, 460 incoming shortwave radiation, and net longwave radiation. All meteorological forcings are given at 3 h and 1.5 • latitude × 1.5 • longitude resolution. The data used covers latitudes from 57 • S to 79.5 • N for the years 1985-2009.
The monthly varying land-surface map used in STEAM

465
consists of 19 land-use types, see Table C1.  (Saxton and Rawls, 2006). Top soil saturation, subsoil field capacity and subsoil wilting point have been assigned to the original 30 resolution, and scaled up to 1.5 • by area weighing.
For evaporation evaluation, we used the LandFlux-EVAL 490 evaporation benchmark products (Mueller et al., 2013) for the years 1989-2005. This data product consists of a merged synthesis from 5 satellite or observation-based datasets, 5 land-surface model simulations, and 4 reanalysis datasets. For runoff evaluation, composite and model runoff fields 495 from the Global Runoff Data Centre (GRDC) were used (Fekete et al., 2000). The model runoff fields are the simulations of the GRDC Water Balance Model (GRDC-WBM), whereas the composite runoff fields (GRDC-Comp) are the model runoff corrected by observed inter-station discharge 500 (Fekete et al., 2000). In addition, we also used ERA-I runoff fields (Balsamo et al., 2011) in our comparison. It should be noted that the ERA-I runoff fields form a separate dataset that does not directly correspond to ERA-I precipitation minus evaporation. The river basin map is based on the global 505 30-min drainage direction map of Döll and Lehner (2002).

Model evaluation
The model evaluation comprises the following model output: total and land-use based evaporation, total and land-use 510 based evaporation partitioning, runoff, irrigation, and irrigation evaporation contribution. Total global fluxes are calculated based on a land area of 133 146 465 km 2 (including Greenland and excluding Antarctica) and for the years 1999-2008. Land-use evaporation is obtained from Eq. 14. Irri-515 gation evaporation contribution was calculated based on the difference in evaporation between STEAM simulations with and without the irrigation routine turned on. Runoff Q from STEAM has been derived from subtracting mean evaporation and mean snow storage changes from mean precipitation: Snow storage changes were subtracted because snow accumulated in glaciers may carry over storage from year to year. Otherwise, most storage changes may be neglected at 525 an annual time scale. Then runoff comparison includes two additional STEAM scenarios: one simulation without irrigation (because irrigation is not always included in land surface models), and one with 5 % uniform reduction in precipitation forcing (because ERA-I precipitation forcing is higher than 530 several other precipitation datasets, see Appendix D).

Time scales of evaporation fluxes
The time scales τ ts of the evaporation fluxes is defined as the mean stock over the mean flux rate of concern j: 535 Figure 1 shows the stock of origin for each evaporation flux. Because both E sm and E t come from S uz , we assumed a stock of soil moisture evaporation S uz, sm and a stock of 540 transpiration S uz, t . To obtain S uz, sm , we multiplied θ top with the assumed top soil depth y top . To obtain the stock S uz, t , S uz, sm was subtracted from the total water available in the unsaturated zone S uz :

Evaporation partitioning: time since precipitation
We are interested in how evaporation partitioning evolves with time after precipitation ceases. To do this, we grouped 555 each grid cell at every time step in categories depending on the time that has past since precipitation. Grid cells at a certain time step that has not received precipitation since n time steps back are placed in the (n+1) th category. Subsequently, evaporation partitioning for each category was retrieved from 560 the model simulation.
In addition, the importance of the evaporation partitioning in relation to rainfall also depends on the evaporated quantity. Thus, we present the portion of evaporation flux during rainy or dry conditions by using evaporation efficiencies β wet and 565 β dry as measures: Here, β wet represents the mean annual portion of an evaporation flux that evaporates during a 3 hour time step with 570 precipitation, and β dry represents the mean annual portion of an evaporation flux that evaporates after experiencing more than 24 hours of no precipitation. To qualify as a wet time step, a 3 hour time step must have >0.01 mm precipitation. The subscript j denotes the evaporation flux of concern. Con-575 struction of these evaporation efficiency measures is useful for answering questions such as: how much of total vegetation interception occurs during rainy periods?

