Climatic expression of rainfall on soil moisture dynamics in drylands

In drylands, characterised by water scarcity and frequent meteorological droughts, knowledge of soil moisture dynamics and its drivers (evapotranspiration, soil physical properties and the timing and sequencing of precipitation events) is fundamental to understanding changes in water availability to plants and human society, especially under a nonstationary climate. Given the episodic and stochastic nature of rainfall in drylands and the limited availability of data in these regions, 15 we sought to explore what effects the temporal resolution of precipitation data has on soil moisture and how soil moisture distributions might evolve under different scenarios of climate change. Such information is critical for anticipating the impact of a changing climate on dryland communities across the globe, especially those that depend on rainfed agriculture and groundwater wells for drinking water for humans and livestock. A major challenge to understanding soil moisture in response to climate is the availability of precipitation datasets for dryland regions across the globe. Gridded precipitation data may only 20 be available for daily or weekly time periods, even though rainstorms in drylands often occur on much shorter time scales, but it is currently unknown how this timescale mismatch might affect our understanding of soil moisture. Numerical modelling enables retrodiction or prediction of how climate translates into dynamically evolving moisture within the soil profile. It can be used to explore how climate data at different temporal resolutions affect these soil moisture dynamics, as well as to explore the influence of shifts in rainfall characteristics (e.g., storm intensity) under potential scenarios of climate change. This study 25 uses Hydrus 1-D, to investigate the dynamics of soil moisture over a period of decades in response to the same underlying rainfall data resolved at hourly, daily, and weekly resolutions, as well as to step changes in rainfall delivery, which is expected under a warming atmosphere. We parameterised the model using rainfall, evaporative demand, and soils data from the semiarid Walnut Gulch Experimental Watershed (WGEW) in SE Arizona, but we present the results as a generalized study of how rainfall resolution and shifts in rainfall intensity may affect dryland soil moisture at different depths. Our results indicate that 30 hourly or better rainfall resolution captures the dynamics of soil moisture in drylands, and that critical information on soil water content, moisture availability to vegetation, actual evapotranspiration, and deep percolation of infiltrated water is lost when soil moisture modelling is driven by rainfall data at coarser temporal resolutions (daily, weekly). We further show that https://doi.org/10.5194/hess-2021-48 Preprint. Discussion started: 4 February 2021 c © Author(s) 2021. CC BY 4.0 License.


Introduction
In dryland regions of the world, water is inherently scarce, and there are tight relationships between rainfall and soil moisture that have implications for the water balance and specifically for groundwater recharge, agriculture, and natural vegetation. 40 Given the brevity of storms and high potential evapotranspiration, it is challenging to understand the influence of rainfall on soil moisture in drylands, yet this information is critically needed within drought-prone dryland regions, where livelihoods are coupled to the regional expression of climate Davenport et al., 2019). There is a disconnect between the spatial and temporal resolution of globally available rainfall data and the local characteristics of precipitation and its translation into soil moisture and water availability to vegetation within dryland regions. Drylands are characterised by low mean annual 45 precipitation totals, but is expressed through rainstorms that are typically intense, short-lived, and spatially restricted (Nicholson, 2011). Dryland regions, which comprise 40 % of total global land mass, are characterised by extremely scarce water resources and limited in situ data on weather and hydrological fluxes (Nicholson, 2011). These regions tend to have sparse precipitation gauge data, so gridded datasets are typically used to understand the relationships between climate variables and soil moisture, even though they may not preserve the inherent intensity characteristics during individual rainstorms, rainfall 50 sequencing, or the time series of the driving evaporative demand. Globally available (gridded) precipitation data are typically resolved at daily, weekly, and monthly temporal resolutions, and all are in common use for understanding historical expressions of climate into the water cycle, yet it is unclear what effect such aggregated timeframes may have on estimates of soil moisture (and the overall water balance) for dryland regions.

