Can limits of plant available water be inferred from soil moisture distributions?

Robust assessment of crop water availability requires effective integration of soil moisture data within the range of field capacity (θFC) to permanent wilting point (θPWP). Emerging needs for spatiotemporally dynamic θFC and θPWP are difficult to achieve with lab determinations. Therefore, we used long‐term data from 182 sites across the United States to evaluate whether soil moisture extremes defined by 95th and 5th percentiles represent θFC and θPWP, respectively. Soil moisture extremes and lab‐measured θFC and θPWP were well correlated (R2 = 0.71−0.92), however, both 95th and 5th percentiles overestimated θFC and θPWP at most depths (RMSE = 6%–16% vwc). Percentiles of soil moisture distribution that corresponded to lab‐determined θFC and θPWP varied widely and were a function of precipitation received at the site and site‐ and soil‐depth specific clay content. These findings imply that while θFC and θPWP may not be broadly represented by soil moisture extremes (95th and 5th percentiles), there may be potential to statistically infer the positioning of θFC and θPWP within long‐term soil moisture distributions using biophysical determinants such as aridity and soil characteristics.


INTRODUCTION
Soil moisture monitoring is a critical prerequisite for assessing droughts/floods, land-atmospheric interactions, hydrological and surface energy budgets, ecosystem functions and services and other applications.In agriculture, soil moisture data are used to inform irrigation management and assess field accessibility/trafficability.Soil moisture is also used to predict evapotranspiration, runoff (streamflow), soil and water quality, and water use efficiency.To fulfill these needs, investments in soil moisture sensing efforts have increased in recent years, including (1) sensor technology adoption by irrigators (USDA-IWMS, 2018); (2) public national and state-level soil and environmental monitoring networks (Cosh et al., 2021;Ochsner et al., 2013;Quiring et al., 2016); and (3) satellite missions providing spatiotemporally continuous data (Babaeian et al., 2019;Entekhabi et al., 2010).Actual water availability for crops, however, is not solely a function of soil moisture but also its magnitude relative to field capacity (θ FC :VWC at soil matric potential of −33 kPa, where VWC is volumetric water content) and permanent wilting point (θ PWP :VWC at soil matric potential of −1500 kPa).Normalized soil water indices based on θ FC and θ PWP are used to track, report, and manage water stress and schedule irrigation based on crop-specific thresholds (Jensen & Allen, 2016), underscoring the need for high-fidelity θ FC and θ PWP Agric Environ Lett. 2023;8:e20113.wileyonlinelibrary.com/journal/ael2 1 of 6 https://doi.org/10.1002/ael2.20113estimates.Although definitions in terms of specific soil matric potential values may not be consistent across all soils and vegetation (Garg et al., 2020;Sperry et al., 1998), use of θ FC and θ PWP is widespread in crop water stress applications as heuristic tools.Robust lab-determination of θ FC and θ PWP using Tempe cells and pressure plates (Richards & Fireman, 1943) may not cater to extensive need for spatiotemporally dynamic θ FC and θ PWP information given substantial cost, time, and skill involved.Remotely sensed soil moisture data are being increasingly employed in monitoring and decision tools, requiring continuous estimates of θ FC and θ PWP .Moreover, soil organic carbon gain under climate-smart soil management practices (Bagnall et al., 2020;Deines et al., 2023Deines et al., , 2019;;Lehmann et al., 2020) is expected to improve θ FC and θ PWP (Lal, 2020), necessitating their temporal characterization and mapping.
Amidst these emerging needs, the idea of leveraging long-term soil moisture observations to identify operational brackets of plant available water capacity lends some optimism.This concept has been relied on for assessing soil moisture-based drought conditions by developing indices (e.g., soil moisture index, soil water deficit, fraction of available water, soil water depletion, fraction of transpirable soil water) that are normalized by soil moisture extremes (AghaKouchak, 2014;Espinoza-Dávalos et al., 2016;Ford & Quiring, 2019;Ford et al., 2015Ford et al., , 2016;;Herold et al., 2016;Krueger et al., 2019;Leeper et al., 2021;Ochsner et al., 2013;Zhao et al., 2020).Extremes defined by 95th and 5th percentiles of θ observations have been used as proxies of θ FC and θ PWP , respectively (Cao et al., 2022;Hunt et al., 2009Hunt et al., , 2016;;Liu et al., 2017;Martínez-Fernández et al., 2015).However, it is yet to be determined if these extremes are of pragmatic value for representing plant available water capacity.In this letter, we leverage large-scale in situ soil moisture monitoring networks and associated soil characterization efforts to evaluate the hypothesis that percentiles determined from long-term soil moisture distributions can be inferred as θ FC and θ PWP across a wide spectrum of soil and environmental diversity.

