Confronting the water potential information gap.

Water potential directly controls the function of leaves, roots, and microbes, and gradients in water potential drive water flows throughout the soil-plant-atmosphere continuum. Notwithstanding its clear relevance for many ecosystem processes, soil water potential is rarely measured in-situ, and plant water potential observations are generally discrete, sparse, and not yet aggregated into accessible databases. These gaps limit our conceptual understanding of biophysical responses to moisture stress and inject large uncertainty into hydrologic and land surface models. Here, we outline the conceptual and predictive gains that could be made with more continuous and discoverable observations of water potential in soils and plants. We discuss improvements to sensor technologies that facilitate in situ characterization of water potential, as well as strategies for building new networks that aggregate water potential data across sites. We end by highlighting novel opportunities for linking more representative site-level observations of water potential to remotely-sensed proxies. Together, these considerations offer a roadmap for clearer links between ecohydrological processes and the water potential gradients that have the 'potential' to substantially reduce conceptual and modeling uncertainties.

Gradients in the water potential (Ψ) of soils and plants form the energetic basis for the transport of water, and elements contained therein, through a connected continuum linking the deepest soil layers to the top of plant canopies (Figure 1). Ψ can be a positive or negative pressure, though it is typically negative -- a tension force -- in unsaturated soils and within plant hydraulic systems. Ψ gradients have been recognized as the fundamental driver of water fluxes between soils, streams, and groundwater for more than a century, and they appear in some of the most foundational equations in hydrology[1] (e.g. Darcy’s Law, Richard’s Equation). Likewise, the critical role of Ψ gradients in driving water flows through the soil-plant-atmosphere continuum has been known for decades[2].

Figure 1:

Water potential links environmental drivers to biophysical responses.

Water flows “downhill” along gradients of water potential in the soils (Ψ, where water potential is relatively high, often >-1 MPa) through the stems (Ψ) to the leaves (Ψ, where potential is relatively low) and eventually to the air (Ψ, where it can be as low as −100 MPa). Water potential also directly controls key biological processes, including microbial function, mortality risk arising from damaged plant xylem, and plant-atmosphere gas exchange. While observations of environmental drivers, soil moisture content (θ) and carbon and water fluxes are broadly accessible from environmental networks and remote sensing, Ψ timeseries are more discrete, sparse, and generally not coordinated or discoverable.

Beyond redistributing water through ecosystems, Ψ is also a direct control of many biophysical processes. Soil water potential (Ψ) regulates flow of water into and out of soil microbe cells and determines their metabolism[3]. In plants, leaf water potential (Ψ) is a key driver of stomatal conductance and photosynthetic carbon uptake[4,5], and its close connection to branch and stem water potential (Ψ) controls the risk of drought-driven xylem embolism and mortality[6,7]. Consequently, most ecosystem services, including water storage, food and fiber supply, and water and climate regulation, are fundamentally linked to Ψ.

While undeniably important for soil and plant function, for reasons discussed in more detail below, Ψ is rarely measured in-situ[8,9], and observations of plant Ψ have historically been limited to destructive and disjunct manual measurements. The objective of this paper is to demonstrate key uncertainties linked to the dearth of soil and plant Ψ data, and to discuss the theoretical and modeling progress that could be enabled with richer and more discoverable information about Ψ. We begin by discussing issues surrounding the measurement, modeling, and synthesis of soil water potential, and then address additional considerations linked to the measurement and prediction of water potential in plants. We then present a road map for creating accessible and open Ψ databases and discuss promising new approaches for detecting Ψ using remote sensing.

Concepts and uncertainties linked to soil water potential

Water flows “downhill” energetically, moving from areas of higher-to-lower potential, such that Ψ gradients are the driving force of subsurface water flows[1]. In most unsaturated soils, Ψ is dominated by the matric potential, which becomes more negative when soils dry, and the effective radii of water-filled pore spaces in the soil become smaller. This process produces the general shape of the water retention curve (also known as the ‘moisture characteristic’ or ‘water release’ curve), which relates Ψ to volumetric soil moisture content (θ). Critically, variation in soil physical properties can cause Ψ to differ by an order of magnitude across soil types, even if soil moisture content is the same[10,11] (Figure 2a).