55
Given the strong feedbacks between rainfall intensity and inherent soil properties such as infiltration capacity, it is not clear how well rainfall data resolved at progressively coarser temporal resolutions might affect the prediction of soil moisture. For example, a 30-minute storm with an intensity of 80 mm/h, would yield 40 mm of rainfall in an hour. Assuming a constant infiltration rate of 20 mm/h, half of this rain would infiltrate into the top layer of soil within an hour (and the other half would pond or run off). However, if this rainstorm was the only rain occurrence that day and precipitation was summed over a day, 60 it would yield 40 mm/day or 1.7 mm/h. This bias towards light, constant drizzle becomes even more extreme for weekly data (0.24 mm/h), and it is clearly not reflective of most dryland rainfall. This so-called 'drizzle effect' is a problem that limits how well the climate system is represented at the land surface, and it has already been identified as a major challenge in climate modelling (Pendergrass and Hartmann, 2014;Stephens et al., 2010). However, it is less well appreciated how this drizzle effect might manifest within historical gridded precipitation datasets (based on gauge data, remote sensing, and/or reanalysis climate 65 https://doi.org/10.5194/hess-2021-48 Preprint. Discussion started: 4 February 2021 c Author(s) 2021. CC BY 4.0 License. modelling) and its broader implications for studies of land surface hydrology that use these data. In this paper, we explore how using rainfall data resolved at different temporal resolutions (hourly, daily, weekly) affect estimates of soil moisture and water availability to plants in dryland regions. We also explore how changes to the delivery of rainfall in the future might affect soil moisture. If precipitation during individual events becomes more or less intense (Trenberth, 2011;Trenberth et al., 2017), it could have major implications for soil moisture, especially in dryland regions, but this effect is currently unknown. We suggest 70 that one of the most critical variables required to understand and predict the soil moisture profile in drylands is rainfall temporal resolution, which is often coarser than the inherent expression of rainfall for hydrological processes.
In order to address these issues, we created decadal timeseries of rainfall inputs at three temporal resolutions (hourly, daily, weekly) with the same seasonal and annual rainfall totals, based on high resolution, long-term data from a densely gauged 75 watershed in SE Arizona. We use these synthetic time series to drive Hydrus-1D, a model based on Richards equation for Darcian flow in unsaturated conditions (Šimůnek et al., 2012), and explore model outputs of soil moisture, actual evapotranspiration, and deep percolation of infiltrated water under different scenarios.

Study Site 80
Our goal was to evaluate the effects of temporal rainfall resolution on soil moisture that could be used to inform modelling strategies for data-sparse dryland regions. To accomplish this, we needed a data-rich site, where we could reasonably evaluate conversion of rainfall into soil moisture. The study was carried out using data from Walnut Gulch Experimental Watershed (WGEW), a 150 km 2 semi-arid watershed in SE Arizona, as a proxy for a testbed to explore the impact of rainfall inputs on soil moisture (Figure 1). WGEW is a typical semi-arid region with brush and grass cover, an elevation ranging from 1250 to 85 1585 m, and annual average precipitation of 350 mm (Renard et al., 2008). Within WGEW, we focused on a location in the Kendall grasslands, where there was available rainfall data from a dense network of rainfall gauges ( Figure 1) with per minute resolution spanning multiple decades (Goodrich et al., 2008;Renard et al., 2008), soil moisture data from a mesoscale cosmogenic neutron apparatus (details below). This site is a typical hot arid climate region with a Köppen-Geiger BWh climate classification and an aridity index between 0.05 and 0.20. This area is characterised by high evaporative demand and intense 90 convective rainstorms during the summer monsoon, which dominate annual runoff in the basin (Osborn and Lane, 1969;Osborn et al., 1979;Goodrich et al., 2008).

Modelling Strategy
In this study we used the Hydrus-1D model v4.17 (Šimůnek et al., 2012) to simulate soil moisture dynamics in response to different rainfall resolutions (hourly, daily, weekly) over a multi-decadal period (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018)(2019). Hydrus-1D has been applied to understand and track water flow through porous media and it has been extensively tested and successfully applied worldwide 100 in many different conditions. The model is driven by time series of rainfall and potential evapotranspiration (PET), from which the water balance is solved numerically to determine net rainfall, infiltration into the upper soil layer, actual evapotranspiration, and drainage below the bottom layer ( Figure 2). Hydrus-1D solves the Richards equation, the unsaturated form of Darcy's Law, to quantify movement of water through the soil, between soil and the atmosphere (via evapotranspiration), and the percolation of soil water into the groundwater aquifer. The model requires parameterisation of soil properties that control water 105 flow through the soil (e.g., hydraulic conductivity), which can be derived from soil texture (van Genuchten, 1980). Details on the model development and theoretical underpinning can be found elsewhere (Šimůnek et al., 2012).

