Soil Moisture Profiles of Ecosystem Water Use Revealed With ECOSTRESS

While remote sensing has provided extensive insights into the global terrestrial water, carbon, and energy cycles, space‐based retrievals remain limited in observing the belowground influence of the full soil moisture (SM) profile on ecosystem function. We show that this gap can be addressed when coupling 70 m resolution ECOsystem Spaceborne Thermal Radiometer Experiment on Space Station retrievals of land surface temperature (LST) with in‐situ SM profile measurements. These data sets together reveal that ecosystem water use decreases with depth with 93% of sites showing significant LST coupling with SM shallower than 20 cm while 34% of sites have interactions with SM deeper than 50 cm. Furthermore, the median depth of peak ecosystem water use is estimated to be 10 cm, though forests have more common peak interactions with deeper soil layers (50–100 cm) in 37% of cases. High spatial resolution remote sensing coupled with field‐level data can thus elucidate the role of belowground processes on land surface behavior.


Introduction
Earth's land surfaces play a major role in the climate system.Both soil and vegetation regulate water, carbon, and energy fluxes between the land and atmosphere, influencing weather patterns and surface temperatures (Yang et al., 2023).Satellite remote sensing and land surface models provide insights into land surface function under climate change and extremes (Green et al., 2019;Humphrey et al., 2018;Zhang et al., 2023).However, while aboveground processes are well-observed, our understanding of subsurface effects on ecosystem function remains a major uncertainty in characterizing and predicting the terrestrial water, carbon, and energy cycles (Vereecken et al., 2022).
Soil evaporation and transpiration are surface energy fluxes that drive surface climate and influence atmospheric boundary layer processes.Soil exfiltration uses moisture from the shallow soil layers (Aminzadeh & Or, 2014;Lehmann et al., 2008) whereas transpiration relies on vegetation root water uptake that can continue deep into the rootzone (Jackson et al., 1996;Schenk & Jackson, 2002).Transpiration might contribute proportionally more to global land surface energy fluxes than soil evaporation according to isotopic tracers (Good et al., 2015;Jasechko et al., 2013;Yang et al., 2023).Given that water availability is a major driver of these global water and energy fluxes (Madani et al., 2017;Nemani et al., 2003), it is imperative to understand the role of the full rootzone profile on land surface function.

RESEARCH LETTER
With early discoveries of very deep vegetation rooting (Nepstad et al., 1994;Phillips, 1963), there have been efforts to understand the role of these deeper roots on ecosystem function.Rooting profiles often extend to 1 m and below (Tumber-Dávila et al., 2022), sometimes into bedrock (McCormick et al., 2021), and are controlled by water table depth (Fan et al., 2017).It is likely that these deeper roots are especially important for ecosystem resilience during droughts (Miguez-Macho & Fan, 2021;Stocker et al., 2023).Nevertheless, while rooting depth databases continue to expand (Tumber-Dávila et al., 2022), the relative role of deeper versus shallower soil layers on plants and broader ecosystem function remains an open question (Nippert & Holdo, 2015).
Our current tools are limited in their ability to observe the control of soil moisture (SM) variations across the full rootzone on spatially integrated ecosystem function.Traditionally, the role of soil water availability on energy fluxes can be understood by evaluating evaporation versus SM relationships (Dong et al., 2023;Feldman et al., 2019;Seneviratne et al., 2010).However, investigations of the role of belowground soil and roots on terrestrial fluxes remains sparse (Javadian et al., 2023).This is in part because rooting profile and soil water isotopic tracer (such as, deuterium) measurements are heterogeneous in global coverage (Tumber-Dávila et al., 2022) and are not always co-located with flux or SM measurements.Recent insights have mainly been gained at well-instrumented, but sparse field locations (Javadian et al., 2023;Kulmatiski & Beard, 2013;Nippert et al., 2012).Furthermore, the longest wavelength microwave observations currently available from satellites (i.e., at L-band frequency) can typically only sense shallower soil layers (Akbar et al., 2018;Feldman et al., 2023;Short Gianotti et al., 2019).Multi-layer modeled reanalysis SM can be used, though they are subject to model biases especially in data-poor regions (W.Li et al., 2021).Gravimetric satellite measurements provide insights into deeper soil stores, but these data are limited by coarse spatial and temporal resolutions and uncertain subsurface representation (Humphrey et al., 2018;Khanal et al., 2023;Rodell & Famiglietti, 2001).There are also efforts to evaluate subsurface moisture with infrared satellites, though they are empirical and context dependent (Babaeian et al., 2018;Burdun et al., 2023;Hain et al., 2009).