Figure 2:

Water retention curve and pedotransfer function (PTF) uncertainty.

Across soil types, Ψ can differ by an order of magnitude for a given soil moisture content (panel a, with curves generated from the van Genutchen model[11],, see methods). Panels b-d illustrate the uncertainty in the water retention curve attributable to PTF parameter uncertainty. The shaded area shows the 90% confidence interval due solely to variation in a single parameter of the van Genuchten model (the ‘n’ shape parameter, which is linked to pore size) within just one standard deviation of its reported distribution for each soil class from a popular PTF[18]. Thick lines in panels b-d are the same as in panel a. The PTF-driven uncertainty in the water retention curve propagates into large uncertainty for modeled fluxes and pools. Specifically, variation in the van Genuchten ‘n’ parameter within again just one standard deviation of its reported range[18] causes the 90% confidence intervals on modeled evapotranspiration (ET), soil moisture content (θ, and Ψ (shaded gray areas, panels e-f) to vary by a magnitude comparable to the mean value of each parameter (thick black line). Simulations were run using the HYDRUS 1-D[79] model for a forest site in Indiana, US[80] during a drought event (see methods for details).

Field observations of θ are common[12], but with a few exceptions[9,13], Ψ is rarely measured systematically in field research settings[8,9]. The reasons why θ became the predominant metric for describing soil water status are not entirely clear[8], but may reflect the fact that no single instrument captures the entire range of Ψ (from saturation to the very dry end), and sensors for measuring Ψ in the field have historically been associated with unique limitations and uncertainty[8,14].

Even if Ψ data were plentiful, strategies for relating θ to Ψ would still be necessary in models to connect water balance equations with potential-driven flows. Most hydrologic and land surface models thus rely on water retention curve models[15], with those proposed by Campbell (1974)[10] or van Genuchten (1980)[11] ranking high in popularity. Pedotransfer functions (PTFs) predict the parameters of water retention curve models using empirical equations driven by a limited set of soil characteristics (typically %sand, %clay, and bulk density[16-18]).

While developing PTFs is an active field[15], PTF parameter distributions are poorly constrained and prevent confident transformation of θ to Ψ. For example, even relatively small variations in a single parameter of the van Genuchten model cause Ψ to vary by an order of magnitude over a wide range of θ (Figure 2b–2d). Soil structure, which differs from soil texture and is governed by biophysical properties, may be a key omission in PTFs[19] explaining some of this uncertainty. For example, growth of roots and mycorrhizae into soil pores, and deposition of root exudates, increase overall water retention[20,21], and macropores can create preferred flow pathways that are challenging to incorporate into PTFs. Moreover, depth into the soil may also affect hydraulic properties by controlling connectivity with root systems and through slowly-evolving changes in soil morphology. Finally, most PTFs assume that the water retention curve is static; but many relevant processes occurring in natural landscapes (including drying-rewetting cycles, fire, and management shifts) may cause time-dependent hysteresis of the water retention curve[22-24].

This uncertainly linked to PTFs propagates through water cycle models in highly consequential ways.[25-26] Prior work performed in the Shale Hills Critical Zone Observatory confirms that van Genuchten model parameters are the dominant source of model uncertainty in a coupled 3-D land-surface and hydrological model[27], and that water retention curve parameters must be measured locally and optimized through data assimilation[28] for watershed hydrologic variables to be predicted with any degree of certainty[29]. Here, using a popular 1-D water balance model, we further demonstrate that uncertainty in a single PTF parameter drives large uncertainty in modeled predictions of evapotranspiration, soil moisture, and Ψ (Figure 2e).

The parameters of the water retention curve are also key sources of uncertainty explaining variability in carbon cycle fluxes from global-scale land surface models. Here, we used a global sensitivity experiment[30] to explore the variability of these parameters along with other key parameters of the ORCHIDEE land surface model[31,32] (see methods for details). The parameters of the water retention curve explained between 10–32% of the modelled GPP variance across three diverse sites (Figure 3). Moreover, when considering the wider set of soil hydrology parameters (including the hydraulic conductivity, field capacity, and permanent wilting point of the soil), the percentage of explained GPP variance increased to 22–53% across sites.

Figure 3.

Water retention curve parameters are a key source of land surface model uncertainty.

