Widespread woody plant use of water stored in bedrock

In the past several decades, field studies have shown that woody plants can access substantial volumes of water from the pores and fractures of bedrock1–3. If, like soil moisture, bedrock water storage serves as an important source of plant-available water, then conceptual paradigms regarding water and carbon cycling may need to be revised to incorporate bedrock properties and processes4–6. Here we present a lower-bound estimate of the contribution of bedrock water storage to transpiration across the continental United States using distributed, publicly available datasets. Temporal and spatial patterns of bedrock water use across the continental United States indicate that woody plants extensively access bedrock water for transpiration. Plants across diverse climates and biomes access bedrock water routinely and not just during extreme drought conditions. On an annual basis in California, the volumes of bedrock water transpiration exceed the volumes of water stored in human-made reservoirs, and woody vegetation that accesses bedrock water accounts for over 50% of the aboveground carbon stocks in the state. Our findings indicate that plants commonly access rock moisture, as opposed to groundwater, from bedrock and that, like soil moisture, rock moisture is a critical component of terrestrial water and carbon cycling. Woody plants across the continental United States make extensive use of water stored in bedrock across diverse climates and biomes.

In the past several decades, field studies have shown that woody plants can access substantial volumes of water from the pores and fractures of bedrock [1][2][3] . If, like soil moisture, bedrock water storage serves as an important source of plant-available water, then conceptual paradigms regarding water and carbon cycling may need to be revised to incorporate bedrock properties and processes [4][5][6] . Here we present a lowerbound estimate of the contribution of bedrock water storage to transpiration across the continental United States using distributed, publicly available datasets. Temporal and spatial patterns of bedrock water use across the continental United States indicate that woody plants extensively access bedrock water for transpiration. Plants across diverse climates and biomes access bedrock water routinely and not just during extreme drought conditions. On an annual basis in California, the volumes of bedrock water transpiration exceed the volumes of water stored in human-made reservoirs, and woody vegetation that accesses bedrock water accounts for over 50% of the aboveground carbon stocks in the state. Our findings indicate that plants commonly access rock moisture, as opposed to groundwater, from bedrock and that, like soil moisture, rock moisture is a critical component of terrestrial water and carbon cycling.
Plant transpiration mediates water and energy exchange at Earth's surface. The circulation of near-surface water by plant roots has consequences for a large number of Earth-system processes, including landscape evolution, ecosystem carbon storage and nutrient delivery to streams 6 . At present, soils (physically mobile regolith) are thought to store the majority of root-zone water. As a result, soil processes underpin the conceptual frameworks and models used to predict environmental change. For example, climate projections rely on large-scale estimates of soil hydraulic properties 7 .
However, plants can source water and nutrients from bedrock 8 , which is exposed or only thinly soil-mantled across much of Earth's terrestrial surface 9 . Unlike soils, bedrock is characterized by relict primary rock structures, such as bedding or joint planes, which manifest distinct hydraulic 10 and biological 11 processes.
Recent field studies have indicated that plants can access substantial volumes of rock moisture 1,4 , defined as plant-available water stored in unsaturated, weathered bedrock 3 . Furthermore, the water storage capacity of bedrock can explain ecosystem distributions and drought vulnerability [12][13][14] . In the face of widespread drought-induced die off 15,16 , massive wildfires 17 and woody encroachment 18 , information about the spatial and temporal patterns of plant-available water in bedrock is needed to appropriately predict water and carbon fluxes under environmental change.
Here we quantify root-zone water storage in bedrock across the continental United States (CONUS) using publicly available data. We estimate lower bounds on the magnitude and frequency of bedrock water use by plants, and map the spatial distribution of plant access to bedrock water.

