Local moisture recycling across the globe

. Changes in evaporation over land affect terrestrial precipitation via atmospheric moisture recycling and 7 consequently freshwater availability. Although global moisture recycling at regional and continental scales are relatively well 8 understood, the patterns of local moisture recycling and the main variables that impact it remain unknown. For the first time, 9 we calculate the local moisture recycling ratio (LMR) as the fraction of evaporated moisture that precipitates within a distance 10 of 0.5° (typically 50 km) from its source, identify variables that correlate with it over land globally and study its model 11 dependency. We derive seasonal and annual LMR using a 10-year climatology (2008–2017) of monthly averaged atmospheric 12 moisture connections at a scale of 0.5° obtained from a Lagrangian atmospheric moisture tracking model. We find that, 13 annually, on average 1.7% (st.dev. = 1.1%) of evaporated moisture returns as precipitation locally, but with large temporal and 14 spatial variability, where LMR peaks in summer and over wet and mountainous regions. Our results show that wetness, 15 orography, latitude, convective available potential energy, wind speed, and total cloud cover correlate clearly with LMR, 16 indicating that especially wet regions with little wind and strong ascending air are favourable for high LMR. Finally, we find 17 that spatial patterns of local recycling are consistent between different models, yet the magnitude of recycling varies. Our 18 results can be used to study impacts of evaporation changes on local precipitation, with implications for, for example,


Introduction 21
Atmospheric moisture connections redistribute water from evaporation sources to precipitation sinks, affecting climates 22 globally, regionally, and locally. These connections are key in the global hydrological cycle and are used to understand the 23 importance of terrestrial evaporation for water availability. As evaporated moisture can travel up to thousands of kilometres 24 in the atmosphere, changes in evaporation can affect precipitation in a large area. An evaporationshed (Van der Ent and 25 Savenije, 2013) describes where evaporated moisture from a specific source region precipitates and therefore, can be used to 26 study (1) the changes in precipitation on a global scale following a change in evaporation in the source region and (2) 27 atmospheric moisture recycling. Globally, more than half of terrestrial evaporated moisture precipitates over land (Van der 28 Ent et al., 2010; Tuinenburg et al., 2020), which is a process called terrestrial moisture recycling. About half of terrestrial 29 precipitation originates from land (Tuinenburg et al., 2020). Hence, terrestrial moisture recycling has an important contribution 30 We study the relations between multiple variables and the 10-year climatology (2008-2017) of local moisture recycling to 129 identify factors that affect recycling. To calculate this 10-year climatology of LMR, for each month, we weighted the multi-130 year (2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017) monthly LMR by multi-year monthly evaporation in the same period: 131 = ∑ ℎ = (4) 132 in which Eyear is the sum of the evaporation of the 12 months. To identify factors that affect LMR, variables that relate to 133 atmospheric moisture and vertical displacement of air, as both higher atmospheric moisture content and ascending air promote 134 precipitation are selected. All these variables are obtained, either directly or indirectly from ERA5 reanalysis data (Hersbach 135 et al., 2020). We downscaled the original resolution from 0.25° to 0.5° by centrally averaging the data. Precipitation, which we expect to correlate positively with LMR given that in Lagrangian moisture tracking models, the amount 139 of moisture that leaves the parcel (i.e., precipitates) scales with precipitation.
(3) Total evaporation as it enhances the 140 atmospheric moisture content and we, therefore, expect it to promote precipitation locally. (4) Wetness (precipitation minus 141 evaporation), as with increasing wetness the downward flux of moisture increases and evaporated water becomes more likely 142 to precipitate, possibly promoting LMR. (4) Convective precipitation and (5) large-scale precipitation, as they scale with 143 precipitation, by definition. Both are included to study whether the type of precipitation is an important factor explaining LMR. 144 (6) Latitude, which is a proxy for processes related to the Hadley cell circulation, which is characterized by strong ascent and 145 descent of air at specific latitudes, which we expect to have an important contribution to LMR, because they respectively 146 enhance and reduce the formation of precipitation (Wang and Yang, 2022). (7) The vertical integral of the atmospheric 147 moisture flux (in northward and eastward directions and the total flux) as it carries the moisture away from its source and could 148 thus reduce LMR. (8) Convective available potential energy (CAPE), which feeds convection and therefore promotes 149 precipitation locally, which could enhance LMR. (9) Vertical wind shear between 650 and 750 hPa of both meridional and 150 zonal winds, as it affects moisture transport in multiple directions and, therefore, we expect it to impact LMR. (10) Total wind 151 speed, as it carries the wind, and therefore, we expect it to correlate negatively with LMR. (11) Total cloud cover as a proxy 152 for condensation processes which possibly enhance LMR (Richards and Arkin, 1998). (12) Boundary layer height, because 153 thinner boundaries need less evaporation to reach saturation of air, and therefore, we expect it will promote precipitation 154 locally. Finally, (13) net surface solar radiation as a proxy for the energy source of convection, and other processes, which we 155 expect to be important for LMR. We calculate shear ( ) using Equation (5). 156 In this equation, v1 and v2 are the wind speed (in zonal and meridional directions) at two different heights (z1 and z2). We 158 identified significant correlations using Spearman rank correlations. It should be noted that a correlation does not imply 159 causality. We exclude oceans, seas and Antarctica from this analysis using the land-sea mask from ERA5. We classify the data 160 6 based on latitude to account for decreasing grid cell size with increasing latitude. Each class has a range of 15° and includes 161 the grid cells on both the Northern and Southern Hemispheres (see Table A1). Between 60° and 90° south, the grid cells do 162 not contain land besides Antarctica, and are therefore not included in the classes. Additionally, we used the Ecoregions 2017 163 data (https://ecoregions.appspot.com/) to study the spatially averaged local moisture recycling of 14 biomes across the globe 164 ( Fig. A4). We study variation amongst biomes, as biomes include information on both biotic factors such as vegetation type, 165 and abiotic factors such as climate. 166 3 Results 167