While networks of SM profile measurements exist, they are typically not collocated with energy flux measurements (Bell et al., 2013;Dorigo et al., 2021;Schimel et al., 2015).However, remote sensing retrievals of energy fluxes are potentially at a high enough resolution to approach the representation scale of in-situ SM sensor measurements (Famiglietti et al., 2008;Goward et al., 2001;S. Li & Sawada, 2022).For example, ECOsystem Spaceborne Thermal Radiometer Experiment on Space Station (ECOSTRESS) retrieves land surface temperature (LST) at a 70 m spatial resolution (Fisher et al., 2020).Additionally, with its sub-daily to 5-day revisits, it measures at varying times of day, especially at afternoon hours when surface fluxes are larger and more responsive to climate (Bateni & Entekhabi, 2012;Feldman et al., 2022;Panwar & Kleidon, 2022).ECOSTRESS can capture ecosystem heterogeneity and function across the lower and mid-latitudes (Cooley et al., 2022;Xiao et al., 2021), and combining these data with point measurements of SM might enable understanding the role of SM profile variations on terrestrial water, carbon, and energy fluxes.
Here, we evaluate to what degree SM across the rooting profile (from US Climate Reference Network (USCRN) and Soil Climate Analysis Network (SCAN) sites) interacts with water and energy fluxes as observed at high resolution from ECOSTRESS.We ask: can coupling between high resolution satellite retrievals and in-situ SM measurements indicate depths of ecosystem water use?What are the vertical soil profiles of ecosystem water use across vegetation types?ECOSTRESS retrievals integrate both soil and vegetation components and thus allow us to evaluate the contribution of different parts of the SM profile to total evaporation.

Data
Level 2 land surface temperatures from ECOSTRESS (ECO2LSTE Version 1), an instrument onboard the International Space Station (ISS), were obtained over the continental US (CONUS) (Hook & Hulley, 2019).ECOSTRESS LST data are gridded at a 70 m resolution with retrieval temporal resolution of several hours to 5 days, with LST retrievals distributed throughout hours of the day.Since the ISS has a precessing orbit, ECOSTRESS measures at different times of day (Cawse-Nicholson, Anderson, et al., 2021).ECOSTRESS Level 3 evapotranspiration (ET) retrievals from the PT-JPL algorithm are used as an ancillary data set here ( Fisher, 2019).All data are obtained between 2019 and 2023 and from April to October when surface energy fluxes are expected to be largest in CONUS.
CONUS is selected as the domain because of its expansive in-situ SM networks with standardized depths of measurements across sites and different vegetation types.Specifically, hourly SM measurements are obtained from two standardized SM networks: USCRN and SCAN (Bell et al., 2013;Dorigo et al., 2021).These sites have consistent SM probe measurements at 5, 10, 20, 50, and 100 cm depths at hourly timesteps that convert dielectric measurements into volumetric units (m 3 m 3 ).Sites were removed that did not provide measurements at all five depths, leaving 228 sites for the analysis.Quality flags were used to remove hours with less reliable measurements.