Robustness
Large uncertainties exist in evaporation partitioning and esti-580 mation of storage capacities. To verify how robust or sensitive the temporal characteristics are to these uncertainties, we performed a sensitivity analysis with two scenarios: transpiration-plus and interception-plus. In transpirationplus, the unsaturated zone storage capacity increased by 20 % 585 and the vegetation and floor interception storage capacity decreased by 50 %. In interception-plus, the increase and decrease in the storages are reversed, see Table 5. STEAM estimates global annual terrestrial evaporation as 555 mm year −1 (i.e., 73 900 km 3 year −1 ), spatial distribution is shown in Fig.2. This is comparable to current global evaporation datasets. In the Water Model Intercomparison Project (WaterMIP), the range of evaporation given by eleven 595 models was 415-585 mm year −1 for the period 1985-1999 forced with WATCH meteorological data (Haddeland et al., 2011). By subtracting global runoff from precipitation products for the years 1984-2007, Vinukollu et al. (2011) arrived at global evaporation rates of 488-558 mm year −1 600 (i.e., 64 000-73 000 km 3 year −1 ). In the LandFlux-EVAL multi-data set synthesis, the global mean evaporation was 493 mm year −1 as given by a combination of land-surface model simulations, observational dataset, and reanalysis data for both the period of 1989-1995and 1989 et al., 2013). Figure 3 shows how STEAM evaporation compares to the LandFlux-EVAL product for 1989-2005. STEAM evaporation is within the inter-quartile range of all LandFlux-EVAL products in the tropics, the United States, parts of Europe, 610 South Asia, northern Russia and large parts of Africa south of Sahel. The upper quartile is mostly exceeded in the boreal forests in the northern latitudes, China, Argentina and the Sahel. Most exceedance of STEAM evaporation is in comparison with the land surface models, and the least with the re-615 analyses data included in the LandFlux-EVAL product. Only a few limited patches in northern Canada, Sudan, Argentina and northern China exceed the LandFlux-EVAL maximum. Seasonally, Fig. 4 shows that Northern Hemisphere spring and summer are generally more in range compared to winter 620 and fall, when STEAM tends to have higher evaporation rates in the northernmost latitudes compared to LandFlux-EVAL. However, LandFlux-EVAL excluded some high evaporation values in the northern latitudes based on physical constraints (Mueller et al., 2013), which consequently eliminates poten-625 tially important winter time interception (Schlaepfer et al., 2014).
Evaporation contributions per land-use type are listed in Table 1, and compared to the other studies in Table 2. The highest evaporation rates are found in irrigated lands, 630 evergreen broadleaf forests, and open waters. This is followed by wetlands, savannahs, deciduous broadleaf forests, natural mosaics, woody savannahs, mixed forests, and rainfed croplands. Evaporation rates in the lower tier include contributions from needleleaf forests, grasslands, and shrub-635 lands. In general, STEAM evaporation is comparable to the estimates of Gordon et al. (2005), the compilation results of Schlesinger and Jasechko (2014) (based on Mu et al., 2011), and the field data from Rockström et al. (1999). The mixed forest evaporation estimate in STEAM is double that of Gor-640 don et al. (2005), but the area is also very different, suggesting substantial differences in forest definition. Closed shrublands in STEAM also produces higher evaporation rates, but because the numbers are for shrublands in general and not closed shrublands in particular, the shrublands comparison is 645 inevitably inconclusive. Some caution is warranted in comparing evaporation rates across studies. Nevertheless, this comparison shows that evaporation estimates in STEAM are within the range of previous estimates.