Article
The magnitude and spatial distribution of D bedrock,Y across California and Texas are reported in Fig. 3a. In any given year, transpiration is at least partially sourced from bedrock over at least 28-30% and 5-10% of the total land areas of California and Texas, respectively (Fig. 3a). D bedrock,Y for all of the CONUS is reported in Extended Data Fig. 3. In some areas, D bedrock,Y exceeds 300 mm and can constitute more than one-quarter of the mean annual precipitation (Extended Data Fig. 4). Bedrock is thus a critical storage reservoir of plant-accessible water over large areas. We focus here on California and Texas because bedrock water use has been documented via field studies in those states (Fig. 3b) and they experience extended dry periods where deficits reflecting storage volumes can accumulate.
Deficit-based methods, such as those employed here, yield lower-bound estimates of root-zone water storage (Methods); however, where there are long, extended dry periods or where energy and precipitation delivery are out of phase, deficit-based estimates of root-zone storage are more likely to approach actual root-zone storage capacity. By contrast, where precipitation occurs year-round or where energy and precipitation delivery are in phase, deficit-based methods will more substantially underestimate root-zone storage capacity. This is because withdrawal from storage (that is, ET) during extended dry periods will cause increases in an accrued deficit, whereas ET during periods with frequent precipitation will not result in a large accrued deficit.
We calculate a bedrock root-zone water storage capacity, S bedrock , which is defined as the largest storage used by woody vegetation over a multiyear time window (2003-2017) that cannot be accounted for by soil water storage capacity (Methods, Extended Data Fig. 5). S bedrock as a percentage of total root-zone storage capacity is reported in Fig. 4, which shows that bedrock water storage often constitutes the majority of total storage capacity in the root zone.
In some locations, the magnitude of D bedrock,Y is relatively consistent across different years, and consequently similar to S bedrock , indicating that plants withdraw similar amounts of bedrock water each year. However, in other locations, such as the southern Sierra Nevada in California and the Edwards Plateau in Texas, S bedrock is often larger than D bedrock,Y (Extended Data Figs. 3,5), indicating that the storage capacity of plant-accessible water in bedrock is much greater than the storage that is withdrawn in a given year. Under these conditions, bedrock may have a central role in plant response to multiyear drought because bedrock water is progressively drawn down to explain the observed ET 20 .
Bedrock water serves as a reservoir for transpiration in locations hosting high aboveground biomass (Extended Data Fig. 2a) across a range of biomes and Köppen climate types, including humid climates (Extended Data Fig. 6). The largest measurements of S bedrock are associated with arid, semiarid and Mediterranean climate types and evergreen forests, savannahs and shrublands (Extended Data Fig. 6, Extended Data Table 1). Bedrock water storage may be particularly important in semiarid shrublands, Mediterranean savannahs and Mediterranean needleleaf forests (Extended Data Table 1).

Rock moisture commonly accessed
Locations where field studies document plant use of unsaturated bedrock water storage (that is, rock moisture) coincide with locations where we calculate positive median D bedrock,Y (Fig. 3b, Extended Data Fig. 7). This corroborates our use of D bedrock,Y as an indicator of ecosystem access to bedrock water stores. Field studies reporting greater than 50% of annual ET derived from rock moisture are shown in Fig. 3b. Some of these sites do not meet our analysis criteria (Methods) and are consequently masked (designated with superscripts in Fig. 3b, Extended Data Fig. 7). This is another indication that our reported values are underestimates of the spatial extent of bedrock water use, and thus the volume of bedrock water accessed (Methods). Although bedrock water storage volumes measured at these sites are calculated using very different methods from those employed here, there is general agreement between D bedrock,Y (shown as blue bars in Fig. 3b) and field measurements of bedrock water storage accessed by plants (shown as circles in Fig. 3b).