For each site, a timeseries of LST (and ET) measurements was extracted for the ECOSTRESS grid cell containing it.Additionally, we used only USCRN and SCAN SM measurements at the hour closest to the timing of the ECOSTRESS overpass.The standard International Geosphere-Biosphere Program land cover types were used to group vegetation types (Kim, 2013), with the exception that all forest types are grouped into a single category.Finally, sand and clay fractions at the 5 cm depth and 250 m resolution were obtained from SoilGrids 250 m to evaluate the role of soil texture and hydraulics (Hengl et al., 2017).

Ecosystem Water Use Analysis
LST is closely representative of the surface energy balance (Bateni & Entekhabi, 2012;Panwar et al., 2019).We thus argue that LST-soil moisture interactions can capture the total ecosystem response to water availability.Namely, higher SM availability should result in more evaporation and a lower LST, unless highly energy limited.The general relationship between surface temperature and surface energy fluxes is illustrated in an example from the FLUXNET US-SRM site near Tucson, Arizona (Figures 1a and 1b) (Scott et al., 2015).In this water-limited grassland, afternoon soil temperature (averaged between 12 and 3 p.m.) correlates negatively with afternoon latent heat flux (r = 0.53; p < 0.05) and positively with sensible heat flux (r = 0.66; p < 0.05) (Figures 1a and  1b).These relationships might differ when considering ECOSTRESS retrievals of local skin temperature, which include vegetation top of canopy temperature.
These points motivate our use of afternoon ECOSTRESS LST as a proxy for ET here.Use of ECOSTRESS LST is desirable because these data are not influenced by land surface process modeling assumptions or errors (present in Level 3 ET with their use of stress functions), and thus ECOSTRESS LST is used in the main analysis.Furthermore, ECOSTRESS LST provides a skin surface temperature over the entire grid cell.If there is more vegetation cover, the vegetation temperature represents more of the LST signal.This is advantageous for our analysis because we are interested in the total ecosystem response to water availability.
Note that conditions in Figures 1a and 1b may represent a best-case scenario for use of ECOSTRESS LST as an ET-proxy given that the site is highly water-limited.Under energy and aerodynamically limited regimes associated with higher water availability, these variables can decouple.We therefore supplementarily repeat the analysis with ECOSTRESS L3 ET to evaluate robustness of results, though we explore nuances with these data (Text S1 in Supporting Information S1).Morning to afternoon diurnal temperature amplitudes are likely more representative of ET than afternoon LST alone (Feldman et al., 2019;Panwar & Kleidon, 2022).However, using diurnal ECOSTRESS LST rates of change (limited the analysis to only days with 2 or more retrievals) would reduce the LST-soil moisture retrieval pairs here by 90%-95%, critically hindering our analysis.
Contemporaneous ECOSTRESS LST retrievals and in-situ SM pairs between 12 and 3 p.m. are used to assess interactions between LST and SM.Specifically, for each site and each of the five soil depths, we compute: where β θ,d is the sensitivity of LST to SM variations at soil depth d; θ is SM at d; β 0 is the y intercept; ε is the residual error.We interpret a statistically significant negative β θ,d as the soil water availability control at depth d on land surface function, including both the soil and vegetation components.The correlation is also computed as a metric of variance explained by scaling β θ,d by σ(SM)/σ(LST) where σ is standard deviation.Note that statistical significance is independent of whether correlation or slope is used.
We refer to negative LST-soil moisture correlations as indicators of "ecosystem water use," though our analysis does not identify causality.Causal inference frameworks can elucidate these directional effects, but their use is hindered here by low sample sizes of LST-soil moisture pairs and irregular temporal sampling.Nevertheless, the coupling between LST and SM should mostly indicate depths of SM used for ET.Namely, increased SM in a layer that is used for ET should result in a reduction in LST.Similarly, an increase in LST due to radiation or advection effects can cause more ET and reduce SM.Scenarios exist for which these directions of causality do not hold, but we expect these cases are less dominant.
