Though the volume of water is small, surface soil moisture generates outsized effects on the global water and energy balance (McColl et al., 2017b). Climate, weather, and flood predictions depend on soil moisture (Entekhabi et al., 1996; Viterbo and Betts, 1999). Feedback between the land and atmosphere can perpetuate soil moisture anomalies differently depending on the climatic regime (Koster et al., 2004; Tuttle and Salvucci, 2016). The duration of soil moisture anomalies depends on the drying rate of soil, which is controlled by complex interactions between soil hydrologic processes, atmospheric conditions, and vegetation state (e.g., Rodriguez-Iturbe, 2000). Therefore, documenting the controls on soil drying is necessary to more fully understand the role of terrestrial hydrologic processes and land–atmosphere interactions in the climate system. In this study, we use satellite data from NASA's SMAP (Soil Moisture Active Passive) mission and ancillary datasets to investigate such processes, specifically direct soil evaporation.

Soil moisture loss mechanisms include runoff, drainage, and evapotranspiration (ET). The relative dominance of each depends on how much time has passed since rainfall (Laio et al., 2001). Runoff, when present, ceases only minutes after rainfall. Drainage occurs on a timescale of hours. Thus both can be ignored when considering drying intervals on a timescale of days, as is documented by the 1–3-day SMAP soil moisture observations (Chan et al., 2016). Assuming drainage is negligible outside of rainy intervals is a common assumption in models of soil moisture dynamics (Federer et al., 2003; Guswa et al., 2002; e.g., Laio et al., 2001; McColl et al., 2017a; Porporato et al., 2004).

At the continental scales observed by SMAP, these assumptions result in a loss function dominated by stage two (water-limited) ET (McColl et al., 2017a). Water-limited ET is itself the sum of two fluxes: (1) direct evaporation from the soil surface, which typically occurs over a depth of several centimeters and (2) transpiration via plant stomata, which removes water from throughout the root zone (Campbell and Norman, 1998; Monteith and Unsworth, 2013). Given the shallow sensing depth of L-band radiometers (Njoku and Entekhabi, 1996), we expect SMAP to capture mainly the former. Accordingly, our study is framed around assessing the controls on direct evaporation on the continental scale. We include surface soil moisture, potential evaporation (PE) rate, vegetation, and soil texture in our assessment, as these quantities are all expected to control direct soil evaporation rates.

Surface moisture (volumetric soil moisture, VSM) supply is known to affect evaporation rates. In the Noah land surface model (LSM) and other models, moisture stress scales evaporation rates according to a piecewise function. Above field capacity, evaporation proceeds at its potential rate (stage one evaporation); below a residual soil moisture content, no evaporation occurs; between the two thresholds, evaporation depends on relative moisture content (Allen, 2000; Chen and Dudhia, 2001).

Atmospheric moisture demand is quantified by the PE rate and is also known to affect evaporation rates. Surface vapor pressure, atmospheric vapor pressure, radiation, albedo, and wind velocity contribute to its calculation (Mahrt and Ek, 1984; Penman, 1948).

Any vegetation present on the land surface introduces two counteracting effects on soil drying that do not exist for bare soil surfaces. Direct evaporation decreases because of shading from the canopy that intercepts solar radiation (Mahfouf and Noilhan, 1991). Transpiration draws moisture from the root zone into the atmosphere, with the depth of this flux depending on root distribution (Schenk and Jackson, 2002).

The soil texture of the land surface, and thus its hydraulic properties, also influences soil drying. Compared with fine-textured soils (clay), coarser soils (sand) have more well-connected pores, allowing water to leave the system more easily at a given moisture level (Campbell, 1974; Clapp and Hornberger, 1978; Cosby et al., 1984; Van Genuchten, 1980). These differences cause volumetric soil moisture to be lower for coarse textures than for fine (Laio et al., 2001). Despite the efforts that have gone into quantifying and modeling the effects of textural differences on soil moisture and heat fluxes, there remains much variability within each texture class, so assigning model parameter values based on continental-scale soil maps has proven to be problematic (Gutmann and Small, 2005; Xia et al., 2015b). Nevertheless, the Noah LSM and other models use soil texture to assign soil hydraulic properties, which dictate the infiltration and redistribution of moisture in the soil column (Chen and Dudhia, 2001).