Evaporation partitioning 650
In STEAM, the dominating evaporation flux is transpiration E t (59 %), followed by vegetation interception E v (21 %), floor interception E f (10 %), soil moisture evaporation E sm (6 %) and lastly, open water evaporation E w (4 %). The global distribution of the annual mean evaporation fluxes is shown in Figs. 2 and 5 (as percentage of total evaporation). Seasonal variations of evaporation fluxes are shown over latitudes in Fig. 6. It is shown that transpiration dominates in the densely vegetated areas in the tropics. In addition, transpiration rates increase over the boreal forests during the Northern Hemisphere summer. Table 3 provides an overview of evaporative partitioning values in the literature and in STEAM. We note that the STEAM global mean transpiration ratio is in good agreement with the literature compilation results presented by Schlesinger and Jasechko (2014) and the LPJ estimate by Gerten et al. (2005), but higher than other land-surface model simulations (Alton et al., 2009;Lawrence et al., 2007;Choudhury et al., 1998;. Jasechko et al. (2013Jasechko et al. ( , 2014 estimated the transpiration ratio to be 670 80-90 % using a combination of isotope measurement techniques and satellite observations at river basin and the global scales. However, their results have been challenged by Coenders-Gerrits et al. (2014) who showed that the transpiration ratio reduces to 35-80 % by using other input data, 675 Schlesinger and Jasechko (2014) who estimated the global transpiration ratio to be 61 % based on literature data compilation, and by Schlaepfer et al. (2014) who argued that Jasechko et al. (2013)'s underlying assumption that isotope ratios of a lake would be representative for the entire catch-680 ment is flawed. A number of possible explanations for the high transpiration ratio bias in isotope studies is also offered by Sutanto et al. (2014). Table 1 shows the annual average evaporation fluxes as a percentage of total evaporation per land-use class. Transpi-  Among the more vegetated land-use types, vegetation interception ratios are highest in forests (21-37 % of E), followed by croplands (17 %), and lowest in the sparsely vegetated land-use types: shrublands, savannahs, grasslands, wetlands, and urban lands (10-14 %). Floor interception values 695 follow the pattern of vegetation interception. Thus, floor interception is generally higher than soil moisture evaporation in forests, whereas soil moisture evaporation equals or exceeds floor interception more often in shrublands and croplands.

700
Reported land-use specific evaporative partitioning in previous research is scarce at the global scale. Lawrence et al. (2007) do not report evaporative partitioning by land use (from simulation using Community Land Model version 3), but map figures indicate that their soil evaporation is higher 705 and canopy interception is lower in savannahs, grasslands, and shrublands occupied areas compared to STEAM. Transpiration ratios of CLM3 are comparable with STEAM in forested and savannah areas, but are much lower (down to < 30 %) in the western US, India, southeastern China, and 710 South Africa. Alton et al. (2009) report global mean transpiration ratios of 49-65 % in forests, 32-60 % in grassland, and 44-51 % in shrublands. The order of magnitude is similar to STEAM, but transpiration ratios for shrublands are lower. Schlesinger and Jasechko (2014)  Sensitivity of STEAM evaporation partitioning to precipitation is analysed by a 5 % uniform reduction of precipitation, see Appendix D.  (Fekete et al., 2000). Thus, the STEAM runoff estimate appears to be slightly higher than some of the previous estimates, but lies within the uncertainty range. Differences can partly be explained by the terrestrial area considered in the studies, as well as relatively high P applied 770 (see Appendix D).
STEAM runoff was also compared to GRDC-Comp, GRDC-WBM, and ERA-I runoff data in 13 major river basins of the world, see Figs. 7 and 8a. The largest deviations for both STEAM and ERA-I from the GRDC-Comp 775 runoff are found in the Congo and Nile river basins. However, because Congo precipitation and runoff estimates are particularly uncertain in general (Tshimanga, 2012), we can not evaluate our Congo evaporation estimate based on this specific comparison. As for the Nile river basin, STEAM uses 780 a static land-use map that does not include seasonal variations in wetland size or presence of reservoirs. Since the Nile contains the Sudd, one of the largest wetlands in the world with a highly variable size, evaporation simulation is challenging in this region, even in fine resolution models includ-785 ing complex processes (Mohamed, 2005;Mohamed et al., 2007). In several of the northern river basins (e.g., the Mississippi, Mackenzie, and Danube), STEAM runoff is low in comparison to GRDC-Comp. There could be multiple reasons for this underestimation: our simplified snow simula-790 tion, our uniform parametrisation of land-use classes across climate zones or simply uncertainties in the forcing data. In support of the latter, the largest uncertainties in evaporation inferred from precipitation and runoff data occur mainly in the higher latitudes (Vinukollu et al., 2011). Table 4 shows that the STEAM evaporation is close to the mean evaporation provided by the WaterMIP (Water Model Intercomparison Project) (Haddeland et al., 2011;Harding et al., 2011), while both the simulated runoff and the used precipitation forcing is substantially lower. In contrast, in 800 the Lena river basin, STEAM runoff is in range while both evaporation and precipitation have a high bias. In the Amazon basin, the default STEAM simulation slightly overestimates runoff, but reducing precipitation forcing by 5 % (see the 95 %-P run in Fig. 7) brings runoff down to the level 805 in GRDC-Comp. Also the comparison with WaterMIP indicates that high bias in Amazon precipitation translates into high runoff. This effect of precipitation reduction can also be noted in particularly the Brahmaputra-Ganges, Congo, and Nile river basins. This is not surprising, because precipita-810 tion uncertainties have been shown to translate almost entirely into uncertainty in runoff in wet regions, but not at all in arid regions (e.g., Fekete et al., 2004). The relative sensitivity of runoff and evaporation fluxes to precipitation is further accounted for in Appendix D.