A sensitivity analysis of key model parameters of the ORCHIDEE land surface model[31,32] was performed to demonstrate the relative importance of each parameter in simulating daily GPP at three contrasting FLUXNET sites: a) a temperate broadleaf forest (Harvard Forest, FLUXNET code US-Ha1[82]); b) a boreal needleleaf forest (Sodankyla, FI-Sod,[83]); and c) a semi-arid savanna (Demokeya, SD-Dem[84]). The Sobol method[30] was used to perform the sensitivity analysis; this method is based on variance decomposition and is able to capture interactions between parameters. More details can be found in the methods.

The dearth of information about Ψ is not only a problem for models, but also confounds observation-driven work. Because θ is widely measured, and Ψ is not, it is extremely common to see key response variables like carbon and water fluxes explained as a function of measured θ[33-35]. These relationships are usually non-linear and threshold driven[36-37]. This is not surprising, as these responses embed site-to-site variability in the water retention curve, which itself is nonlinear and threshold-driven (Fig. 2a–d). The shape of these response functions thus depends very much on whether Ψ or θ is chosen as the driving variable[38]. Indeed, the relationship between gross primary productivity (GPP) and soil water status is more linear and less spatially heterogeneous when Ψ, as opposed to θ, appears on the x-axis (Figure 4). Likewise, substantial skill in predicting soil respiration can be gained when model functions are driven explicitly by Ψ[3]. Thus, more abundant and aggregated site-level Ψ information could reduce conceptual uncertainty about how ecosystem fluxes respond to soil water deficits, and permit other sources of spatio-temporal variability to be more discernable.

Figure 4:

Soil water potential better explains variability in GPP when compared to soil moisture content.

Across four AmeriFlux sites for which site-specific water retention curves were measured[38,85–87], the relationship between GPP (normalized by its well-watered rate) and Ψ (bottom row) is more linear than the relationship between GPP and θ (top row). Moreover, cross-site heterogeneity in the response functions is reduced when it is Ψ, as opposed to θ, on the x-axis (compare panel e to panel j). GPP estimates were obtained from AmeriFlux, with site codes given in parentheses. Error bars indicate one standard error of the mean, which is quite small for some of the binned averages. See methods for more details.

Plant water potential: Key concepts and controversies

The effective radii of evaporating water surfaces within plant cell walls are extremely small, resulting in tension forces strong enough to pull water upwards from soils, where it is already tightly bound, to the leaves. Thus, the difference between Ψ and Ψ is the driving force for transpiration, which is closely coupled with photosynthetic carbon uptake. Moreover, branch and stem water potential (Ψ), which are coupled with Ψ, interact with anatomical features of the plant’s water transport system to determine the risk of xylem embolism that can lead to mortality[6,7,39-41]. Stomatal regulation of gas exchange is also critical for buffering plants from the very low water potential of the atmosphere (see Figure 1), which is extremely sensitive to relative humidity.

Historically, observations of plant Ψ have been limited to manually collected “snapshots” (e.g. with a pressure chamber[43]). These data have proven indispensable for shaping our theoretical understanding of how plants respond to soil water stress[6,7,40,44]. However, because pressure chamber measurements are destructive and labor intensive, they are typically limited to weekly or seasonal temporal resolutions. While the weekly timescale is well matched to soil drying, it is too coarse to capture faster-acting hydrodynamic processes, including stomatal response to vapor pressure deficit (VPD[45]) and the depletion and refilling of plant water pools over the course of a day[46]. Moreover, with some exceptions[47], Ψ and Ψ are not often monitored over long time periods (e.g. years to decades), and centralized databases and networks for time series of Ψ do not yet exist.

The discrete and undiscoverable nature of plant Ψ observations limit our ability to characterize the distributions of the minimum plant water potentials that are so critical for determining plant mortality risk[41]. The gap also limits understanding of how plant and soil water potential are coordinated and coupled. For example, a fundamental assumption in plant eco-physiology is that Ψ and Ψ are equilibrated with Ψ across the root zone in pre-dawn hours[48]. This assumption has allowed eco-physiologists to circumvent the Ψ data scarcity problem by relying on pre-dawn Ψ observations as a proxy for root-zone Ψ – an approach that treats the plants as an instrument for recording the soil water environment. Yet experiments have shown that nighttime transpiration – while small – can still occur[49,50], lowering pre-dawn Ψ and decoupling it from Ψ[51]. Synthetic assessments of pre-dawn equilibrium are hindered by the absence of nocturnal Ψ observations collected together with data on Ψ and/or stem water flows (e.g. from sap flux), or at least often enough to determine if stationarity in pre-dawn Ψ, which should be a hallmark of equilibrium, has been achieved.