Input Data
We used the data-rich WGEW as a representative testbed for understanding the impacts of rainfall resolution and rainfall intensity on soil moisture in dryland regions. Event precipitation data were acquired from rainfall digital gauging station (No. 82) within WGEW (Figure 1;  and to depths of order ~10 -1 m using cosmogenic neutrons. COSMOS is a mesoscale measure of soil moisture based on an area and depth averaging. Soil moisture from cosmic-ray neutron sensors from Kendall site was obtained directly from the COsmic-ray Soil Moisture Observing System (COSMOS) (Zreda et al., 2012). Typical corrections to measured neutron counts 120 related to variations of atmospheric pressure and water vapor, and solar intensity were applied following guidelines in the literature (e.g., (Rosolem et al., 2013;Schrön et al., 2017)). In addition, the cosmic-ray neutron sensor at Kendall was calibrated against independent set of soil samples taken within the sensor footprint as part of its original deployment and all information is available in the COSMOS network (e.g., dry soil bulk density, soil lattice water, soil organic carbon, and sensor calibration coefficient N0). Additional data quality control steps were taken by removing unrealistic estimates of soil moisture by the 125 cosmic-ray neutron sensor that fell either below zero volumetric (m 3 m -3 ) soil moisture content or above the site-specific https://doi.org/10.5194/hess-2021-48 Preprint. Discussion started: 4 February 2021 c Author(s) 2021. CC BY 4.0 License.
To obtain rainfall intensity per event from the historical WGEW rainfall record, we divided reported event precipitation depth (mm) by event duration (min), and then aggregated the resulting set of events into hourly precipitation data ( Figure 3). We subsequently aggregated this synthetic hourly time series of precipitation into daily and weekly series, ensuring that the total rainfall volume is the same in each time series (Figure 4). The temporal spreading of rainfall data from hours to weeks has a very noticeable impact on the mean values of precipitation (Figures 4a-4c) and on the peak magnitudes of events . Although the WGEW rainfall record goes back to the 1950s, we modelled precipitation inputs from the year 2000, in order to ensure that the input rainfall series coincide with the available reference crop evapotranspiration (ETO) data (a measure of potential evapotranspiration).  (1) where Δ is the slope of the saturation vapor pressure-temperature (kPa o C -1 ), Rn is net radiation calculated at the crop surface in MJ m -2 h -1 , γ is psychrometer constant (kPa o C -1 ), u2 is average hourly wind speed measured at 2 m above ground level (m s -1 ), T, is average air temperature at 1.5 m above ground level ( o C), ea and es are saturation vapor pressure and mean actual vapor pressure respectively all measured at 1.5 m from the ground level (kPa). 165 where k represents the extension coefficient for solar radiation at global scale, and is a function of angle of the sun, leaf arrangement and distribution of plants. Here we adopted k of 0.463, as in previous work (Tuttle and Salvucci, 2016;Šimůnek et al., 2012;Chen et al., 2014). The partitioned hourly ETO values were used as model input to compute the water balance and were kept constant over all simulations, regardless of the timescale or intensity of the input precipitation. The model is spatially 180 discretised in 1cm layers and we integrate soil moisture outputs over 0-30 cm, 30-50 cm and 50-100 cm. At the bottom of the soil, we used a free drainage water flow boundary (zero-gradient) condition to simulate a freely draining soil profile, assuming the water table is deep enough to not influence water redistributions in the soil profile. Hydrus-1D provides various relevant water balance outputs including soil moisture at multiple depths through the soil profile, actual evapotranspiration, and free drainage, which gives information on deep percolation of infiltrated rainfall that will likely contribute to groundwater recharge. 185 Rainfall gauging station number 82 at WGEW lies in the Elgin-stronghold soil complex (https://www.tucson.ars.ag.gov/dap/1993soils.htm). By averaging 9 data points covering the whole Elgin-stronghold soil complex within the Kendall, we obtained averaged soil texture of ~79 % sand, ~11 % silt, and ~9 % clay (Breckenfeld et al., 1993). We used the neutral network prediction module in Hydrus-1D, based on the Rosetta model (Schaap et al., 2001), to 190 determine soil water retention and suction parameters required to compute soil moisture. To parameterise saturated hydraulic conductivity (Ks), we used published data from rainfall simulation experiments in WGEW (Wainwright et al., 2000). The soil for the Kendall site is represented in Hydrus-1D as a uniform soil profile with parameters shown in Table 1. 195

Climate Change Scenarios
In order to examine the potential effects of climate change on rainfall and soil moisture in a dryland region, we utilised the were robust, we produced 10 realisations for each grid location per climate scenario, equivalent to 200 years of simulation 215 time per scenario.