Studying evaporation and its contribution to total ET is necessary to more completely understand and model the flux of water from the land surface to the atmosphere (Kool et al., 2014). However, continental-scale evaporation from soil is notoriously difficult to measure directly. Lysimeters and chamber measurements provide information over extremely small areas (∼ 10 m2 or less) (e.g., Herbst et al., 1996; Stannard and Weltz, 2006). Soil drying rates determined from satellite-based observations can provide an estimate of surface evaporation rates on a large scale (McColl et al., 2017a). This requires that the depth supplying evaporation is sampled by the sensor and that vertical redistribution within the soil is negligible. Evaporation largely draws from the top several centimeters of the soil column, within the sensing depth of L-band radiometers (Njoku and Entekhabi, 1996), although under extreme dry conditions the evaporative front will propagate down to greater depths. In addition, L-band sensing depth varies slightly with moisture content (Njoku and Kong, 1977), though quantifying the effect of this change is beyond the scope of this study.

In situ observations have allowed for investigation into how different environmental factors control soil drying rates and thus evaporation rates (Cavanaugh et al., 2011; Detto et al., 2006; e.g., Kurc and Small, 2004, 2007). Only recently has satellite remote sensing of soil moisture advanced sufficiently to make it possible to monitor drying rates on a large scale (Entekhabi et al., 2010; Kerr, 2006), thus allowing scientists to evaluate the physical controls on soil drying across a wider range of conditions. McColl et al. (2017a) studied global soil drying dynamics by fitting SMAP surface soil moisture observations to an exponential decay function. They found shorter drying timescales in areas that have higher aridity indexes and higher soil sand content, although the effects of soil texture were relatively minor. Their results confirm the expected roles of atmospheric demand and soil texture on soil drying, and thus presumably also on direct evaporation. They noted, however, that there was substantial unexplained variance in drying timescales, citing vegetation as a likely factor. In addition, McColl et al. (2017a) only considered static descriptors of the physical environment at each location: neither aridity index nor soil texture vary through time. Yet, the soil drying process represents a dynamic interplay between hydrologic, climatic, and ecosystem processes (Rodriguez-Iturbe, 2000).

Here, we do not fit an exponential decay model to drydown events as in McColl et al. (2017a). Instead we use calculations of changes in surface soil moisture through time, which Shellito et al. (2016b) showed to provide a similar depiction of soil drying as the exponential approach. Importantly, by analyzing simple soil drying rates we avoid introducing new uncertainties into our analysis associated with fitting an exponential decay model.

Our study builds upon McColl et al. (2017a) in two ways. First, we examine the effects of dynamic controls on soil drying, rather than static covariates. Soil moisture supply, PE rate, and vegetation cover are observed or calculated coincident with both the time and location of the SMAP drying observations. This approach allows us to obtain results indicative of time-varying hydrological mechanisms and processes, rather than of the overall climate (e.g., aridity index) or soil type in each location.

Second, we compare our results to output from the Noah LSM (Ek et al., 2003; Xia et al., 2012a). Noah development and validation have been more focused on reproducing heat fluxes and runoff than soil moisture (Chen and Dudhia, 2001; Xia et al., 2012a, b). A comparison between Noah and the North American Soil Moisture Database shows it is able to capture broad features of soil moisture variations (Xia et al., 2015a). Accurately simulating specific soil drydowns is more difficult. By analyzing Noah soil drying rates alongside the observations from SMAP, we can offer guidance into the strengths and limitations of Noah-simulated soil moisture dynamics. In addition, the model helps guide and confirm our understanding of the relationship between surface drying rates and surface evaporation rates.

The drying rates observed by SMAP and simulated by Noah decrease with time since rainfall cessation, consistent with prior results using similar data (McColl et al., 2017a; Shellito et al., 2016b). Thus, as the soil dries following precipitation (Fig. 6a), the median drying rates (Fig. 6b) and median equivalent evaporation rates (Fig. 6c) also decrease. The form of the relationship between drying rate and VSM is described in more detail below.

Although SMAP and Noah both show that drying slows with increased time since rainfall, considerable differences exist. Directly following precipitation, drying rates are higher based on SMAP data than simulated by Noah, by a factor of 2 (Fig. 6b). However, SMAP drying rates decrease more quickly through time. After about 10 days, Noah drying rates are equal to or slightly faster than SMAP rates. The SMAP and Noah equivalent evaporation rates are similar (∼ 0.8 mm day-1) directly following rainfall (Fig. 6c). Evaporation from Noah occurs over twice the approximate SMAP sensing depth, making up for the differences in drying rate (Fig. 6b) and contributing to higher equivalent evaporation rates than from SMAP after ∼ 5 days.