MATERIALS AND METHODS
Daily soil moisture data at depths of 5, 10, 20, 50, and 100 cm were obtained for all Soil Climate Analysis Network (SCAN) and United States Surface Climate Observing Reference Networks (USCRN) sites for which corresponding lab-measured θ FC and θ PWP were available (Figure 1a), resulting in 182 sites in total (116 SCAN sites and 66 USCRN sites).At each site, all available θ records were included to capture long-term variability.Time series of θ at each site consisted of missing records, the proportion of which varied significantly from site-to-site.Any missing records for any depth in the soil profile were removed.Since only observed θ were to be used to infer distributions from sufficiently long time series available

Core Ideas
• Assessing water availability requires integrating soil moisture (θ) observations with θ FC and θ PWP .• Strong need exists for spatiotemporally dynamic θ FP and θ PWP estimates for diverse applications.• Despite reasonable association, 95th and 5th θ percentiles had limited capability to accurately estimate θ FC and θ PWP across 182 sites.• Predictive performance varied with site aridity and soil depth, requiring site-specific θ percentiles.• Site-specific percentile-based θ FP and θ PWP estimates were influenced by precipitation and clay content.
at each site, no gap filling/interpolation was attempted.Sensor performance is negatively affected in frozen soils (Zhang et al., 2011) and insignificant soil water extraction is observed due to little or no vegetation growth in colder temperatures.Thus, the data were subset to only include records from April to October.Upon filtering, datasets at individual sites ranged from 9 to 24 years.At each site and each constituent depth, 95th and 5th percentiles as well as percentiles corresponding to lab-determined θ FC and θ PWP were calculated using quantile function in R (R Core Team, 2023).True determination of θ FC and θ PWP at SCAN and USCRN sites was conducted using lab-measured θ and soil water (matric) potential and consequent development of soil water retention curves.For SCAN, we used the soilDB package (Beaudette et al., 2023) to obtain site-specific van Genuchten model (Van Genuchten, 1980) parameters measured by NRCS and determine θ FC and θ PWP .For USCRN, we obtained predetermined θ FC and θ PWP presented by Wilson et al. (2016), as averaged from soil analysis at three separate pits surrounding each site.We used simple goodness of fit (R 2 ) and root mean squared error (RMSE) to evaluate agreement between measured θ FC and θ PWP and its soil moisture distribution-based proxies.We also calculated the percentiles corresponding to the lab-measured θ FC and θ PWP within the observed θ distribution at each site to understand their relative positioning within long-term records.Long-term (1991Long-term ( -2020) ) mean annual precipitation at each monitoring site was obtained from PRISM database (PRISM Climate Group, 2023), and ranged from 850 to 1765 mm across all sites.

RESULTS
The 95th and 5th percentiles of soil moisture observations correlated well (R 2 = 0.71−0.92)with lab-determined θ FC and θ PWP (Figure 1b), implying that site-to-site variation in extremes is sensitive to inter-site differences in θ FC and θ PWP .
In general, 95th percentiles had slightly better correlation to θ FC than what was observed for θ PWP at all five depths.Despite the correlations, it does not seem advisable to use 95th and 5th percentiles as proxies of θ FC and θ PWP given substantially high RMSE.θ FC was overestimated by 13%-22% when relying on 95th percentile, showing an RMSE of 11%-16% VWC across soil depths.Similarly, θ PWP was overestimated by 2%-53% with RMSE ranging from 6% to 13% across soil depths.However, significant underestimation in θ PWP was observed for 5 cm layer (33%).Performance generally decreased with increasing soil depths.Only 10% and 26% of the sites showed 95th and 5th percentiles that were within 2% VWC of lab-determined θ FC and θ PWP , respectively.Similarly, around 50% and 25% of the sites showed deviation <5% VWC when percentile-based estimates were compared to lab determined θ FC and θ PWP , respectively.The differences between field-measured soil moisture extremes and lab determinations can be partially attributed to the generally low soil water extraction by the sparse and shallow-rooted surface covers at these reference sites.At most sites, the surface cover is mainly grass, but also include other vegetation types such as sparse desert grasses and shrubs, and prairie grasses.Although surface covers are representative of the site's native land use, they are generally shallow rooted, preventing soil moisture from reaching lower bounds.
Given the ineffectiveness of blanket extremes (i.e., 95th and 5th percentiles) across all sites to represent θ FC and θ PWP , it is appropriate to ask the question "where do labdetermined θ FC and θ PWP lie within long-term soil moisture distribution at each site and depth therein?"We found that percentiles representing lab-determined θ FC and θ PWP varied largely across sites and depths (Figure 1c).Median θ FC values range roughly between 40 and 80 percentiles and decrease for deeper soil depths.Median θ PWP values range roughly between 0 and 20 percentiles across soil depths.Soils remain saturated for longer periods of time as depth increases, explaining lower percentiles corresponding to lab-determined θ FC .Median θ PWP percentiles of zero at deeper layers imply that soils never dried sufficiently to attain θ PWP .
The site-specific percentiles that were appropriate for representing θ FC and θ PWP were strongly determined by site characteristics.Multiple linear regression models developed between percentiles representing depth-specific θ FC and θ PWP and mean annual precipitation, clay content, sand content, and soil organic carbon at USCRN sites revealed that both precipitation (range:850-1765 mm) and clay content (range:1%-66%) were statistically significant drivers (p < 0.001) of θ FC percentiles.The same was true for θ PWP percentiles, except that only precipitation was significant at 20, 50, and 100 cm depths.Notably, the percentiles corresponding to θ FC and θ PWP decreased as precipitation increased and clay content decreased.Together, precipitation and clay content explained 34%-53% and 10%-49% of the variance in θ FC and θ PWP percentiles, respectively.Thus, both soil texture and site wetness determined the level of percentile that was suitable for a given site.