Likewise, the water potential information gap limits understanding of how soil and plant water potential are coupled at mid-day. The relationship between mid-day Ψ and the root-zone Ψ is frequently used to classify plant water use strategies[44,52,53]. For example, plants with conservative water use strategies (“isohydric” species) close stomata quickly as Ψ declines, whereas “anisohydric” plants keep stomata open longer, sustaining gas exchange but with more rapid declines in Ψ that may increase the risk of xylem embolism. The (an)isohydry framework is popular but controversial, with several studies highlighting critical interactions with other environmental drivers beyond Ψ[54-56], including VPD[57]. Moreover, coordinated observations of sapflow, enhanced with data on soil and stem water potentials, hold great promise for understanding how the dynamics of hydraulic conductance of different plant organs influence whole-plant hydraulic physiology[58]. Plant hydraulics schemes relying on concepts like isohydry are rapidly being incorporated in hydrologic and Earth system models[59-61]. Benchmarking and testing these schemes would benefit from open and spatially representative databases of plant and soil Ψ timeseries, measured together at a temporal frequency (e.g. hourly) over which key drivers like VPD vary.

Coordinated observation of plant and soil Ψ could also offer new perspectives on the critical role of root hydraulic function. Pre-dawn observations of Ψ and Ψ from multiple depths could reveal interspecific patterns in functional rooting depth – a trait that is difficult to measure by other means and partially responsible for model difficulty in capturing plant drought responses[62]. When complemented with data on Ψ and/or root sap flow, profile observations of Ψ would also illuminate the important but poorly understood consequences of hydraulic redistribution of water from wetter to drier soil layers through plant roots[63-64]. While root Ψ is difficult to measure with pressure chambers, it could be monitored more easily with psychrometers or other techniques for continuous observation of plant Ψ. Data on root Ψ, especially when paired with laboratory-derived root xylem vulnerability curves, would also be useful for understanding the dynamics of root hydraulic conductance, noting that roots may be among the most vulnerable components of the plant hydraulic system[65-66]. Finally, differences in Ψ and root Ψ could also improve our understanding of gradients in Ψ occurring at the root-soil interface[67].

Strategies to address the water potential information gap

Recent advances in measurement technology have substantially improved the ease and reliability of Ψ observations. In the lab, sensor improvement has reduced the time necessary to generate the “wet end” of the water retention curve[68]. A second instrument, typically a dew-point potentiometer, is required to capture the dry end of the curve, but this step proceeds relatively quickly. While the instrumentation and expertise necessary to characterize water retention curves may be siloed within soil science disciplines, this barrier could be easily overcome through cooperative arrangements and/or knowledge transfer. At the same time, technology is improving for more confident observation of Ψ in-situ[8]. Tensiometers, which are accurate when soil is relatively wet (e.g. Ψ > −0.1 MPa), are widely used in agricultural settings for the purposes of irrigation scheduling. In the drier range, soil matric potential can be measured using psychrometry or from dielectric measurements, with several commercial sensors available at a relatively low cost (e.g. the Teros 21 product, Meter Group). While the accuracy of sensors like these is greatest when Ψ is above −2 MPa, this is still lower than the wilting point of many plant species[8].

With respect to plants, psychrometers permitting continuous and long-term observation of both Ψ and Ψ are becoming more widely and commercially available (e.g. the PSY1 products, ICT International), drawing from a long history of psychrometric approaches for measuring plant water potential[69]. Stem psychrometers can now be deployed on branches and boles of some species for weeks to months at a time[55], and evidence is mounting that high-frequency Ψ and Ψ data can indeed improve our understanding of plant water use strategies and dynamics[55,70]. Psychrometers are still relatively expensive, best suited for broadleaf and non-resinous species, and sensitive to biases linked to temperature fluctuations and wounding effects. Thus, for now, psychrometer data is best viewed as complimentary to pressure chamber measurements. Nonetheless, for many plants, these instruments allow for the collection of Ψ and/or Ψ data at the hourly timescales necessary to be harmonized with observed carbon and water fluxes (e.g. from sap flux and flux towers) and to more rigorously test model frameworks.