In addition to the more rapid decrease in drying rates observed with SMAP shown in Fig. 6, there are also obvious geographic differences between SMAP and Noah drying rates (Fig. 1c and d). Median values of drying rates are calculated in a 25 pixel (5 by 5) moving window. Data are displayed at the center of each window whenever at least 100 drying rates are contributing to the median. Median drying rates from SMAP and Noah are similar in the southwest region of the continent where precipitation events are small and infrequent. In these areas, soil moisture remains near its residual value for much of the time. As a result, the large differences that exist immediately after rain events (Fig. 6b) occur infrequently and do not influence the median values shown in Fig. 1c and d. In wetter regions, median drying rates from SMAP are considerably faster than simulated by Noah, at least partly because data from several days after precipitation events affect the median value.

To further investigate the causes of the different drying rates shown above, we consider the variables we understand to control the soil drying process: moisture supply (surface VSM), atmospheric demand (PE), vegetation cover, and soil texture.

The remotely sensed data from NASA's SMAP mission, combined with modeled data, provide insight into the environmental factors that affect surface soil moisture dynamics and direct evaporation from soil. The results presented here are based on correlations between environmental factors and soil drying rates, so it is not possible to prove cause–effect relationships exist. However, the results are consistent with ground-based observations (e.g., Kurc and Small, 2004) and physics-based relationships included in models (e.g., Laio et al., 2001). Therefore, results confirm that these fundamental relationships exist on the continental scale.

SMAP data show that the land surface dries rapidly immediately after rainfall. With time, the soil dries and the drying rate slows, the latter approaching zero after ∼ 10 days. Noah simulations also exhibit this trend, but the drying rates are slower directly after rainfall and persist at nonzero values for longer than SMAP rates do, indicating a more linear drying process (Fig. 6).

To constrain the factors affecting the drying process, we simultaneously consider the supply of water (soil moisture) and the atmospheric demand for it (PE). When the surface soil is wet, only atmospheric demand should limit evaporation rates (stage one evaporation). When the surface soil is dry, moisture supply further restricts evaporation rates (stage two evaporation). Our study finds that, on SMAP spatial and temporal scales, the satellite is observing a system that is predominantly water-limited (McColl et al., 2017a). This is supported by the following observations (Fig. 7).

In most cases, drying and evaporation rates are linearly related to soil moisture content. Such consistent dependence on moisture supply indicates water-limited conditions. In contrast, energy-limited conditions would show evaporation rates to be insensitive to soil moisture.

It is possible that SMAP drying rates are slightly exaggerated due to decreases in L-band sensing depth that accompany wet soil. After rainfall, moisture in the top couple of centimeters could dominate the signal, leading to the entire 0–5 cm sensing depth being assigned a moisture level that is only present at the very surface. As the soil dries and becomes more evenly distributed within the sensing depth, such abnormalities would dissipate.

The calculated values for evaporative efficiency are quite low (Fig. 5); 90 % of values are below ∼ 0.25. This suggests evaporation from the surface satisfies only a small fraction of the atmospheric demand. The fraction is not larger for one or more of the following reasons: (1) on this spatial and temporal scale, evaporation is highly water-limited; (2) most plant roots are deeper than ∼ 5 cm, so transpiration draws water from depths below the surface layer and thus does not contribute to the SMAP-based accounting of drying analyzed here; (3) similarly, if any evaporation draws moisture from below SMAP's sensing depth, it will not be accounted for here; and (4) the sampling interval of SMAP is too low to capture the fastest evaporation rates, which occur soon after rainfall. Only 6.5 % of the calculated drying rates include an observation from within the first 12 h after precipitation, when the fastest drying and highest evaporative efficiency is likely to occur. For example, in Fig. 3, the first observation, t1, is almost 2 days after rainfall cessation.

While VSM and PE clearly influence soil drying, SMAP observations also show that vegetation plays an important role in determining soil drying rates. These effects are most obvious once drying rates are normalized by PE because vegetation (as indicated by NDVI) and PE both tend to be greatest in the summer months (Fig. 8). Evaporative efficiency decreases with increasing vegetation cover (Fig. 9c), unless the soil is dry and evaporative efficiency is close to zero. This means that at a given moisture level and PE rate, the surface soil of a parcel of land will dry more slowly if it has vegetation on it than if it does not. This is consistent with ground-based observations that show direct evaporation can be limited by vegetation (e.g., Breshears et al., 1998).