LMR obtained from output of UTrack 168
Annually, on average about 1.7% (st. dev. = 1.1%) of terrestrial evaporated moisture recycles locally. LMR shows spatio-169 temporal variation ( Fig. 1) with peaks over elevated (e.g., the Atlas Mountains and Ethiopian Highlands) and wet areas (e.g., 170 Congo Basin and Southeast Asia) and minima over arid regions (e.g., Australia and the Sahara Desert). Additionally, we find 171 peaks in LMR during summer (i.e., during DJF for the Southern Hemisphere and during JJA for the Northern Hemisphere). 172 This seasonality is especially strong over mountainous and wet areas. For the mid-latitudes, especially the Mediterranean Basin 173 shows seasonality with peaks in summer (JJA). However, seasonality is largest at low latitudes. Within the tropics we find 174 some spatial differences. First, LMR in the Congo Basin and Southeast Asia exceed LMR in the Amazon Basin. Second, 175 recycling in the Congo Basin and Southeast Asia peaks in JJA and recycling in the Amazon Basin peaks in DJF, which is the 176 wet season for a large part of the Amazon.

182
We calculated recycling on a 1.5° grid using both the dataset by Link et al. (2020), which we refer to as rWAM2-layers, and the 183 dataset by Tuinenburg et al. (2020) (upscaled to 1.5°), which we refer to as rUTrack, to study the model dependency of local 184 recycling. We find that the global spatial patterns of rUTrack and rWAM2-layers agree (Fig. 2 & Fig. A5). However, the magnitude 185 of rWAM2_Layers is larger than rUTrack over mountains, the tropics, and the high latitudes. RUtrack is larger than rWAM2-layers over 186 drylands and deserts (e.g., the Sahel region and Western Asia) (Fig. 2). Globally, the difference between rUTrack and rWAM2-layers 187 and its variation is largest around the equator (Fig. A6). On average, the relative difference between UTrack and WAM2-188