Ultimately, addressing environmental questions at policy- and management-relevant scales requires the collection and standardization of observations across many sites. This need has motivated the recent development of many environmental observation networks, including highly-centralized initiatives like NSF’s National Ecological Observatory Network (NEON[71]), as well as more bottom-up networks like AmeriFlux[72] and FLUXNET[73] and the new international SAPFLUXNET network[74]. Other approaches include “network-of-networks” cyberinfrastructure like the International Soil Moisture Network,[13] which aggregates soil moisture observations from dozens of individual networks.

Both bottom-up and top-down approaches could be useful for building new Ψ networks. On the one hand, centralized and standardized deployment of new Ψ sensors, ideally in locations that are already nodes of other networks, would have the advantage of uniformity in instrumentation and data quality control that facilitates cross-site synthesis. On the other, a community-driven effort to aggregate and redistribute both existing and new Ψ data could follow the highly successful ‘coalition’ model employed by networks like AmeriFlux[72], increasing the discoverability of data while allowing room for innovation at the site level. Even a concerted effort to generate and/or collect laboratory-based water retention curves from existing network sites could substantially constrain how much of the non-linearity in the response of fluxes to observed soil water content can be explained by soil physics (e.g. see Fig. 4). The success of a water potential network would be maximized with: a) a focus on collecting data from sites that also support continuous plant- and/or stand-scale carbon and water fluxes, b) cyberinfrastructure to support the discoverability and distribution of these databases; c) a focus in at least some locations on within-site spatial heterogeneity in Ψ dynamics, to better understand of how many observation points (and at what depths) are necessary to substantially improve model skill; and d) training programs, such as summer short-courses or distributed graduate seminars, to transfer knowledge about how to interpret network observations and to share best practices for sensor deployment.

Even with well-developed observation networks, it is not possible to measure key physiological variables like Ψ everywhere and all the time. Thus, strategies for linking these variables to proxies observable from space are required for regional- and continental-scale work, with microwave remote sensing representing a particularly promising approach. Microwave observations can be used to determine vegetation optical depth (VOD), which is sensitive to plant water content[75] and should be monotonically related to Ψ[76,77]. Comparison of observed Ψ with either spaceborne[78] or tower-based[70] radiometry confirms that VOD and Ψ follow similar dynamics, especially after accounting for the effect of changing biomass and leaf area. However, the exact relationship between VOD and Ψ is influenced by vegetation type[76], and further study of this relationship is currently hindered by the sparsity of Ψ data.

Importantly, microwave remote sensing observations can be made at night, which raises the question: can nocturnal microwave remote sensing of Ψ be used to infer dynamics of root-zone Ψ? Answering this question requires a critical understanding of when and where pre-dawn Ψ is equilibrated with root-zone Ψ. This knowledge gap can be addressed with network observations of Ψ from psychrometry, or observations of plant and soil water potential collected in the same site, which could then guide the design and interpretation of both tower- and satellite-mounted microwave remote sensing systems. The approach will also require further refinement of retrieval algorithms for separating the contribution of plant and soil water content, for example by leveraging emerging approaches for the remote sensing of vegetation structure[77].

In conclusion, we have highlighted how more numerous, discoverable, and continuous observations of soil and plant Ψ can improve not only our conceptual understanding of biophysical processes throughout the soil-plant-atmosphere continuum, but also serve as a much-needed new tool for benchmarking and calibrating hydrologic and land-surface models and remote sensing products. While in-situ and site-specific observations of Ψ, Ψ, and Ψ may not yet be “easy,” recent advancements in sensor technology have certainly made them easier than in decades past. The time is right for a new focus on the collection of these data in the field, and the development of new networks to aggregate observations across sites complemented by new approaches for integrating these observations into Earth system models.