Model evaluation
Simulated soil moisture saturation (θs) values in units of percent were compared with θs from COSMOS over the period 2010-2014 ( Figure 6). In general, we found good agreement between modelled and observed θs and the simulated and observed 220 distributions. Although the distributions are statistically different based on the Kolmogorov-Smirnov statistic (p< 2.2e-16), due to the larger range of the COSMOS data ( Figure 6), the Mann-Whitney U test showed the medians of COSMOS and Hydrus-1D output are statistically similar (p=0.5774). It is clear that the Kendall soil is very dry with a median value of ~18 % saturation, but there are clearly short-lived wet periods wherein high rainfall produces spikes in soil moisture (represented as outliers in both simulated and observed distributions of θs in Figure 6). From this analysis, we concluded that Hydrus-1D 225 simulations are broadly representative of measured soil moisture at our study site, enabling further investigation of the effects of temporal rainfall distribution and intensity.

Effects of rainfall temporal resolution on soil moisture 230
We investigated the impact of changing rainfall resolution in terms of the overall distribution of rainfall inputs over multiple decades. Based on statistical analysis via the Kolmogorov-Smirnov test, we found highly significant differences between hourly v. daily, hourly v. weekly, and daily v. weekly rainfall (Figure 4; p<<2.2e-16). This suggests that deriving rainfall from gridded data products resolved at different temporal scales can dramatically affect the input time series of rainfall, even if the overall seasonal and annual precipitation totals are the same. 235 We subsequently analysed the impacts of temporal averaging hourly precipitation over coarser timescales on the resulting soil moisture. The output histograms of θs for the three input rainfall resolutions (hourly, daily, weekly) and for the three depth intervals are shown in Figure 7. They illustrate that strong differences in the temporal expression of rainfall have a large impact on the resulting soil moisture for all depth intervals. At each soil profile layer, the soil moisture values produced by the three 240 rainfall resolutions are highly significantly different from each other (p<2.2e-16) for both comparisons of distributions (Kolmogorov-Smirnov) and their medians (Mann-Whitney), where the median declines with increasing time interval of rainfall aggregation.
Another interesting aspect of these plots is how shifts in soil moisture distributions arising from different temporal resolutions 245 of input rainfall affect the time below wilting point (WP), which has important implications for plant (and crop) water availability. The frequency of values that cross below the WP threshold progressively increases (lower soil moisture) with increasing temporal resolution of rainfall, moving from hourly to weekly (Figure 7a-c). These are important results, which were anticipated based on the marked differences between distributions of input rainfall (Figure 4). They indicate the 'drizzle effect' in which temporal aggregation leads to a steady supply of precipitation to the land surface at very low intensity, is 250 counteracted by actual evapotranspiration losses for the upper 30 cm of soil, leading to a drier soil column (Figure 7a). This phenomenon underlines the importance of resolving rainfall at the appropriate temporal scale to characterise the infiltration and resulting soil moisture and water availability to relevant vegetation at different rooting depths. Remarkably, the weekly and daily input rainfall distributions yield soil moisture values for the upper 30 cm of soil that are below the WP 7-8 % more The amount of P lost via AET decreases as we move from weekly to hourly rainfall resolutions ( Figure 10). For example, over the period 2000-2019 at WGEW, hourly precipitation produced cumulative AET of 3720 mm, while the daily simulation 270 generated ~10 % more AET, followed by the weekly simulation with 28% more AET (Fig.10). The deep percolation below the model domain decreased with reduced rainfall resolution from the hourly cumulative value of 2146 mm to ~18 % less for https://doi.org/10.5194/hess-2021-48 Preprint. Discussion started: 4 February 2021 c Author(s) 2021. CC BY 4.0 License. daily rainfall, and 49 % less for weekly rainfall Out of the cumulative P of 5945 mm that enters the water balance, AET tends to be more than double the value of drainage for all the three resolutions (Fig.10).
275 Figure 8: Cumulative AET (a) and cumulative drainage (b) produced by modelling the three rainfall resolutions. The cumulative P for all the three resolutions is 5945 mm. The AET for hourly, daily, and weekly resolutions when expressed as a percentage of total precipitation is respectively is 62 %, 69 % and 80 % while the percentages for drainage when expressed as percentage of P is 36 %, 30 % and 18 %, respectively.