815
The standard deviations between the multiyear mean runoffs in GRDC-Comp (which we here consider as the benchmark runoff) and the other runoffs (GRDC-WBM, ERA-I, and STEAM) are shown in Fig. 8b and c. Among the compared datasets, STEAM runoff deviates the most 820 from GRDC-Comp when Congo is included and the least when Congo is excluded. Note also that omitting irrigation in STEAM increases the runoff deviation to GRDC-Comp, and that reducing precipitation decreases this deviation. Thus, the wet bias in ERA-I precipitation probably explains some of 825 the runoff overestimations we notice in STEAM.

Irrigation
The simulated mean gross irrigation is 1970 km 3 year −1 , and the simulated mean increase in evaporation from irrigation is 1134 km 3 year −1 . The irrigation hotspots in especially In-830 dia, south-eastern China, and the central US coincide well with where evaporation is enhanced by irrigation input. Our estimates are comparable to previous estimates. Gross irrigation was estimated at 2500 km 3 year −1 by Döll and Siebert (2002), at 2353 km 3 year −1 by Seckler et al. (1998), and at 835 1660 km 3 year −1 by Rost et al. (2008). The latter study did, however, not take into account recharge to the groundwater. Evaporation contribution by irrigation was simulated at 1100 km 3 year −1 by Döll and Siebert (2002). While higher evaporation contributions have also been reported in the lit- The modelled global average time scale (Eq. 27) is 1.3 h for vegetation interception and 7.7 h for floor interception, but 42 days for soil moisture evaporation and 274 days for transpiration in areas with mean evaporation rates higher than 850 0.01 mm d −1 . Evaporation rates from vegetation cover and floor are large compared to their respective stocks, resulting in small time scales for interception. In contrast, the stocks in the unsaturated zone are many times larger than the interception stocks, and cause the time scales of soil moisture evapo-855 ration and transpiration to extend from days and months. The use of an area reduction factor (see Eq. 21 and 22) leads to interception storage capacities that are smaller in the model than in reality, thus, presumably causing some underestimation of the interception time scales. Nevertheless, the robust-860 ness test (Table 5) shows that the magnitude of all evaporation time scales (except for transpiration) are relatively robust against uncertainties in storage capacities. Figure 9 shows the spatial distribution of mean terrestrial residence time scales (i.e., stock divided by flux) of the par-865 titioned evaporation fluxes (Eq. 27). We see that time scales are in general prolonged over the tropics, and over the cold northern latitudes. This finding is consistent with the transpiration response time scale provided by Scott et al. (1997). Over the tropics, evaporation rates are high, but the stocks are also relatively larger. The time scales of floor and soil moisture evaporation are extended in the tropics, because these fluxes there are suppressed by the relatively high vegetation interception and transpiration.
The temporal variation of the evaporation fluxes at differ-875 ent latitudes is displayed in Fig. 10. Seasonality is distinct for all evaporation fluxes, in particular for transpiration time scales. While the mean latitude transpiration time scale can extend to over 500 days in the mid-latitude winter, it falls well below 100 days in the summer.
Regions and seasons with extremely high transpiration time scales (>300 days) largely coincide with low transpiration in the north, whereas high transpiration rates coincide with intermediate or low time scales (<100 days). On the contrary, relatively high vegetation interception time scales 885 seem positively correlated with high vegetation interception in the tropics, (compare Fig. 2 and 9). This difference can be explained by the limiting factor to evaporation. Transpiration time scales approach infinity as the stock is still wet, whereas vegetation interception time scale often approaches 890 zero when vegetation interception is caused by depletion in vegetation interception stock rather than in evaporative demand. Thus, the high transpiration time scales in the north should be understood as the result of declining evaporative demand, whereas the high vegetation interception time scales 895 in the tropics can be interpreted as the result of a steady and ample supply of precipitation to the vegetation interception stock.