Bedrock water storage used by plants can commonly occur in the form of rock moisture (Fig. 3b, Extended Data Fig. 7); however, D bedrock,Y and S bedrock do not discriminate between rock moisture (unsaturated) and bedrock groundwater (saturated). Even in field settings, partitioning plant water use between the unsaturated and saturated zones remains challenging, yet the distinction between them is germane to mechanistically modelling biogeochemical and hydraulic processes. Rock moisture use has been confirmed under circumstances that might commonly be attributed to groundwater use. For example, Hahm et al. 21 have shown that oaks relied on rock moisture to sustain dry season transpiration at an oak savannah site where groundwater remains within 3 m of the surface throughout the year. Insensitivity of ET to extended drought is another tool used to attribute groundwater as a transpiration source; however, storage capacity in the unsaturated zone can produce similar insensitivity of ET to drought 12 . These circumstances suggest that misattribution of rock moisture as groundwater is likely, and that rock moisture use by woody plants may be common.  The root-zone water storage capacity is partitioned into soil and bedrock components. b, The extent of woody vegetation is coloured by soil thickness, which could also be considered the depth to bedrock because only areas mapped as underlain by bedrock are shown. Landcover data were sourced from the USGS NLCD 40 and soil thickness from the USDA gNATSGO 41 . All raster maps in all figures and Extended data were plotted in QGIS 42 , with map data generated in Python in the Google Colaboratory environment. All raster data are publicly available and were processed using the Google Earth Engine Python application programming interface (API).

Implications of bedrock water uptake
Although it has long been recognized that woody plants root into bedrock 22 , the widespread and routine transpiration of bedrock water reported here suggests that the dynamics of bedrock water storage may be as fundamental to understanding terrestrial water and carbon cycling as soil moisture. Across the western United States in particular, large volumes of water are stored in bedrock and released back into the atmosphere on an annual basis. For example, our deficit analysis suggests that in California alone, 20 km 3 (16.2 million acre-feet) of water can be extracted from bedrock by woody plants annually. This is approximately equal to the volume of water stored in all of the state's reservoirs combined 23 , and about three times the state's annual domestic water use 24 . Although our study is limited to the CONUS, bedrock water use by woody vegetation has also been documented in a wide range of environments globally [25][26][27][28][29][30][31][32] . Investigation of biological and hydraulic processes in the bedrock rhizosphere is a frontier research area [4][5][6] . New studies are needed to clarify the role of bedrock water storage under projected shifts in global precipitation regimes, including multiyear drought and alternation between extreme wet and dry years. In the 2011-2016 California drought, for example, forest ecosystems with access to rock moisture exhibited diverse responses, from insensitivity 12 to vulnerability 20 . This motivates new field-based observational studies of belowground structure and bedrock water storage dynamics across diverse lithological, climatic and ecological settings to clarify the different ways in which bedrock water storage mediates ecohydrological processes 33,34 .
Plant bedrock water use, and specifically the use of rock moisture, occurs in critical locations for water supply, including the Sierra Nevada, the recharge zone of the Edwards and Trinity aquifers, and the headwaters of the Colorado River (Figs. 2, 3), which together supply water to at least one-quarter of the US population. Given that the dynamics of rock moisture have the potential to regulate the timing of groundwater recharge and runoff 35 , bedrock water storage may be critical to water resource planning.
Woody ecosystem dependence on stored subsurface water will probably increase in the future as plant community ranges shift 36 , snowpack declines in high-elevation and high-latitude regions, and many environments undergo a transition from energy-limited to water-limited conditions 37 . Thus, the availability of bedrock water storage may be key to predicting large-scale vegetation dynamics, including the stability or vulnerability of ecosystem carbon storage, under climate change.
Long-term, intensive monitoring studies are increasingly documenting mechanisms by which roots in bedrock impact ecosystem function 13 , groundwater and stream chemistry 38 , and rates of soil production and weathering 6 . Although bedrock water storage in the humid eastern USA may be largely undetectable via a deficit-based water balance, substantial circulation of water in bedrock may be occurring. This could lead to largely unmeasured drivers of carbon cycling 39 . Thus, bedrock water storage dynamics are likely key to understanding the sensitivity of carbon, water and latent heat fluxes to changes in climate.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-021-03761-3.   Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Literature compilation of rooting in bedrock
Available English-language published evidence of rooting into bedrock is included in our literature compilation 51 , which builds on several past compilations 1,[52][53][54][55] . Each entry includes information about rooting, climate, soil and bedrock properties. A subset of sites report use of rock moisture by vegetation. For these entries, where possible, we report estimates of the contribution of rock moisture to evapotranspiration, as well as any estimates of plant-available soil and rock moisture water storage capacities (Fig. 3b, Extended Data Fig. 7).