Factors underlying LMR 193
For each latitude class we calculated the Spearman rank correlation coefficient (ρ) ( Table 1). Below we discuss only 194 statistically significant (p<0.05) correlations with ρ ≥ 0.4, indicating a moderate correlation. These correlations are emboldened 195 in Table 1. We find that LMR correlates positively with total precipitation (tP) and wetness (P-E) for all classes between 15° 196 and 75°. In addition, between 15° and 30°, LMR correlates strongly with tP (ρ = 0.80. Furthermore, large scale precipitation 197 (between 15° and 45° and between 60° and 75°) and convective precipitation (between 15° and 45°) correlate positively with 198 LMR. The highest correlation between LMR and convective precipitation is found between 15° and 30° latitude. Here LMR 199 also correlates positively with evaporation and CAPE, which enhances convective precipitation. Despite the low correlation 200 between LMR and CAPE for most of the latitude classes, high CAPE clearly relates to LMR, as the skewed profile in the 201 scatter density plot indicates that only a small amount of the grid cells with a relatively high CAPE have a low LMR (Fig 3). 202 Furthermore, the presence of clouds also correlates with LMR. Between LMR and total cloud cover, a positive correlation 203 holds between 15° and 45°, and a negative correlation holds between 60° and 75°. The vertical integral of the eastward and 204 northward moisture fluxes correlate less with LMR compared to vertical fluxes (e.g., precipitation) as for the higher latitudes, 205 8 the northward moisture flux correlates positively with LMR (between 60° and 75°) and the eastward moisture flux correlates 206 negatively with LMR (between 75° and 90°). However, wind speed correlates negatively with LMR for the lower latitudes 207 (between 0° and 45°). Furthermore, LMR correlates positively with orography between 30° and 75°. We find that for high 208 elevation, LMR is always relatively high ( Fig A7). Additionally, LMR correlates negatively with boundary layer height 209 between 45° and 60°. Finally, LMR correlates negatively with shear at 650 hpa in the meridional direction (between 75° and 210 90°) and latitude (between 60° and 75°), However, we find an oscillating relation between LMR and latitude (Fig 4), which is 211 not captured by the Spearman rank correlation coefficients. This pattern indicates high values of LMR over the equator (0°) 212 and 60° north, and low values around 30° north and south. Orography seems to disrupt the relation between latitude and LMR 213 causing peaks in LMR around 35° north and 20° south (Fig 4). LMR does not correlate to surface net solar radiation for any 214 latitude. However, for low surface net solar radiation (<0.75*10 6 J/m 2 ) holds that LMR increases with increasing surface net 215 solar radiation (Fig 3).

Factors underlying LMR 231
Moisture recycling affects humanity by influencing water security, agriculture, forestry, regional climate stability and and which factors affect it. We find that LMR, defined as the fraction of evaporated moisture that precipitates within a distance 236 of 0.5° (typically 50 km) from its source, varies over time and space, peaking in summer and over elevated and wet regions. 237 First, we identified latitude, elevation, and Convective Available Potential Energy (CAPE) as important factors influencing 238 LMR (Fig. 5). These variables all promote convection (Roe, 2005; Scheff and Frierson, 2012; Wallace and Hobbs, 2006), 239 strongly suggesting a dependency of LMR on convection. Convective storms develop due to unstable conditions resulting in 240 precipitation locally (Eltahir, 1998) and a higher CAPE results in more rainfall (Eltahir and Pal, 1996;Williams and Renno, 241 1993). The pattern of LMR across latitudes also coincides with updraft and downdraft of air caused by the Hadley cell 242 circulation (Wallace and Hobbs, 2006). Around the equator and 60° north and south, air ascends, where we find a high LMR. 243 Around 30° north and south, air descends, where we find a low LMR. Deviations from this pattern correspond to higher 244 elevations which promote LMR through orographic lift. Overall, our results suggest a positive relation between convection 245 and LMR. 246