The higher the interception ratios, the lower the evaporation time scales on land (also in consistency with e.g., Scott 900 et al. (1995)), and the faster the overall feedback to the atmosphere. The regions that have a high vegetation interception ratio ( Fig 5) coincide with the regions with low atmospheric moisture recycling length scales (van der Ent and Savenije, 2011). This suggests that tropical interception is very impor-905 tant for vegetation to maintain atmospheric moisture in the air, and could constitute a large portion of local recycling due to immediate feedback. However, moisture supplied to continents in general (van der Ent et al., 2010), the world's most important croplands (Bagley et al., 2012), or for rainfall 910 dependent regions (Keys et al., 2012) also relies on remote evaporation sources, which could account for a large part of transpiration. For such cases, upwind modifications that result in changed transpiration rates (e.g., changes in vegetation species, rainwater harvesting practice, CO 2 concentrations) 915 may play a larger role for downwind regions than changes in interception. A detailed investigation of the role of interception and transpiration for local and remote moisture recycling is performed in Part 2.
6.2 Evaporation partitioning in relation to time since 920 precipitation Figure 11 shows the mean latitudinal evaporation ratios by time since precipitation last occurred. Mean latitudinal transpiration ratio is up to 40 % during the wet time steps with precipitation, but can amount to up to 90 % after just a few 925 dry 3 hour time steps. The largest increase in transpiration ratios with time since precipitation are seen in the cold northern latitudes, where moisture availability is expected to exceed evaporative demand. On the contrary, the vegetation interception ratio is high (up to approximately 60 %) during 930 wet time steps, but falls drastically to almost no interception within 6 hours. Similarly to transpiration, soil moisture evaporation ratios generally increase with precipitation-free hours. However, the steepest increase in soil moisture evaporation ratios are found in the equatorial band where the total 935 soil moisture evaporation is very low. Table 5 shows that transpiration and soil moisture evaporation occur both during wet and dry conditions, whereas vegetation and floor interception evaporation occur almost exclusively during time steps with precipitation. The table shows 940 that 31 % of all transpiration occurs during time steps that have endured more than one day of no precipitation, when no vegetation interception occur. Instead, 96 % of the vegetation interception occurs in a 3 hour time step with precipitation, whereas only 45 % of transpiration evaporates in 945 such conditions. Noteworthy is also that these evaporation efficiency numbers (Eq. 29) are robust to changes in the evaporation partitioning: for example, the 96 % vegetation interception efficiency persists even when the vegetation interception ratio varies between 12 and 27 %. In other words, 950 even with large differences in the evaporation ratio, interception is likely to occur almost exclusively within the wet period, whereas transpiration may have a substantial time lag between the moment water enters the soil and exits through a plant's stomata. In for example the field study of Farah et al. 955 (2004), transpiration at a tropical woodland site continued for two months after rainfall. This also explains why transpiration dominates in the dry season and could have substantial effects on moisture recycling patterns (which will be analysed in Part 2). Furthermore, although a change in evapora-960 tion partitioning does not change the vegetation interception and transpiration efficiencies, it changes the total evaporation efficiency and the overall temporal distribution of evaporation.