Of the soil depths with statistically significant β θ,d values, the depth of the most negative β θ,d is considered the soil depth of peak ecosystem water use.β θ,d is chosen here as the metric of sensitivity over correlation because correlation can be the same for different slopes.However, we also note that the SM temporal standard deviation tends to decrease with depth (Gruber et al., 2013).This decrease in variability might artificially increase the β θ magnitude in deeper layers, and thus bias peak coupling to deeper layers.We therefore also estimate the peak ecosystem water use using the maximum correlation magnitude to understand the sensitivity of peak ecosystem water use estimates to the choice of these metrics.
Note that LST (or ET) relationships with SM tend to be nonlinear with nearly linear relationships in the waterlimited (drier) regime and little relation between these variables in the energy-limited (wetter) regime (Bassiouni et al., 2020;Fu et al., 2022;Koster et al., 2009;Seneviratne et al., 2010).However, the linear LST-soil moisture slope will still capture bulk sensitivity of energy fluxes to SM in the water-limited regime despite the presence of some energy limitation (Dong et al., 2020).
An example analysis is shown in Figure 1 for an Arizona grassland (Figures 1c-1h).The 10 and 20 cm soil depths have statistically significant (p < 0.05) LST-soil moisture relationships, interpreted as ecosystem water use at these depths.Additionally, the depth of peak water use for this site is 10 cm since it has the largest magnitude β θ,d across all depths with statistically significant relationships.

Data Quality Control
Correlation magnitudes between ECOSTRESS LST and in-situ SM are unlikely to approach unity for several reasons: the spatial scale mismatch between in-situ SM measurements and satellite LST retrievals, LST retrieval noise due to cloud cover and satellite measurement artifacts, some non-linearity in LST-soil moisture relationships for wetter locations, factors other than SM influencing LST, and the use of LST-soil moisture pairs from different hours of the afternoon (given that radiation has its own impact on LST).We partially mitigate several of these factors here.
LST retrievals below 273 K (as induced, for example, by frozen ground and cloud cover) were removed.Additionally, very high LST retrievals occur seemingly randomly at some sites.Visual analysis of several of these cases revealed ISS solar arrays entering the ECOSTRESS field of view.Removal of retrievals above 340 K greatly reduced these cases.Furthermore, only dry days were considered by removing days when the 5 cm SM increased by more than 0.01 m 3 /m 3 (a detection threshold above noise indicating that moisture was added to the ecosystem) since the last data pair.This mostly removes rainfall events and interception contributions to LST that do not directly represent SM control on surface energy fluxes.Results do not appear sensitive to this threshold.Sites with less than 10 samples after quality control were removed.
We found that the formal ECOSTRESS quality flags were overly restrictive for our application.Namely, including quality flags reduces the median number of LST-soil moisture data pairs from 33 to 21 (between 12 and 3 p.m.), thus reducing the percentage of sites with significant LST-soil moisture interactions from 39% (90 sites) to 20% (46 sites).When we visually evaluated the cause of this reduction, we found many false positive detections of clouds; the flags removed potentially good quality LST retrievals at individual sites where clouds were only in the vicinity (example in Text S2 of Supporting Information S1).Furthermore, removing LST retrievals below 273 K and using only dry days were already effective methods for removing poor quality data.Nevertheless, the analysis was repeated supplementarily using ECOSTRESS quality flags to assess their impact on the results.

Methodological Limitations and Additional Tests
Given that these are correlation-based analyses as demonstrated in Figure 1, we acknowledge several barriers that will obscure the interpretation of SM's "control" on LST and ET and how this control varies with soil depth.