Climate change scenarios and rainfall characteristics 280
Based on the understanding gained from changing input rainfall resolution, we further explored the impact of hypothetical scenarios of climate change on soil moisture using only hourly timeseries of rainfall, since this temporal resolution well captures the influence of dryland rainfall characteristics on soil moisture ( Figure 6). The two climate scenarios investigated here (increased and decreased rainfall intensity with no change to seasonal or annual precipitation totals) are distinct from the 'historical' baseline timeseries generated stochastically from rain-gauge data in WGEW over the 20-year simulation period 285 ( Figure 8). Rainfall intensities, in terms of both distributions (Kolmogorov-Smirnov) and medians (Mann-Whitney), are significantly different between the three climate scenarios (p<2.23e-308), for example, where the 'decreased storminess' scenario produces more storms (at lower intensity) and the 'increased storminess' scenario produces fewer, but more intense storms relative to the 'historical' baseline timeseries (Figure 8).  Figure 9: Empirical cumulative distribution functions (eCDFs) of 'historical' (black), 'decreased storminess' (yellow) and 'increased storminess' (red) climate scenarios precipitation intensities.

290
The hourly rainfall timeseries for the baseline (historical) and the two climate change scenarios were used to examine the relative impacts of these differences in overall rainfall intensity characteristics on soil moisture dynamics using Hydrus-1D. 300 The modelled soil moistures using the three climate scenarios were all significantly distinct from one another (Table 2).
Relative to the historic 'baseline' conditions, the distribution of soil moisture produced by the 'decreased storminess' scenario is generally lower (skewed to the left) for all soil depth intervals, while it is higher (skewed to the right) for the 'increased storminess' scenario ( Figure 10). The differences between the distributions become larger with soil depth. For example, the differences in time below the wilting point between the 'decreased storminess' simulation and the other two simulations 305 increases sharply in the 30-50 cm interval, and again for the 50-100 cm interval. It is clear that the 'historical climate' and 'increased storminess' simulations generate significant wetting fronts that persist at deeper soil depths, such that there is a minority of the distribution below the wilting point (Figs 10a and 10b). Additionally, we detected a crossover between the https://doi.org/10.5194/hess-2021-48 Preprint. Discussion started: 4 February 2021 c Author(s) 2021. CC BY 4.0 License.
'historical climate' and 'increased storminess' scenarios in the 50-100 cm depth interval, whereby the historical distribution of θs exceeds that of the decreased storminess up to the 15th percentile, above which the 'increased storminess' distribution of 310 θs for the higher intensity rainstorms yields an overall wetter soil ( Figure 10). 30-50 cm, and (c) 50-100 cm. The top 0-30 cm of the soil profile is below WP 71 %, 94 % and 62 % of the time, respectively for historical, decreased storminess, and increased storminess, respectively. At the 30-50 cm soil depth, θs was below WP 41 %, 88 % and 23 % of the time for historical, decreased storminess, and increased storminess, respectively. The bottom 50-100 cm interval of the soil had θs values WP 30 %, 56% and 18 % for historical, decreased storminess, and increased storminess respectively during the whole simulated period.

320
Partitioning of precipitation into AET and deep percolation (drainage below the model boundary) for these climate change scenarios supports the expectation of lower (higher) cumulative loss of water through evapotranspiration for increased (decreased) storminess. Essentially less intense rainfall infiltrating into upper soil layer tends to be lost more easily than intense rainfall that penetrates deeper in the profile through a pronounced wetting front. Consequently, the results show more water draining below the bottom of the soil profile in the 'increased storminess' scenario than in the other two. Note that the use of 325 stochastic ensembles of rainfall results in more variability in these results and perhaps more overlap than might otherwise be expected.  cumulative AET is 3138 mm, 2964 mm and 3003 mm for reduced storminess, historical climate, and increased storminess scenarios respectively while the cumulative drainage is 1247 mm, 129 3mm and 1376 mm for reduced storminess, historical climate, and increased storminess scenarios, respectively.