965
This paper developed and evaluated the global hydrological land-surface model STEAM, and used the model output to analyse the terrestrial temporal characteristics of different evaporation fluxes on land. STEAM is designed to 1) be tailored for coupling with the atmospheric moisture recycling model WAM-2layers, 2) be flexible for land-use change by land-use parametrisation and by including representation of features particularly important for evaporation (e.g., phenology and irrigation), 3) remain simple, transparent and computationally efficient, and 4) simulate evaporation and evaporation partitioning in line with current knowledge.
The ability of STEAM to simulate evaporation and evaporation partitioning realistically was evaluated by comparison with other modelling studies, global datasets, and reported values from field studies. STEAM's total terrestrial evapora-980 tion rate (73,900 km 3 year −1 ) is comparable with previous estimates -lower than reanalysis products, but higher than other land-surface models. Reasons for this include that we do not add water in data assimilation as in reanalysis, and compared to other land-surface models we use a relatively 985 high precipitation input and also include irrigation and wetlands. Overall, STEAM simulates global evaporation partitioning within the range of previous estimates: 59 % transpiration, 21 % vegetation interception, 10 % floor interception, and 6 % soil moisture evaporation. The global mean transpi-990 ration ratio in STEAM is similar to or somewhat higher than other land-surface models, and in line with the recent literature compilation study of Schlesinger and Jasechko (2014). Vegetation interception ratios in forests are comparable with both the findings from a global satellite based estimate of 995 interception (Miralles et al., 2010) and with reported values from field studies in the tropics. In agreement with previous studies (McNaughton and Jarvis, 1983;de Bruin and Jacobs, 1989;Teuling et al., 2010), STEAM also simulates higher transpiration ratios in short vegetation types than in forests.

1000
Simplifications in STEAM include neglect of runoff routing, groundwater, and sublimation processes. Koster and Milly (1997) and Koster and P. Mahanama (2012) concluded among others that compatibility between runoff and evaporation formulations can be important due to interaction through 1005 soil moisture. Dry season evaporation might also be underestimated by the neglect of groundwater (Miguez-Macho and Fan, 2012) and hydraulic redistribution of soil water by roots (Lee et al., 2005). Crop simulations presently also do not follow sowing and harvesting dates. The neglect of sublimation 1010 can further cause underestimation of interception (Schlaepfer et al., 2014). Nevertheless, the model evaluation analyses and the sensitivity tests suggest that that the current model setup is a reasonable simplification for the research questions asked.