Sensing and sampling uncertainty
Direct comparisons of soil moisture observations with labmeasured θ FC and θ PWP should be interpreted with caution, given the uncertainties associated with both.First, in situ sensor outputs are subject to complex uncertainties owing to soil texture, salinity, sensor technology, and others (Datta et al., 2018;Leib et al., 2003;Sharma et al., 2021), including sensors employed at SCAN and USCRN sites, that is, Hydra (Stevens Water Monitoring Systems, Inc.) soil probe (Jabro et al., 2018;Seyfried et al., 2005;Vaz et al., 2013).These discrepancies between soil moisture sensor outputs and θ FC and θ PWP information derived from lab/surveys have also been noted by Evett et al. (2019), and can lead to misrepresentation of crop water availability.With much diversity in sensor technology used on commercial farms, producers rarely have the capacity or access to sensor-specific and soil-specific calibration functions.Second, spatial soil variability encountered at fine scales make selection of soil sampling locations for lab measurement of θ FC and θ PWP challenging, further adding uncertainty to interpretation of actual crop water availability.For example, soils at USCRN sites were characterized from samples originating from three pits surrounding the site (Wilson et al., 2016), and these samples had a mean coefficient of variation of 0.16 and 0.12 for θ FC and θ PWP , respectively.Thus, the differences between soil moisturebased percentiles and lab-determined θ FC and θ PWP cannot be entirely classified as errors, given the uncertainties associated with sensor output as well as spatial variance of lab-determinations.

Significance and applications
This research has shown that soil moisture extremes are quite sensitive to the θ FC and θ PWP of the soil being monitored, however, they are subject to large error when compared to lab determinations.Thus, drought tracking via soil moisture indices that rely on 95th and 5th percentile to represent θ FC and θ PWP (Hunt et al., 2009;Martínez-Fernández et al., 2015, 2016;Spennemann et al., 2020) should be judged with caution, especially when large spatial domains are assessed with significant variation in aridity and soil texture.Sitespecific percentile levels would be superior to using a fixed percentile level (e.g., 95th and 5th) for all sites.Site-to-site differences in suitable percentiles used to represent θ FC and θ PWP were governed by site precipitation and clay content in the soil.At individual sites, operational limits of plant available water can be inferred using a combination of statistical and visual approaches.It is important that sufficient soil moisture records from contrasting regimes are included (e.g., immediately following heavy wetting event and during dry spells).In most pragmatic situations, long-term records are scarce, however, research shows that 3 years of data can result in robust percentiles (Ford et al., 2016).It is relatively easy to visually identify apparent θ FC from time series data following major wetting events for calculation of soil water depletion (Vories & Sudduth, 2021).By filtering soil moisture records to only include datapoints soon after major wetting events, the general tendency of a soil's upper bound of plant available water can be determined.However, at wetter and clayey sites, soil can remain near θ FC for extended periods of time, which means that lower percentiles may be better suited than 95th percentile.What these appropriate percentile levels are will depend on site wetness and soil texture, as observed from distributions presented in Figure 1c.Similarly, subsurface conditions during concurrent dry spells, periods of high evaporative demand and root water uptake can be statistically studied to determine the lower bound.Although satellite-derived soil moisture products have coarser resolutions, similar approaches can be used for inferring large-scale operational estimates of θ FC and θ PWP .For instance, satellite missions such as SMAP have currently assimilated 8 years of root zone soil moisture data, which may be sufficient to identify operational bounds at grid-level.Future research may evaluate such estimates from remotely sensed soil moisture products (Peng et al., 2021) against θ FC and θ PWP derived from gridded soil survey data (Soil Survey Staff, 2023).Tracking soil moisture extremes over time may help capture and monitor anticipated improvements in plant available water capacity due to widespread adoption of soil health (also known as regenerative agriculture, climate-smart) practices.

AU T H O R C O N T R I B U T I O N S
Meetpal S. Kukal: Conceptualization; data curation; formal analysis, funding acquisition; investigation; methodology; project administration; software; visualization; writingoriginal draft.Suat Irmak: Investigation; project administration; resources; supervision; validation; writing-review and editing.

C O N F L I C T O F I N T E R E S T S T A T E M E N T
The authors declare no conflicts of interest.

D A T A AVA I L A B I L I T Y S T A T E M E N T
The datasets used for the analyses are publicly available at the sources mentioned in Section 2.

F
I G U R E 1 (a) Location ofSCAN and USCRN sites included in this study; (b)  regression between soil moisture extremes (90th and 5th percentiles) and laboratory-determined volumetric water content (VWC) at field capacity (θ FC ) and permanent wilting point (θ PWP ) at each soil depth; and (c) site-level distribution of percentiles corresponding to lab-determined θ FC and θ PWP across all sites at each soil depth.SCAN, Soil Climate Analysis Network; USCRN, United States Surface Climate Observing Reference Networks.