Discussion
Drylands are water-limited environments facing immense challenges under climate change. In the most food insecure dryland 335 regions of the world, the lack of ground-based meteorological stations increases the uncertainty in forecasting soil moisture and crop yields over seasonal time horizons, and also in the projection of changes to the water balance under climate change.
As a substitution for local precipitation information, data from individual meteorological stations is often gridded, sometimes using remote sensing data and/or global climate model reanalysis data, to obtain spatially consistent characterisations of rainfall (and ET) to quantify the water balance and thereby determine plant-available water and associated crop yields. These 340 calculations then typically form the basis for decision support tools and famine forecasts Brown et al., 2017).
Given that most gridded rainfall products generated in this manner are available on daily or coarser temporal resolutions (e.g., daily data for Global Precipitation Climatology Centre-GPCC and 5-day averages for CHIRPS (Funk et al., 2015)), we explored how these temporal resolutions affect estimates of soil moisture, especially in drylands where rainfall is often delivered in brief rainstorms that are challenging to represent in climate and hydrological models (Nicholson, 2011;Singer et 345 al., 2018). For example, rainstorms often occur on timescales of hours and even portions of hours, delivering water to the land surface at high intensity, where it infiltrates into dry soils on flat ground at high rates. Through our 1D modelling, we found systematically lower soil moisture distributions for progressively coarser temporal resolutions of input rainfall (Figure 7), despite the same total precipitation. This phenomenon arises due to nonlinear 350 interactions between the rainfall resolution and the Richards equation (Wang et al., 2009), and corresponding subsurface propagation of the 'drizzle effect', a term usually restricted to climate modelling, but here used to refer to the effect of temporally averaging precipitation. If a single intense rainstorm of 100 mm/h lasting for 50 minutes is averaged over a day or a week, it will yield a constant input of very light rainfall (Figure 4), which may be easily counterbalanced by AET losses over a diurnal cycle ( Figure 5). Further exploration of the effect of rainfall resolution on soil moisture showed that the differences 355 between hourly input rainfall versus daily and weekly values becomes more pronounced with depth in the soil, as the effects of rainfall resolution propagate through the wetting front. In particular, the coarser temporal resolutions lead to higher AET losses of infiltrated precipitation (Figure 8), lower soil moisture with more time below the wilting point at deeper depths ( Figure 7), and less deep percolation (Figure 8), which might support more deeply rooting plants and crops, or recharge groundwater aquifers via diffuse recharge. If one were to propagate these soil moisture values into crop yields for these shallow 360 rooting depths, they would dramatically underestimate the water availability to critical crops and therefore underestimate crop yields and overestimate the risk of crop failure and corresponding famine for many years of the historical time series. Our results also indicate that moving from hourly to daily to weekly rainfall resolutions in hydrological modelling may lead to 16 % and 45 % underestimation of groundwater recharge (deep percolation), respectively. This result of higher moisture availability at deeper depths could have important implications for more deeply rooted forage crops and native vegetation that 365 serves as a critical food source to livestock in pastoralist communities in dryland regions around the world.
When seasonal rains fail in dryland regions that rely on rainfed subsistent agriculture and pastoralism, it threatens livelihoods and often requires large-scale humanitarian response from governments and NGOs to avert famine. The rural communities within the Horn of Africa drylands (HAD) are particularly prone to climatic shocks and they tend to have low socio-economic 370 levels, and low adaptive capacity to climatic shocks. Hence, recent severe droughts have dramatically increased food insecurity from ~7 million people in 2011 to ~35 million people in 2017, leading to livestock loss and major water shortages. The region has a total population of 160 million people, 70 million of that population live in areas classified as having extreme food shortage, other 40 % of the population is undernourished (Lusamba-Dikassa et al., 2012;Margulis, 2012). Over the last four decades, droughts in HAD have become more frequent and more severe (Lyon and DeWitt, 2012;Liebmann et al., 2014;Funk 375 et al., 2019), reducing soil moisture for plants and affecting groundwater reserves. Therefore, there is increasing need to improve characterisation of drought (and flood) impacts on water stores such as soil moisture (Gruber et al., 2019). Current analyses of soil moisture are often based solely on remote sensing data, which only provides information on the top ~5-10 cm of the soil profile. Other work calculates soil moisture and crop yields based on globally available precipitation data at daily or weekly resolutions . Our analysis suggests that in dryland regions, where precipitation is often 380 delivered in brief, intense storms, it is important to model the impacts of rainfall forecasts on water storage using precipitation data at high temporal resolution, in order to more faithfully capture its effects. Indeed our results from 1D modelling ( 8) corroborate the results form previous work suggesting that rainfall intensity is closely linked with groundwater recharge (Adloff et al., In Review;Taylor et al., 2013;Cuthbert et al., 2019).