1015
Our analyses show a striking difference in mean annual global time scales between the different evaporation fluxes: 95-434 days for transpiration, 42-46 days for soil moisture evaporation, 5.2-11.6 hours for floor interception, and 1.1-1.6 hours for vegetation interception. The time scales also 1020 vary greatly over the seasons and latitudes. Most transpiration occurs several hours or days after a rain event, whereas interception is immediate. We find that 31 % of all transpiration occurs in time steps that have endured more than one day without precipitation, when no vegetation interception 1025 occurs. Instead, 96 % of the vegetation interception occurs in a 3 hour time step with precipitation, whereas only 45 % of the transpiration occurs in such conditions. Uncertainties in parametrising storage capacities affect the evaporation partitioning ratios, but have a smaller effect on the relative differ-1030 ences in temporal characteristics. Only the transpiration time scales are significantly changed by changed storage capacity, but are still substantially different from the interception time scales. We note that high vegetation interception ratios coincide with high local evaporation recycling, which suggests 1035 that tropical interception may have an important role for vegetation to maintain atmospheric moisture in the air. This will be subject to further investigation in Part 2.
STEAM runs at the same temporal and spatial scale as the atmospheric moisture recycling model WAM-2layers, and 1040 can be used in both one and two-way coupling. One-way coupling, i.e., forcing WAM-2layers with STEAM output, is used in Part 2 to investigate the differences in moisture recycling between direct and delayed evaporation fluxes. Twoway coupling, i.e., feeding induced changes in precipitation 1045 from WAM-2layers back to STEAM, can be applied in later studies to investigate the effect of land-use change on moisture recycling. Although WAM-2layers does not simulate precipitation, such analyses are possible by assuming that changes in terrestrial evaporation proportionally alters the at-1050 mospheric moisture content or the precipitation with continental origin.
The importance of land use for the hydrological cycle, the climate, and the Earth system as a whole has been stressed in many studies (e.g., Feddema et al., 2005;Gordon et al., 1055 2005; Rockström et al., 2009a). Thus, changes in evaporative partitioning following e.g., land-use change may have implications and provide answers for landscape resilience, drought development, and effects on remote fresh water resources. The differences in moisture recycling patterns be-1060 tween delayed and direct evaporation fluxes constitutes the case for investigation in Part 2 for the present day situation. Future research should also extend to land-use change scenario analysis to quantify and improve the assessment of land-use change effects on global fresh water resources.         . Evaporation partitioning with time since precipitation over terrestrial latitudes (1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008). Results are discussed in Sect. 6.2.       (Haddeland et al., 2011;Harding et al., 2011). The ERA-I precipitation used to force STEAM and the WFD (Watch Forcing Data) precipitation used to force WaterMIP are also shown for each compared river basin. Results are discussed in Sect. 5.3.

B1 Input variables to the Penman-Monteith equation
The vapour pressure deficit D a is defined as: The psychrometric constant γ [kPa K −1 ] is where p is the atmospheric pressure [kPa], and ξ mw is the 1095 ratio of the molecular weight of water vapour to that for dry air [0.622]. Net radiation is calculated by: where α is albedo, R sw is the incoming shortwave radiation and R net, lw is the outgoing net longwave radiation. In reality, albedo varies with angle of reflection and the surface properties such as snow cover change and soil wetness. Here, we assume α to be fixed for each land-use type, see Table C1.

1105
Daily ground heat flux G is derived from interpolating monthly ground heat flux G month (Allen et al., 1998) G month = 0.07(T month+1 -T month−1 ).
There are three types of aerodynamic resistances used in 1110 STEAM: the aerodynamic vegetation resistance r a, v , the aerodynamic floor resistance r a, f , and the aerodynamic water resistance r a, w . They are expressed as follows (Shuttleworth, 2012): .
The aerodynamic roughness length z 0 [m] is estimated from:

B2 Surface stomatal resistance
Surface resistance applies only to transpiration and soil moisture evaporation, since interception and open water evapora-1145 tion occur without resistance. The surface stomatal resistance r s, st of vegetation is simulated by the Jarvis-Stewart equation (Stewart, 1988), taking into account of solar radiation, vapour pressure deficit, optimum temperature, and soil moisture stress: where r s, st, min is the minimum surface stomatal resistance dependent on land-use type and specified in the land-use lookup table (Table C1), i LA,eff is the effective leaf area index (unit 1155 leaf area per unit ground area that is actively participating in transpiration) and f are the four stress functions for incoming shortwave radiation R sw in W m −2 , vapour pressure deficit D a , mean daily temperature T mean and soil moisture θ uz (Stewart, 1988). Effective leaf area index i LA,eff is adapted 1160 from Allen et al. (2006) and Zhou et al. (2006) as: The stress functions vary between 0 and 1. The stress function of soil moisture f (θ uz ) is the same as in Eq. (19). The 1165 other stress functions as follows (Jarvis, 1976;Zhou et al., 2006;Matsumoto et al., 2008): where c R is the radiation stress parameter fixed at 100 (Zhou et al., 2006), D 0.5 is the vapour pressure deficit halfway between 1 and c D2 set at 1.5 kPa, c D1 is the first vapour pressure parameter set at 3, and c D2 is the second vapour pressure 1175 stress parameter set at 0.1 (Matsumoto et al., 2008). Optimum temperature T opt [K] is based on elevation a.s.l. Z [m] and latitude ω [rad] (Cui et al., 2012): Graphical representations of the stress functions are presented in Fig. B1. Under unfavourable conditions where at least one of the stress functions equals zero, r s, st is assumed to be 0.58 d m −1 (50 000 s m −1 ), corresponding to the molecular diffusivity of water vapour through leaf cuticula 1185 (Tourula and Heikinheimo, 1998). If i LA is zero, no transpiration is allowed.