Soil moisture is vertically correlated within the soil column to various degrees (Akbar et al., 2018;Feldman et al., 2023;Ford et al., 2014).Infiltration, soil exfiltration, and soil-root hydraulic processes like hydraulic redistribution tend to couple SM across depths (Cardon et al., 2013;Nadezhdina et al., 2010).Additionally, if root water uptake reduces SM at a given depth, neighboring soil layers will also dry out.Indeed, for the sites evaluated in the analysis, SM daily anomalies between sensors at 5 and 50 cm depths have a median correlation of 0.62, while this correlation is 0.68 between 50 and 100 cm depths.This coupling between SM of different layers may increase LST-soil moisture coupling at depths that are not directly contributing to root uptake (transpiration) or soil evaporation, while potentially dampening the signal of the layers that do contribute directly.Strong coupling between SM layers thus acts to create a uniform distribution of LST-soil moisture coupling.However, if we find profiles of water use in our analysis that are different from uniform, then physical processes are likely overcoming these limitations.
The β θ,d magnitude will be reduced for energy-limited ecosystems which can obscure the soil profile of ecosystem water use.As such, high energy-limitation at such a site might show no LST-soil moisture coupling across the different layers, despite some ecosystem use of SM.Therefore, in wetter, energy-limited locations, our approach might be less applicable.However, some water-limitation should be detected across CONUS in the northern hemisphere spring and summer months of our study even in wetter locations (Schwingshackl et al., 2017;Sehgal et al., 2021).We thus expect that significant LST-soil moisture interactions will still be detected across different soil layers, especially since the same evaporative regime tends to persist throughout the vertical soil column (Dong et al., 2022).
The seasonality of SM and LST can impart spurious relationships that can bias LST-soil moisture strength of coupling (Tuttle & Salvucci, 2017).These effects are partly reduced by focusing the analysis on April-October.However, we perform a supplementary analysis with daily anomalies by removing the seasonal cycle from LST to SM.To compute the seasonal cycles, we use smoothed monthly means across the years for ECOSTRESS LST and smoothed daily averages for the SM time series, as in previous works (Feldman et al., 2019).These deseasoned results are not in the main analysis because there is often a low sample size and higher uncertainty in computing a seasonal cycle across years at a given location.
We assume that the point-scale SM measurement is representative of the ECOSTRESS 70 m grid.However, SM is highly variable in space (Western & Blöschl, 1999).Therefore, using in-situ sensor measurements to represent integrated SM at the ECOSTRESS 70 m grid scale will introduce some error (Crow & Wood, 2002;Gruber et al., 2013).Since these representation errors will effectively be present as random measurement error of the SM grid average, low spatial representativeness of the in-situ SM would force β θ,d to zero.As such, any significant LST-soil moisture relationships found here would overcome these upscaling errors.Furthermore, since in-situ SM error tends to decrease (with representativeness increases) with depth (Gruber et al., 2013), the errors will force β θ,d magnitudes to erroneously increase more with depth, suggesting deeper ecosystem water use than reality.

Results
The most salient feature of our approach is the ability to view and evaluate soil profiles of ecosystem water use (Figure 2).Of the 228 sites, 39% (or 90) have a significant, negative LST-soil moisture relation at one or more soil depths (Figure 2a; p < 0.05).As is visually evident across these sites, there is a high occurrence of upper soil layer water use across biomes (Figure 2).Some differences appear, however.The forested sites have a relatively higher occurrence of significant LST-soil moisture interactions at deeper 50-100 cm soil layers (Figure 2b).By contrast, the grassland and cropland sites exhibit more frequent LST-soil moisture interactions in shallow soil layers (Figures 2c and 2e).LST-soil moisture correlations at the peak water use depths have a median of 0.41 (interquartile range of 0.48 to 0.37) with a reduction at wetter, energy-limited sites as expected (Figure S1 in Supporting Information S1).