Landcover and soil datasets
To determine woody landcover, we used the evergreen, deciduous, mixed forest and shrub/scrub landcover classes reported by the United States Geological Survey (USGS) National Land Cover Database (NLCD) 40 at 30-m resolution. To determine areas underlain by bedrock within 1.5 m of the surface, and the available soil water storage capacity for those areas, we use the United States Department of Agriculture (USDA) Gridded National Soil Survey Geographic Database (gNATSGO) product at 90-m resolution 41 . gNATSGO data are generated using soil data from field surveys and subsequent laboratory analysis 41 . These surveys are occasionally repeated and the newest data are validated against historical surveys before replacement in the official nationwide database 41 . To determine where bedrock underlies shallow soils, we use the gNATSGO product, which reports depths of soil restrictive layers for the classifications of lithic, densic and paralithic bedrock. Our calculation of bedrock water storage considers only areas where bedrock has been encountered within 1.5 m of the surface (Fig. 1, Extended Data Fig. 1). The 1.5 m depth is chosen because soil water storage capacity (S soil ) is only available across the CONUS to 1.5 m depth. Although bedrock water may be accessible to plants in areas with greater than 1.5 m soil depth, we exclude these areas because we cannot quantify S soil there. We note that in practice, the interface between soil and bedrock has not been systematically mapped and the terminology used for defining that interface can be inconsistent 9 . The contact between soils and underlying bedrock can also be gradational and challenging to determine in the field. For example, saprolite, which can be defined as highly weathered bedrock that retains the original fabric of the rock, is often, but not always designated as a 'C' or 'Cr' horizon by the gNATSGO soil survey, and thus categorized as a soil in our study. Therefore, S soil can include saprolite.
We estimate S soil as the 'soil available water storage' (AWS) reported by the gNATSGO database 41 (Extended Data Fig. 2b). This AWS product is calculated as the storage volume, in units of depth, between field capacity (−1/10 bar or −1/3 bar) and wilting point (−15 bars) and is measured for each soil layer until contact with a bedrock restrictive layer. For each layer within a given soil profile, gNATSGO reports a high, low and likely value of AWS, which they take a thickness weighted average of to generate three estimates of profile total AWS. Here we use the highest reported value to represent the AWS of any given layer to avoid underestimating soil water storage. As the AWS product does not account for water stored between field capacity and saturation in soils, we tested the sensitivity of our results to the inclusion of this excess water by reporting S bedrock and median D bedrock,2003-2017 for a hypothetical test case of double S soil (Extended Data Fig. 8). We double S soil to approximate the volume of water between field capacity and saturation. Doubling of S soil necessarily reduces the magnitudes of S bedrock ; however, the spatial area of positive S bedrock is reduced by only 35%, indicating that underestimation of soil water storage capacity by a factor of two would still lead to a large volume of bedrock water use across the CONUS.

Masking procedure
We employ three masking criteria to constrain our analyses to places where (1) woody landcover occupies at least 75% of the 500-m pixel, (2) all soils within the 500-m pixel are underlain by bedrock and less than 1.5 m deep, and (3) total evapotranspiration is less than total precipitation from 2003 to 2017 (Extended Data Fig. 1). The first two masking criteria restrict our calculations to places where water storage deficits could be explained by water extraction by woody plants from bedrock, because bedrock is near the surface and woody plants are present. The third masking criterion is employed to remove locations where outputs exceed inputs over long timespans, indicating either errors in fluxes or unmeasured fluxes entering the rooting zone, such as fog, dew, irrigation or lateral flow in soils. Bedrock water storage could be accessed in areas that do not meet these criteria, and indeed, there are several studies that report plant use of bedrock water in locations that are masked (Fig. 3b, Extended Data Fig. 7). However, in locations where our masking criteria fail to account for fog, dew or lateral inputs of water, bedrock water storage capacity may be overestimated (Methods).