250
Second, we find that wetness is an important factor underlying LMR as LMR significantly correlates with precipitation and P-251 E (precipitation minus evaporation). Furthermore, both large-scale and convective precipitation significantly correlate with 252 LMR. This is surprising, as convection promotes precipitation locally (Eltahir, 1998); therefore, we expected a stronger 253 correlation between LMR and convective precipitation than between LMR and large-scale precipitation. As both correlations 254 are similar, this suggests that the type of precipitation does not affect LMR. Although convection is a local-scale process (i.e., 255 having a spatial scale below 100 km) (Miyamoto et al., 2013), remotely evaporated moisture can be transported to a region 12 with high convective activity and then precipitate as convective precipitation (Jana et al., 2018;Liberato et al., 2012). In that 257 way, the precipitation type is independent of the distance between moisture source and target location and therefore does not 258 relate to LMR. Total cloud cover correlates both positively (between 15° and 45°) and negatively (between 60° and 75°) with 259 LMR. Total cloud cover correlates with precipitation, convective precipitation, and large-scale precipitation for all latitudes 260 except between 60° and 75° (Tab. A2). Due to the positive correlation between LMR and precipitation and the absence of a 261 correlation between precipitation and total cloud cover at these latitudes we can statistically explain the negative correlation 262 between total cloud cover and LMR. Physically, this result is harder to explain. Our results describe the importance of 263 convection underlying LMR at lower latitudes, where total cloud cover correlates with convective precipitation. For higher 264 latitudes, the importance of convection underlying LMR decreases, and we therefore expected also the correlation between 265 total cloud cover and LMR to decrease but not to become negative. Likely, another process that we cannot identify with our 266 analysis causes the correlation between total cloud cover and LMR to be negative. Overall, we find that wetness enhances 267 LMR independent of the precipitation type. 268 269 Unexpectedly, we do not find a clear correlation between the vertical integral of the atmospheric moisture flux and LMR. 270 However, for the lower latitudes (between 0° and 45° latitude), LMR correlates to wind speed (at 10 and 100 m) which carries 271 evaporated moisture away from its source location, enhancing the moisture flux. Therefore, horizontal moisture fluxes at 272 specific altitudes are better for our analysis than the vertical integral of the moisture flux. However, since wind carries moisture 273 away from its source, we expected that wind speed and LMR would also correlate for the higher latitudes (latitude above 45°). 274 It could be that for the higher latitudes, a more significant amount of moisture is present at higher latitudes, explaining why 275 LMR and wind at 10m do not correlate. However, wind speeds at 650 hpa and 750 hpa also do not correlate to LMR for these 276 latitudes (Tab. A2). 277 278 Despite the importance of vertical shear in atmospheric moisture tracking models (Van der Ent et al., 2013), we do not find a 279 correlation between local moisture recycling and vertical shear between 650 and 750 hPa. Shear is the friction between air 280 layers that minimizes complete mixing, which for some regions around the world is strongest between 650 and 750 hPa 281 (Dominguez et al., 2016). A possible explanation is that due to its small spatial scale, the temporal scale of LMR is also small, 282 which may prevent the air reaching 700 hPa within the spatial scale of LMR. Furthermore, it is possible that our study design 283 is insufficient to capture the relation between LMR and shear throughout the year over the globe. We aimed for a general 284 analysis to identify the main factors that influence LMR. A more detailed study that distinguishes between different seasons 285 and isolates different climate zones is necessary to identify more factors that influence LMR as some factors might be more 286 important during a specific season. For example, convection occurs more during summer than during winter, and therefore, 287 might have a stronger correlation with LMR during summer. Besides, some factors are shape and size dependent similar to 288 LMR, while other factors are not dependent on grid cell size and shape. This might cause bias in the results of the Spearman 289 analysis. Furthermore, due to the many interactions within the Earth system and, consequently, between the variables included 290 13 in our study, it is impossible to determine the true drivers of LMR. However, the correlations do indicate how changes in the 291 environment might affect LMR. 292