These points are further illustrated in Figure 3. Ninety-three percent of sites have significant LST-soil moisture interactions at 20 cm or shallower (Figure 3a).There is also substantial coupling of LST with deeper soil layers: 34% of sites have cases of ecosystem water use at or deeper than 50 cm (Figure 3a).Similarly, the depth of peak ecosystem water use is typically in shallow soil layers across all sites (Figure 3b).The median depth of peak ecosystem water use is 10 cm, with a 25th percentile of 5 cm and 75th percentile of 20 cm (Figure 3b).Eighty percent of sites have their depth of peak ecosystem water use at 20 cm or shallower, while the other 20% have this depth at 50 cm or deeper (Figure 3b).
Vegetation types differ in their soil depths of LST-soil moisture interaction.While all vegetation types together show greater than 68% occurrence of LST-soil moisture interaction in the 5 cm layer, forests have a higher (38%) occurrence of interactions in the 50 cm layer than the herbaceous vegetation types (<26%) (Figure 3a).Even more pronounced are the differences in their depths of peak ecosystem water use; grasses and crops have their peak interaction in mostly shallower layers, or at 20 cm or shallower in at least 87% of cases (Figure 3b).This occurrence falls to 63% for forests, with the other 37% of cases having peak interaction in deeper layers (50 cm or below).Similarly, the median depth of peak ecosystem water use is 20 cm in forests but decreases to 5 cm in grasslands and 10 cm in croplands (Figure 3b).
While there are not clear impacts of sand and clay fraction across all soil depths, finer soils appear to be associated with less LST-soil moisture coupling in deeper soil layers.For example, the peak maximum water use occurs in 10% of cases below 50 cm in the sites with the highest clay fractions (top third percentile), compared to 29% in the lowest clay fractions (bottom third percentile) (Figure S2 in Supporting Information S1).A similar conclusion is drawn with sand fractions (Figure S3 in Supporting Information S1).
The overall result that interactions are more common in shallower layers remains when repeating the analysis using ECOSTRESS ET (Figure S4 in Supporting Information S1), LST anomalies with the seasonal cycle removed (Figure S5 in Supporting Information S1), and LST at morning (6:00-10:00) and evening (16:00-22:00) hours (Figures S6 and S7 in Supporting Information S1).Differences in results occur in some cases for specific vegetation types, though this is likely due to differences in the sites sampled.Additionally, we confirm that there are no sampling biases due to denser sampling in the shallower soil layers (5, 10, and 20 cm); the same qualitative conclusions are obtained when considering only SM sensors at 5, 50, and 100 cm depths (Figure S8 in Supporting Information S1).The results are qualitatively similar when repeating the analysis after removing LST data using ECOSTRESS quality flags (Figure S9 in Supporting Information S1).Finally, repeating the analysis using LSTsoil moisture correlation to determine these peaks results in similar findings though with more frequent peak ecosystem water use in shallower soil layers as expected (Figure S10 in Supporting Information S1; see Section 2).

Discussion
Several points provide confidence in our determined soil depths of ecosystem water use.First, the determined pattern of decreasing LST-soil moisture interactions with depth is expected (Figure 3).Soil evaporation mainly influences shallower layers (Aminzadeh & Or, 2014;Lehmann et al., 2008).Rooting density and water use profiles for transpiration also tend to decrease with depth (Jackson et al., 1996).Indeed, while rooting profiles extend deeper than the shallower soil layers, many field studies find active root water uptake from the upper soil layers with progressive decreases with depth (Case et al., 2020;Feldman et al., 2023;Hahn et al., 2021;Kulmatiski & Beard, 2013).Furthermore, forested ecosystems are found to have deeper water use (Figures 2 and 3), which is expected because isotopic tracer experiments find deeper tree root water uptake (Asbjornsen et al., 2008;Case et al., 2020;Kulmatiski et al., 2010;Schulze et al., 1996).The deeper water use in forests can also be due to greater vegetation cover and thus less shallow soil evaporation contribution to total ET (Text S3 in Supporting Information S1).Additionally, sites with finer soil textures also have reduced LST-soil moisture coupling in deeper soil layers (Figures S2 and S3 in Supporting Information S1), likely a result of higher soil hydraulic resistances reducing infiltration to deeper layers and reducing root water uptake in deeper soil layers.We stress that results should be interpreted as the soil depths at which moisture is used for evapotranspiration, and not solely root water uptake patterns.Since soil evaporation also influences our depth estimates, root water use patterns are likely deeper than those estimated here.Soil moisture sensors are still needed at deeper layers than 100 cm to more comprehensively evaluate the role of deep roots on ecosystem function (Stocker et al., 2023;Tumber-Dávila et al., 2022).Nevertheless, root water uptake from much deeper soil layers might be less frequent, such as under drought conditions (Miguez-Macho & Fan, 2021); our results here apply to mainly nominal climatic conditions.