Calculation of root-zone water storage capacity and maximum annual root-zone storage deficit
Here we use a statistically interpolated precipitation data product (Oregon State's PRISM daily precipitation 56,57 ) and a remotely sensed evapotranspiration product (Penman-Monteith-Leuning Evapotranspiration V2 58,59 ) to estimate the minimum magnitude of root-zone water storage capacity (S r ) following the method developed by Dralle et al. 60 , which adapts the original method of S r estimation from Wang-Erlandsson et al. 61 to account for the presence of snow. All raster processing was conducted using the Google Earth Engine 62 Python application programming interface (API).
The method takes a mass-balance approach and is therefore broadly applicable, not requiring place-based soil or plant-community parameterization 63 . Specifically, the technique tracks a root-zone storage deficit (D) as a running, integrated difference between water fluxes exiting (F out ) in units of length per time [L/T] and entering (F in [L/T]) the root zone, here taken to be evapotranspiration (ET) and precipitation (P), with F out = ET and F in = P. This is accomplished by first computing the accumulated difference between F out and F in over a given time interval t n to t n+1 : where C 0 is the threshold percentage of areal snow cover deemed non-negligible, here chosen as 10%. This avoids attributing evapotranspiration from snowmelt recharge into the rooting zone to unreplenished water storage depletion. Snow data are acquired from the Normalized Difference Snow Index (NDSI) snow cover band from the 500-m MODIS/Terra data product 64 .
With this, the root-zone storage deficit at any given time is defined iteratively as: Following these equations, D at any given time represents a lower bound on the volume of water that plants have used that must have been withdrawn from root-zone storage without replenishment by precipitation. The deficit is effectively 'reset' to zero during wet periods, because the updated D(t n+1 ) equals the maximum of 0 and the previous deficit plus the current difference between outgoing and incoming fluxes. Over the course of a year or many subsequent seasonal cycles, the maximum value of D represents the largest amount of subsurface water storage space that must have been used to supply ET.
Here we report two deficit-related quantities: the observed maximum root-zone storage deficit in water year Y (D max,Y ) and the maximum root-zone storage deficit over the period of record (2003-2017), taken as a lower bound on the actual root-zone storage capacity, S r . D max,Y is calculated for a given water year Y (that is, from 1 October in year Y − 1 to 30 September in year Y) first by assuming the root-zone storage deficit on 1 October is zero, then tracking that deficit through to the end of the water year. D max,Y is the maximum value of the deficit time series over that water year. The procedure for computing S r is similar, but the deficit time series is computed over the period of record. That is, D is taken to be zero on 1 October 2003 and is tracked continuously until 30 September 2018. S r is then taken to be the maximum value of this multiyear deficit time series. Importantly, S r and D max are conservative lower estimates of water storage capacity and do not account for all possible withdrawal (see 'Assumptions and limitations of deficit-based calculations of bedrock water storage').

Bedrock root-zone water storage capacity and annual bedrock root-zone water storage
To quantify the root-zone storage capacity that cannot be accounted for by soil water storage capacity, S bedrock , we subtract the soil water storage capacity from S r , making sure to bound S bedrock at zero: We perform a similar calculation to quantify the annual bedrock root-zone water storage capacity, D bedrock,Y , which is the maximum annual root-zone storage deficit that cannot be accounted for by soil water storage capacity: To attribute D bedrock,Y and S bedrock to transpiration of bedrock water by woody plants, we assume that evaporation is restricted to the soil layer, such that evaporation fluxes are accounted for by subtraction of S soil from D max,Y or S r . Note that we use the highest AWS value reported. Therefore, S bedrock and D bedrock,Y are conservative lower bounds, as we use the upper bound on S soil and the lower bound on S r and D max,Y , respectively. The sensitivity of S bedrock to S soil is discussed above in 'Landcover and soil datasets'.