regional patterns 293
To zoom in on the importance of each of the different factors underlying LMR for various areas across the globe, we determined 294 LMR for the major global biomes (Fig. A8). LMR is highest for the wet tropics (between 0° and 15° north and south) and 295 montane grasslands and lowest for desert-like biomes in both the Northern and Southern Hemisphere (between 30° and 45° 296 north and south), confirming the importance of wetness, orography, and latitude. However, in the tropics (between 0° and 15° 297 latitude), we do not find any correlation between LMR and precipitation, evaporation, wetness, or orography. Possibly, due to 298 the abundance of water and energy to evaporate, there is LMR under all circumstances, except for when the wind speed is 299 high. Comparing LMR for each biome between both hemispheres indicates that some of the factors underlying LMR are more 300 robust than other ones for some biomes. In the Mediterranean biomes, located between 30-40° north and south, air generally 301 descends due to the Hadley cell circulation. As a result, these biomes are expected to have low LMR. Although we find a low 302 LMR for the Mediterranean biomes in the Southern Hemisphere, we find a relatively high LMR for the Mediterranean biomes 303 in the Northern Hemisphere. The Spearman rank analysis indicates that at these latitudes, wind speed correlates with LMR, 304 which may explain the difference between both hemispheres..

The spatial scale of the local moisture recycling ratio 318
We study local moisture recycling on a spatial scale of 0.5°, which is approximately 55 km around the equator and 50 km on 319 average globally for all land cells. Instead of recycling within one grid cell (r1), we studied the recycling of evaporated moisture 320 within its source grid cell and its 8 surrounding grid cells. Compared to r1, this r9 includes all moisture flows with a length 321 14 scale of typically 50 km. For r1, moisture flows with a length smaller than 50 km can occur close to the border of grid cells 322 and therefore, r1 by definition underestimates the actual recycling. These moisture flows are accounted for in r9. 323 324 However, defining LMR on a grid scale gives complications. First, the longitudinal distance for a grid cell size decreases with 325 latitude, resulting in different sizes and shapes, which makes it difficult to compare LMR among all grid cells. For the low-326 and mid-latitudes, the variation in grid cell size affects LMR only slightly, as confirmed when LMR for each grid cell was 327 scaled to a single area (Fig. A9). Therefore, we believe that the variation in grid size causes only a small bias in the statistical 328 analysis, as the largest fraction of the land surface is at the low-and mid-latitudes, and moisture recycling is less important for 329 the higher latitudes. However, it should be noted that for similar wind speed, LMR will be lower in smaller grid cells than 330 larger grid cells. Second, the spatial scale of recycling is strongly dependent on regional differences such as biome type, the 331 dominating winds, and the proximity to mountains. For instance, with increasing distance to the Andes mountains the median 332 making it unknown at what spatial scale moisture recycling is the dominant process for precipitation. Therefore, we believe 336 that a grid-based approach to systematically study LMR globally is a solid approach to define and study the physical processes 337 at a spatial scale >50 km through, for instance, the Spearman analysis to study the underlying processes. However, our 338 definition of LMR is not sufficient to identify processes on a spatial scale smaller than 50 km that might be relevant. 339