Despite several sources of error that would confound determining LST-soil moisture relationships between high resolution satellite retrievals and in-situ measurements, it is promising that 39% of sites have significant interactions at a median correlation of 0.41.Furthermore, LST-soil moisture correlations strengthen when considering only measurement pairs between 1 and 2 p.m. to a median of 0.58 (interquartile range of 0.54 to 0.67), though the number of sites is greatly reduced (to 14% of sites) mainly due to low sample size.Nevertheless, with more years of ECOSTRESS sampling, the data can be constrained further to determine these profiles of ecosystem water use more accurately.
The use of quality flags can also increase the correlation magnitude, but only marginally and at the cost of large reductions in sample size.If poor quality datapoints were included, the LST-soil moisture correlation should hypothetically approach zero.However, our results show that the linear fits did not greatly degrade without quality flags, likely due to using other quality control methods; the mean LST-soil moisture correlation is 0.41 without flags, while it is 0.45 with these flags.Meanwhile, the median LST-soil moisture pairs per site decreases from 33 to 21.In fact, the analysis is more sensitive to the time of day of the observations than to whether quality flags are used.For example, if the analysis diurnal period is extended from 12-3 p.m. to 11 a.m.-4 p.m., the median correlation decreases to 0.32 because LST is strongly sensitive to the diurnal cycle of radiation, which-if we do not focus on a short afternoon time window-can confound LST sensitivity to SM.Nevertheless, with more years of ECOSTRESS samples, quality flags can likely be used in the analysis without a critical loss of samples.
The frequent occurrence of statistically significant (p < 0.05) LST-soil moisture relationships found here also suggests that high resolution satellites can be used jointly with in-situ measurements.Specifically, if in-situ SM did not at least partially represent the true integrated SM within the 70 m ECOSTRESS footprint, then no LST-soil moisture relationships would be found.As such, our findings show that the effectively point-scale sensor measurement is still partly representative of a 70 m grid scale, suggesting that the confounding effects of topography, rainfall, and vegetation heterogeneity are not prohibitively large within this pixel scale (Charpentier & Groffman, 1992;Western & Blöschl, 1999).Representativeness and scaling errors, at least for SM, tend to be static and thus less likely to affect evaluations of the slope or correlation-based metrics used here (Chen et al., 2019).
Beyond an evaluation of soil water use, such a demonstration here encourages addressing other science questions with the concurrent use of high-resolution remote sensing and ground measurement networks, where advantageous.Available high-resolution satellites (<100 m), like NASA Earth surface Mineral dust source InvesTigation (EMIT), NASA Global Ecosystem Dynamics Investigation (GEDI), and NASA Cyclone Global Navigation Satellite System (CYGNSS) and upcoming SBG and NASA-ISRO Synthetic Aperture Radar (NISAR) (Cawse-Nicholson, Townsend, et al., 2021) can be viably used in conjunction with field measurements to address an array of terrestrial water, carbon, and energy cycle questions.A.F.F. was supported by both the NASA ECOSTRESS science team and by a NASA Terrestrial Ecology scoping study for a dryland field campaign (ARID).K.C. N. was supported by the ECOSTRESS mission and the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration.Government sponsorship is acknowledged.The United States Department of Agriculture-Agricultural Research Service is an equal opportunity employer.The authors acknowledge their use of SM measurement network data, especially USCRN and SCAN, and compiled by the ISMN.This work used eddy covariance data acquired and shared by the FLUXNET community, including AmeriFlux.The FLUXNET eddy covariance data processing and harmonization was carried out by the ICOS Ecosystem Thematic Center, AmeriFlux Management Project and Fluxdata project of FLUXNET, with the support of CDIAC, and the OzFlux, ChinaFlux and AsiaFlux offices.The authors thank Prajwal Khanal and an anonymous reviewer for their constructive comments on the manuscript.