In comparison to SMAP, the effects of vegetation on drying in Noah are minimal (Fig. 9e–f). These results can be understood given the model's formulation of evapotranspiration (Chen and Dudhia, 2001). Direct soil evaporation only occurs over the fraction of land surface not shaded by the canopy (1-FG), so evaporation decreases with more vegetation. Transpiration only occurs over the fraction of land surface that has a canopy (FG), so transpiration increases with more vegetation. Our results show these two effects are balanced in dry and intermediate wetness soils. Soil in the wettest tercile exhibit slightly higher evaporative efficiencies from bare soil than from vegetated soil (Fig. 9f). The Noah results indicate that in the top 10 cm of soil, moisture from rainfall enhances evaporation from bare ground more than it does transpiration from vegetated ground.

Vegetation effects are much greater as observed by SMAP than simulated by Noah. Therefore, the direct evaporation flux in Noah should be greater (at a given VSM and PE), which would result in higher surface drying efficiencies when vegetation is sparse. Betts et al. (1997) and Ek et al. (2003) both modified Noah's bare soil evaporation function to magnify the decrease in evaporation as the surface dries. The SMAP observations suggest further adjustments are needed.

The results in Fig. 10 show that the sensitivity to soil texture is too high in Noah. SMAP shows only small differences in drying dynamics related to texture classes (here and McColl et al., 2017a), whereas Noah simulations indicate variations of a factor of 5 between sand and other texture classes. Noah-simulated results conform to the expectation that coarser soils (sand and sandy loam) dry faster than fine-grained soils (silt loam). (Soil infiltration and redistribution parameters are indeed selected according to texture class; Chen and Dudhia, 2001). On the contrary, our SMAP-based results show the role of soil textures to be less important than the other factors analyzed here. It is possible that improved soil texture maps could bring SMAP and Noah results into closer agreement, but the heterogeneity of soils even within a single texture class (Gutmann and Small, 2005) makes it unlikely that the solution will come from improving texture maps alone.

The differences in behavior between SMAP and Noah can also partially be attributed to differences in sensing and simulation depths. We expect thicker surface layer dynamics to be dampened when compared to a thinner layer (e.g., Rondinelli et al., 2015). However, changing Noah's layer one soil depth from 0–10 to 0–5 cm only increases soil drying rates by 6 % (Fig. 12), implying that the model structure itself prevents Noah from accurately reproducing the surface soil moisture dynamics observed by SMAP.

SMAP-observed and Noah-simulated soil moisture and drying rates decrease with time since precipitation. SMAP drying rates are faster than Noah-simulated drying rates in the first 8 days after rainfall, but slower afterwards. Because Noah's top soil layer is twice the depth that SMAP senses, its equivalent evaporation rates are nearly the same as SMAP's soon after precipitation and higher afterwards.

SMAP-observed and Noah-simulated soil drying rates both vary linearly with soil moisture content, evidence that continental-scale soil moisture dynamics operate in a water-limited system.

Equivalent evaporation rates from SMAP and Noah rarely exceed 1 mm day-1. Expressed as evaporative efficiency, 90 % of the calculated rates fall below 0.27 (SMAP) or 0.21 (Noah). These values are far below unity, providing further evidence of a water-limited environment. However, accounting for transpiration of water from below the 5 or 10 cm surface layer would shift total ET efficiencies closer to 1.

SMAP and Noah both show that high atmospheric demand for moisture (high PE) increases the sensitivity of drying rates to soil moisture content.

More vegetation amount, indicated by higher NDVI, decreases the surface drying efficiency: SMAP shows a 3-fold evaporation efficiency decrease between sparsely vegetated and densely vegetated pixels. This suggests that the decreases in evaporation from canopy shading are not offset by increases in transpiration from the shallow soil layer. Noah shows a much smaller decrease in evaporative efficiency, only for wet soils, suggesting a deficiency in the model structure.

Soil texture class has a small influence on SMAP drying dynamics. Noah drying dynamics are strongly affected by soil texture class, as prescribed by its soil hydraulic property parameterization.

This study has demonstrated that SMAP exhibits unprecedented soil drying sensitivity across at least three of the dimensions analyzed (soil moisture supply, PE rate, and vegetation cover) that is largely missed by Noah (e.g., Figs. 6, 7, and 9). This provides encouraging evidence of the potential to use SMAP observations to improve model portrayals of soil drying.