Appendix C Primary land-use parameters
The parameters used to describe land use include maximum 1190 and minimum leaf area index i LA,max and i LA,min , maximum and minimum plant height h max and h min , depth of the unsaturated zone (or rather active rooting depth) y uz , albedo α, minimum stomatal resistance r s, st, min and floor roughness z 0,f . Land-use parameters considered include those used in 1195 other large-scale land-surface or hydrological models (Federer et al., 1996;van den Hurk et al., 2000;van den Hurk, 2003;Zhou et al., 2006;Bastiaanssen et al., 2012), and studies of specific land-use properties (Scurlock et al., 2001;Zeng, 2001;Breuer et al., 2003;Kleidon, 2004). The range of 1200 parameters in the literature can sometimes be significant and contradictory, due to discrepancies in scale, parameter definitions, and methods of parameter estimation. The choice of land-use parameters is therefore not simply taken as a mean from the literature values investigated, but rather based on 1205 the preservation of the internal consistency of STEAM, manual calibration and priority for literature values with higher relevance. In addition, some land-use types are assumed to contain water, either as water below vegetation or as open water. The land-use parameters used in the model are shown 1210 in Table C1, and the parametrisation of water fractions are presented in Table C2.    berth et al., 2007). Table D1 provides an overview of the sensitivity of runoff 1225 and evaporation fluxes to a uniform 5 % reduction in precipitation. A number of observations can be noted. First, the mean annual STEAM runoff is clearly more sensitive (−10.95 %) to precipitation reduction compared to evaporation (−1.78 %). Second, among the evaporation fluxes, soil 1230 moisture evaporation (−2.95 %) and transpiration (−2.32 %) respond most strongly, whereas the vegetation (−0.89 %) and floor interception (−0.65 %) evaporation fluxes reduce only marginally. This is logical, because interception stocks are already small and depend more on rainfall frequency than 1235 on rainfall amount. Third, the increase in open water evaporation (+0.25 %) is small, and can be explained by decreases in vegetation interception that translated into increases in available energy for water evaporation in wetlands and rice paddies. Fourth, the relative reduction in snow accumula-1240 tion (−14.63 %) is high since snow melt is unchanged. Last, the global mean evaporative partitioning is changed only insignificantly towards lower transpiration ratio. The sensitivity of transpiration is highest over the US, Australia, the subtropical South America and Africa, and 1245 other areas that at least during part of the years are water constrained. In the wet tropics, transpiration rates do not react to precipitation reductions. Vegetation interception experiences an insignificant relative decrease, which is highest in the north and highest in the tropics. This is probably caused 1250 by a combination of lower original interception rates in the boreal forests, and the relatively higher dependence on high rainfall frequency in the tropical forests.
This uniform perturbation of precipitation forcing indicates that STEAM evaporation is much less sensitive to pre-1255 cipitation than runoff. This can be explained by the fact that evaporation is constrained by potential evaporation, which relates to other factors than just precipitation. In wet regions where soil moisture is close to saturation, any excess precipitation would more likely lead to increase in runoff rather 1260 than evaporation. The sensitivity of runoff to precipitation data is also reported in the literature (e.g., Fekete et al., 2004;Materia et al., 2010) and supports the view that runoff comparisons will not accurately describe how well land-surface models estimate evaporation when precipitation is uncertain.