•
High resolution satellite retrievals of land surface temperature can reveal ecosystem water use when coupled with soil moisture (SM) networks • Across vegetation types, evaporation tends to mainly use water from the upper soil layers with a decrease of SM use with depth • Grassland sites tend to have more frequent ecosystem water use of upper layer SM than forested sites Supporting Information: Supporting Information may be found in the online version of this article.

Figure 1 .
Figure 1.Example analysis of ecosystem water use depths in an Arizona grassland.(a)-(b): FLUXNET demonstration of relationship between afternoon (12-3 p.m.) soil temperature at 5 cm and afternoon (12-3 p.m.) (a) latent heat flux and (b) sensible heat flux in a dry grassland mixed with some tree cover in Santa Rita Experimental Range in Arizona (US-SRM).(c)-(h) ECOsystem Spaceborne Thermal Radiometer Experiment on Space Station land surface temperature relationship with Soil Climate Analysis Network soil moisture at a dry grassland site in Walnut Gulch Experimental Watershed in Arizona ("WalnutGulch#1") at (c) 5 cm, (d) 10 cm, (e) 20 cm, (f) 50 cm, and (g) 100 cm depths.P values are displayed in each subplot.(h) β θ,d (LST-soil moisture slope) at each depth.Red markers indicate statistically significant β θ,d values ( p < 0.05).

Figure 2 .
Figure 2. Soil profiles of ecosystem water use by site and across different vegetation types.(a) Map of in-situ soil moisture US Climate Reference Network and Soil Climate Analysis Network sites with at least one soil depth where LST-soil moisture relationships are significant ( p < 0.05).Land use and land cover types are denoted by different symbols.(b)-(e) The depths of ecosystem water use for sites classified as (b) forests (c) grasslands (d) croplands, and (e) cropland and natural mosaic.Open shrublands and woody savanna are shown in Figure S11 of Supporting Information S1 given their lower sample size.Each column of dots (often with a vertically oriented line connecting them) represents a single site (that also appears in panel (a)).Dot symbols denote that a significant ( p < 0.05), negative LST-soil moisture relationship exists at that depth.The colors of the dot symbols denote the steepness of β θ at that depth.Black outlines around the dot symbol denote the determined depth of peak water use based on maximum magnitude β θ .For visual clarity, sites in each panel are ordered by this determined depth.

Figure 3 .
Figure 3. Profile of ecosystem water use across all sites.(a)Percentage of sites with a significant ( p < 0.05), negative LSTsoil moisture relationship at that soil depth.These values evaluate each soil depth individually and thus range from 0% to 100% at each soil depth (they do not add to 100% across all depths because interactions are possible at any depth).While all sites in Figure2aare included in the "All" category, vegetation types are only shown for those that include more than 10 sites.(b) Soil depths of peak ecosystem water use (maximum magnitude, significant LST-soil moisture steepness, p < 0.05) as a discrete cumulative distribution function (though with the cumulative distribution function on the x axis).For a given soil depth, the values are the percentage of sites that have their peak water use at or below that depth.The median for each category is therefore the value at the 50% occurrence rate.For display purposes, the depths associated with different vegetation types are arbitrarily staggered by ±2 cm in (b).