Methods

Water retention curve uncertainty:

The water retention curves in Figure 2 were created using the van Genuchten water retention curve model[11] relating Ψ to θ. As described in more detail in the Supplementary Information, most parameters of the model were held constant within each soil type, specified as the mean values reported in the updated ROSETTA pedotransfer function[18] (see Supplementary Table S1). The ‘n’ parameter was allowed to vary by randomly selecting a value from a uniform distribution bounded by ±1 standard deviation as reported for the ROSETTA PTF[18]. Overall, this was a conservative approach; drawing the values of n from the full distribution reported for each soil type expands the range of predicted Ψ by orders of magnitude.

The HYDRUS 1-D simulations:

Uncertainty in the water retention curve linked to pedo-transfer uncertainty (e.g. as Figure 2a–d) was propagated through predictions of Ψ and θ (at depths of 15 cm) and surface evapotranspiration (ET, cm day) using the HYDRUS 1D soil water dynamics model[79]. Fifty simulations were performed for the Bradford Woods deciduous forest site in south-central Indiana, where the HYDRUS 1D model had been previously calibrated[80]. In general, model settings were left unchanged, with a few exceptions as discussed in more detail in the Supplementary Information. The soil at Bradford Woods is characterized by a 40 cm depth AP horizon dominated by sandy loam, and a BW Horizon dominated by silt loam from a depth of 40 cm to 208 cm. The very bottom of the soil layer (depths 208 – 230 cm) was prescribed to be clay loam. The parameters of the van Genuchten model used in the HYDRUS simulations are shown in Supplementary Table S2, where again most were held constant, but n varied for the sandy and silt loam layers by drawing it from within one standard deviation of its distribution reported in the updated ROSETTA PTF[18]. The shaded areas in Figure 2e–f thus illustrate the resulting variation in ET, Ψ, and θ due solely to variability in n.

The ORCHIDEE GPP sensitivity analysis:

The ORCHIDEE land surface model (CMIP6 version)[31,32], which is the terrestrial part of the IPSL (Institute Pierre-Simon Laplace) Earth system model, was used to explore the sensitivity of modeled GPP to uncertainty in a wide range of parameters. ORCHIDEE relies on the van Genuchten model to calculate Ψ, as well as the hydraulic conductivity and diffusivity required to solve the Richard’s diffusion equation. ORCHIDEE discretizes the first 2 m of the soil column over 11 layers. For this experiment, we ran ORCHIDEE over three single mesh locations using local half-hourly forcing data to drive the model at each site (see Table Supplementary Table S3), and considered modelled GPP at a daily time-step. The sensitivity analysis results shown in Figure 3 were generated using Sobol’s method[30], using the SALib python package[81] to sample the parameter space and execute the SA algorithms. Briefly, the model was run using different parameter ensembles, with parameters varied within their reported ranges of uncertainty. Then, each modeled GPP timeseries was compared to GPP derived from flux tower observations. The variance of simulated GPP was then decomposed into fractions which can be attributed to each parameter tested. These results shown in Figure 3 capture both independent and interactive contributions of each parameter to the total variance. When interactions are removed, the independent contribution of water retention curve parameters is still significant, and actually increases for the semi-arid site (see details in Supplementary Section 3).

The AmeriFlux GPP analysis:

Half-hourly or hourly data from the four flux towers referenced in Figure 4 were acquired from the AmeriFlux network (ameriflux.lbl.gov) and subjected to a standardized quality control, gapfilling, and partitioning approaches. The sites and quality control procedures are described in more detail in Supplementary Table S5. The methods used to determine the relationship between GPP and soil moisture are similar to those previously used to explore the relationship between surface conductance and soil moisture[35]. Briefly, analysis was constrained to the peak of the growing season to limit bias linked to phenological variation in LAI. Estimates of Ψ for each site were determined from site-specific water retention curves[38,82-84]. The data were then sorted into nine bins representing the 15th, 30th, 45th, 60th, 70th, 80th, 90th, and 100th quantiles of the observed values of soil moisture content in each site. Within each bin, data were constrained to relatively high light (net radiation > 300 W/m2) conditions with VPD limited to 1 ≤ VPD ≤ 1.5 Pa in US-MMS, US-TON, and US-MOz, and 1.5 ≤ VPD ≤ 2 kPa in the more arid US-SRM site. The mean GPP, Ψ, and θ were then calculated for each bin using the filtered data, and normalized by the maximum bin-averaged value observed at each site.