Model and definition dependencies 340
It is important to note that the typical length scale of moisture recycling, as defined by Van der Ent & Savenije (2011), allows 341 for a comparison of regional moisture recycling for different regions around the world due to its independence of the region's 342 size and shape (Fig A10). The typical length scale of evaporated moisture recycling decreases with increasing recycling. It 343 peaks over deserts and is small over the tropics and mountainous regions (Fig A9), overlapping with the spatial pattern of 344 LMR. However, this typical length scale does not allow for the quantification of the amount of recycled moisture and therefore, 345 it is difficult to apply this metric to study the impact of evaporation changes due to land-use change. Therefore, studies that 346 aim to quantify moisture recycling locally may best use recycling ratios. However, studies that aim to compare recycling 347 among different regions can best use the typical length scale of recycling. WAM2-layers and found a similar spatial pattern with high recycling over mountainous and tropical regions and low recycling 354 over desert-like regions. These recycling ratios also have a larger magnitude than LMR. However, it is not straightforward to 355 interpret the differences in recycling ratios as both models use different input data (i.e., ERA5 and ERA-Interim). To assess 356 the possible role of the models in causing the difference in moisture recycling, we describe the main differences between the 357 models. First, WAM2-layers calculates the atmospheric moisture recycling on a larger temporal and spatial scale than UTrack, 358 A larger grid cell size and time step increases the likelihood of evaporation and precipitation taking place within the same 359 small amount of time, which might result in an overestimation of recycling within one grid cell. Second, WAM2-layers 360 generates moisture flows using two vertical layers; therefore, strong winds at specific vertical levels will be described in less 361 detail, reducing estimated moisture transport and enhancing estimated moisture recycling within a single grid cell. Differences 362 between rUTrack and rWAM2-layers are highly visible over mountainous regions where wind experiences relatively strong friction, 363 highly impacting the wind. Finally, different approaches are used to include vertical mixing in the two models. some previous studies used different methods to calculate regional moisture recycling for a specific area, such as isotope 369 measurements (e.g., An et al., 2017) and bulk recycling models (e.g., Burde & Zangvil, 2001). The most common recycling 370 models are modifications of Budyko's model (Budyko, 1974;Burde and Zangvil, 2001), which are 1D or 2D analytical models. 371 These models assume that the atmosphere is completely mixed, meaning that evaporated water directly mixes perfectly with 372 advected water throughout the entire water column. Because of this assumption, first, these models overlook fast recycling, 373 which describes local showers that yield precipitation before the evaporated water is fully mixed. Excluding fast recycling 374 causes models to underestimate terrestrial moisture recycling for some regions (e.g., Amazon Basin) (Burde et al., 2006b). 375 Second, these models ignore the influence of vertical shear, which causes a significant error (Dominguez et al., 2020). Our 376 method minimises the errors due to fast recycling and vertical shear because of two model aspects. First, at each time step, 377 each parcel has a small chance of getting mixed, causing each parcel to move approximately once in the vertical direction 378 every 24 hours, additional to the displacement based on reanalysis data of vertical winds. This process minimizes complete 379 mixing and reduces the error due to shear and fast recycling. Second, the error due to fast recycling also becomes smaller 380 because lower atmospheric levels contribute more to the total precipitation than higher levels due to the skewed vertical 381 moisture profile. WAM2-layers accounts for vertical shear as it models two vertical atmospheric layers of which the interface 382 is located at the height at which shear typically occurs. These two layers are both completely mixed and therefore, compared 383 to bulk models, WAM2-layers better represents the distribution of moisture throughout the atmospheric column. As an 384 alternative method, moisture flows can be calculated on a smaller time step to increase the interactions between different wind 385 components, resulting in a better representation of turbulence (Keune et al., 2022). Despite the error reduction, the 386 representation of fast recycling in UTrack should be studied in more detail, as fast recycling is expected to influence LMR 387 significantly. 388 LMR is calculated as a ten-year average. This period of ten years might miss multi-year climate variability such as the El Niño 389 Southern Oscillation and the North Atlantic Oscillation. The time series of atmospheric moisture connections provided by Link 390 et al. (2020) allowed to study inter-annual variation in relatively local recycling. This shows that recycling is dependent on 391 multi-year atmospheric phenomena. During the major El Niño event of 2015-2016, the northeast of South Africa had a lower-392 than-average local recycling ratio (Fig. A11) for 2015. This pattern coincides with the impact of wetness during El Niño years, 393 consistent with the hypothesis that wetness enhances LMR. Furthermore, strong events such as heat waves and droughts might 394 affect the multi-year annual mean. For example, we clearly find lower recycling over Russia during 2010, which may relate to 395 the 2010 heatwave in eastern Europe and Russia. Overall, for these multi-year and strong events we find that, for regions that 396 face wetter-than-normal conditions, LMR is enhanced, and for regions that face drier-than-normal conditions, LMR is reduced. 397 Hence, drought events might result in a decrease in LMR as seen for the 2010 heat wave event in Europe and Russia. However, 398 not for all inter-annual climate variability modes we find a clear impact on moisture recycling. It may be that these phenomena 399 do not affect wetness throughout the entire year, and therefore, annual means might not represent them well. indicating that only a small amount of the evaporated moisture returns as precipitation locally. For irrigated agriculture in 412 regions that are characterized by a high LMR, a relatively large amount of the evaporated water returns to its source, which 413 reduces the amount of water that is necessary for irrigation. Terrestrial evaporation is an important source for precipitation and 414 freshwater availability (Keune and Miralles, 2019). Therefore, spatial planning using LMR might improve agricultural water 415 management. 416 417 Global climate change likely affects atmospheric moisture connections due to changes in atmospheric dynamics. For example, 418 due to global warming, tropical atmospheric circulation may weaken (Vecchi et al., 2006), and the Hadley cells may move 419 poleward (Shaw, 2019), which will affect the updraft and downdraft of air around the globe, which we found to be important 420 processes underlying LMR. Furthermore, climate change has different opposing impacts on storm tracks which have an 421 important role in moisture transport by transporting latent heat poleward (Shaw et al., 2016). Furthermore, in a warmer climate 422 continental recycling is predicted to decrease and precipitation over land would be more dependent on evaporation over the 423 ocean (Findell et al., 2019). However, our study does not account for any impacts of climate change. As our results indicate 424 that wetness and convection enhance LMR, LMR may change due to, for example, drying and wetting of regions, changes in 425 Hadley cell circulation, and circulation in the tropics. Furthermore, climate change enhances the risk of droughts (Rasmijn et 426 al., 2018;Teuling, 2018) and LMR might be used to study drought resilience globally. As for a high LMR a local drought 427 might drastically impact the local water cycle. 428