Assumptions and limitations of deficit-based calculations of bedrock water storage
The methods we use to estimate the spatial pattern and magnitude of bedrock water use will provide a lower bound on bedrock water storage capacity, because (1) we employ a deficit-based water balance, (2) we use the largest available estimate of soil water storage capacity, and (3) we use masking criteria to exclude areas where alternative mechanisms might reasonably account for evapotranspiration. Here we explore the assumptions and limitations of our approach. Deficit-based calculations of root-zone storage yield lower-bound estimates because they rely on fluxes to infer storage dynamics. That is, deficit-based methods cannot 'detect' the presence of a storage element if that storage does not supply a flux over the period of record of the flux datasets. For this reason, actual root-zone storage capacity will always exceed deficits measured through water-balance methods. Thus, in the absence of systematic error, the deficit is a lower bound on storage capacity. In addition, we make an assumption that bedrock water storage is only accessed when soil water storage is exhausted. If bedrock water is accessed at the same time as soil water storage, then our water balance calculation would result in additional underestimation of bedrock water storage capacity.
We assume that tracking the fluxes of precipitation (F in ) and evapotranspiration (F out ) into and out of a pixel, respectively, results in a lower-bound estimate of root-zone water storage deficit. In addition to the reasons listed elsewhere, this is also because the deficit is minimized by ignoring any fluxes out of the pixel that occur by mechanisms other than evapotranspiration, such as downward drainage or runoff. We acknowledge that not all precipitation entering the root zone leaves as evapotranspiration; however, by imposing that F out is represented by evapotranspiration alone, the deficit represents a lower bound on root-zone storage capacity. Including any additional fluxes in F out would act to increase the deficit. As drainage is challenging to quantify, we follow deficit-based calculation methods (for example, Wang-Erlandsson et al. 61 ) and do not attempt to quantify it. Instead, we report the lower bound of root-zone storage, which occurs when F out occurs by evapotranspiration only.
Underestimating input fluxes (F in ) leads to overestimating S bedrock and D bedrock,Y . F in could be underestimated where fog, dew, irrigation or lateral flow (across pixels) is important. Fog and dew may be important sources of water, but are probably only important in a small subset of the areas where we report S bedrock and D bedrock,Y . By masking locations where evapotranspiration exceeds precipitation over long time periods, we exclude locations where additional inputs to the root zone are required to explain the observed evapotranspiration data. However, lateral transport of water in the subsurface could still occur without causing evapotranspiration to exceed precipitation in the long term, in which case S bedrock and D bedrock,Y would be overestimated. By removing entire 500-m pixels where any soils exceed 1.5 m depth, we tend to exclude convergent parts of the landscape, which can have thicker soils. These areas are the most likely to experience lateral inputs of water into the root zone. Nonetheless, additional research is needed to constrain lateral water flows within hillslopes to better understand water availability to plants.
Systematic errors in the data products used in our water balance could lead to overestimation of storage. One limitation of the deficit method is that it relies on taking the integrated (summed) difference between precipitation (F in ) and evapotranspiration (F out ) such that error in either flux will accumulate and could be large relative to small deficit estimates. S bedrock across the CONUS is shown in Extended Data Fig. 5. We compare this result to bedrock water storage deficit estimates obtained using the root-zone water store capacity (S r ) dataset of Wang-Erlandsson et al. 61 (who used different P and ET data products at a coarser spatial resolution) shown in Extended Data Fig. 9. The patterns of bedrock water storage capacity remain similar, which suggests that the general spatial trends and magnitudes in bedrock water storage are robust to choices in input data products.
As remotely sensed ET and P datasets and in situ measurements of bedrock water storage become available, such datasets could be used to create increasingly accurate estimates of bedrock water use following the workflow presented here.