385
To overcome the challenges of representing projections of precipitation in dryland regions and its impact on soil moisture, where evaporative demand is extremely high, we opted to investigate climate change using bespoke scenarios that simply modify the rainfall intensity for individual storms (termed 'storminess' here) by ±25 %, while maintaining the same seasonal distribution of total rainfall (Singer et al., 2018). This approach, consistent with previous observations that changes in rainfall 390 intensity may arise within higher resolution rainfall data (Barbero et al., 2017;Guerreiro et al., 2018), enables characterization of the impact such intensity changes may have on the water balance for a region. An ensemble of multiple stochastic simulations done in this manner provide a more robust picture of the likely impact that changes in rainfall intensity will have on soil moisture, AET, and deep percolation. Specifically, we found that systematic differences between inputs (stochastic simulations of stationary historical rainfall, of increased storminess, and decreased storminess (Figure 9) lead to markedly 395 different soil moisture distributions (Figure 10), as well as in notable differences in AET and drainage below the bottom boundary ( Figure 11). The differences in soil moisture between simulations become larger with depth, as higher intensity rainfall is characterized by deeper wetting fronts that promote longer retention of infiltrated moisture. Figure 12 sheds more light on this effect for a series of rainfall events over the years 2008 and 2009, where the penetration depth of individual rainfall events declines from hourly to daily to weekly rainfall resolutions (black arrow shows same rainfall event in each series). It is 400 also apparent from these soil moisture profiles in time series that higher rainfall intensity leads to higher overall moisture content at deeper depths (more yellow colouring in Figure 12), and also likely contributes to groundwater recharge.  These seemingly small differences could have a major influence on water availability to vegetation in dryland regions, and thus rural livelihoods in subsistence regions like HAD. In summary, if rainfall in regions such as HAD becomes more intense due to atmospheric warming, it may support more higher soil moisture and higher groundwater recharge, therefore supporting 410 more sustainable rainfed agriculture, pastoralism, and human drinking water supplies from wells. However, under historical climate or the decreased storminess climate scenario (at least based on the data for Walnut Gulch), the landscape would remain dry for much of the year, creating challenging conditions to raise crops or grow hydrophilic fodder crops.
These results suggest great care should be taken in hydrological modelling to represent input rainfall at the most appropriate 415 temporal (and spatial) resolution to best represent the relevant hydrological processes that affect key water balance components: infiltration, evapotranspiration, and runoff. If rainstorms occur at subhourly timescales, it is not sensible to use daily or weekly rainfall to drive such a model. Instead, we recommend the use of the highest rainfall resolution available (often 15-min for many gauging locations). If there are severe limits to the availability of such data, stochastic models that resolve https://doi.org/10.5194/hess-2021-48 Preprint. Discussion started: 4 February 2021 c Author(s) 2021. CC BY 4.0 License. rainfall at such scales is recommended to drive models of hydrological partitioning (Singer et al., 2018). And given the large 420 uncertainties in projecting future changes in precipitation from global climate models, it is further recommended to explore a range of methods for simulating future rainfall scenarios with trends that seem plausible or which are based on physical concepts. Ultimately, to make progress in forecasting and predicting the effects of rainfall changes to soil moisture (and by extension, runoff, and groundwater recharge) and the cascading impacts to vegetation and human society.

Conclusions 425
In this study we evaluated the impacts of rainfall data resolution as well as impacts of climate change on water partitioning upon reaching the soil surface in dryland environment. We used the Hydrus-1D model, driven by historical and stochastic inputs, and applied it to estimate the soil moisture profile at the Kendall site of WGEW. There we demonstrated the importance of high-resolution rainfall data in best characterising the soil moisture and time below wilting point and explored the potential effects of shifts in rainfall intensity associated with climate change. We discussed the broader implications for subsistence 430 communities living in drylands globally. Our straightforward modelling approach yielded new insight into the links between the climate system and soil moisture storage and the water balance in drylands.
Competing interests. The authors declare no conflict of interest.

435
Author's Contribution. IK, KM, RR and MBS designed this study. IK performed the model Simulations MBS and KM provided the model data for the climate change simulations. KM, RR and MBS assisted IK in result analysis and discussions.
All authors contributed to writing and revising the manuscript.