429
We expect that LMR can be helpful also in other ways. Specifically, we expect the concept of LMR can be used to study how 430 changes in evaporation, due to for example afforestation, affect the local water cycle beyond merely a loss of moisture. Overall, LMR gives us better insight into the atmospheric part of the local water cycle and terrestrial evaporation as a source 438 for local freshwater availability. 439

Conclusions 440
We calculated the local moisture recycling ratio (LMR) from atmospheric moisture connections at a spatial scale of 0.5°. LMR 441 is the fraction of evaporated moisture that precipitates within a distance of 0.5° (typically 50 km) from its source. On average, 442 1.7% (st.dev. = 1.1%) of global terrestrial evaporation returns as precipitation locally, with peaks of approximately 6%. LMR 443 peaks in summer and in wet and elevated regions. We find that orography, precipitation, wetness, convective available potential 444 energy, and wind affect LMR. In addition, latitude correlates with LMR, which likely indicates the importance of the ascending 445 air and descending air related to the Hadley cell circulation. Furthermore, by comparing LMR calculated using different models 446 we found that the spatial pattern of LMR is not model-dependent, yet, the magnitude of LMR is strongly dependent on the 447 model. LMR defines the local impacts of enhanced evaporation on precipitation and thus its role as a source for local freshwater 448 availability. Therefore, LMR can be used to evaluate which locations may be suitable for regreening without largely disrupting 449 the local water cycle.

Code availability 506
The code that was used to calculate the local moisture recycling ratio and for the analyses is available from the corresponding 507 author upon reasonable request. 508

Data availability 509
The atmospheric moisture connections from Tuinenburg et al., (2020)