Landscape‐Scale Plant Water Content and Carbon Flux Behavior Following Moisture Pulses: From Dryland to Mesic Environments

Rain pulses followed by interstorm drying periods are the fundamental units of water input into ecosystems on subweekly time scales. It is essential to understand landscape‐scale vegetation responses on these unit time scales as they may describe sensitivity of landscape water, carbon, and energy cycles to shifts in rainfall intensity and frequency, even if the average seasonal precipitation remains unchanged. Because pulse investigations are primarily carried out in drylands, little is known about the characteristics and extent of ecosystem plant pulse responses across the broader range of climates and biomes. Using satellite‐based plant water content (from vegetation optical depth) and plant carbon uptake observations from eddy covariance towers across the continental United States climate gradient (dry to humid), we characterize large‐scale plant carbon and water uptake responses to rain pulses during spring and summer months. We find that while all ecosystems in the study region show discernable plant water content and carbon flux responses to rain pulses, drier ecosystems exhibit more frequent and longer duration responses. Unlike mesic environments, drylands show significantly different responses under varying antecedent soil moisture and pulse magnitude conditions; the largest water and carbon uptakes follow large pulses on initially wet soils. We detect soil moisture thresholds primarily in drylands, which can partly explain dryland vegetation's different responses under dry and wet conditions. We conclude that vegetation responds to individual pulses of water availability across all climates and therefore a range of ecosystems are sensitive to rainfall distributions beyond simple seasonal precipitation totals.


Introduction
Rain pulses provide major inputs of plant available soil water to ecosystems on subweekly scales (Yang et al., 2008). A soil moisture drydown period follows these pulses where soil water is lost through transpiration, soil evaporation, and drainage. After accounting for seasonal modes of water input (i.e., snowmelt), these fundamental pulse time units can additively scale up to describe the annual water cycle at a location (Eagleson, 1978). Given that vegetation exerts a strong control on the global water, carbon, and energy cycles (Jasechko et al., 2013), it is essential to understand terrestrial biosphere behavior on this time scale to, for example, address how rainfall regimes and their changes impact these cycles (Knapp et al., 2002). However, seasonal and annual-scale environmental controls on plant productivity have received more attention (Nemani et al., 2003). We therefore have inadequate knowledge of general plant pulse dynamics and their influence on seasonal and annual vegetation behavior across biomes.
Primarily, only dryland ecosystems have been evaluated in the context of pulsed water inputs (Collins et al., 2014;. Dryland vegetation has been hypothesized to be driven by infrequent, unpredictable rain pulses under a pulse-reserve paradigm (Noy-Meir, 1973). This paradigm states that, after rainfall, plants upregulate (increasing sensitivity to external stimuli) with growth and carbohydrate storage, followed by depletion of reserves . Continental-scale evidence exists for this paradigm (Feldman et al., 2018). While field studies have not directly assessed such growth and storage mechanisms (Collins et al., 2014), they have instead evaluated leaf gas exchange and plant hydraulic (i.e., via predawn water potential) responses to rain pulses (Huxman, Snyder, et al., 2004;Sala & Lauenroth, 1982;Williams et al., 2009). Overall, these studies generally agree that small rain events induce ecosystem respiration efflux responses (e.g., the "Birch" effect, Jarvis et al., 2007) and larger rain events additionally result in great-Abstract Rain pulses followed by interstorm drying periods are the fundamental units of water input into ecosystems on subweekly time scales. It is essential to understand landscape-scale vegetation responses on these unit time scales as they may describe sensitivity of landscape water, carbon, and energy cycles to shifts in rainfall intensity and frequency, even if the average seasonal precipitation remains unchanged. Because pulse investigations are primarily carried out in drylands, little is known about the characteristics and extent of ecosystem plant pulse responses across the broader range of climates and biomes. Using satellite-based plant water content (from vegetation optical depth) and plant carbon uptake observations from eddy covariance towers across the continental United States climate gradient (dry to humid), we characterize large-scale plant carbon and water uptake responses to rain pulses during spring and summer months. We find that while all ecosystems in the study region show discernable plant water content and carbon flux responses to rain pulses, drier ecosystems exhibit more frequent and longer duration responses. Unlike mesic environments, drylands show significantly different responses under varying antecedent soil moisture and pulse magnitude conditions; the largest water and carbon uptakes follow large pulses on initially wet soils. We detect soil moisture thresholds primarily in drylands, which can partly explain dryland vegetation's different responses under dry and wet conditions. We conclude that vegetation responds to individual pulses of water availability across all climates and therefore a range of ecosystems are sensitive to rainfall distributions beyond simple seasonal precipitation totals.
FELDMAN ET AL. er water uptake, transpiration, and primary production responses (Chen et al., 2009;Q. Guo et al., 2016;Tang et al., 2018;Xu et al., 2004). Some studies note that antecedent soil moisture conditions explain differences in response magnitude and duration (J. S. Guo and Ogle, 2019;Potts et al., 2006;Tang et al., 2018). However, there have been few efforts to reconcile whole-plant pulse responses beyond individual sites Reynolds et al., 2004). As such, calls to scale leaf to ecosystem plant pulse behavior and evaluate pulse behavior beyond drylands (i.e., woody vegetation in mesic regions) remain relatively unaddressed (Weltzin & Tissue, 2003;. Scaling up plant behavior beyond drylands is becoming possible with recent satellite missions and streamlined field tower networks. Since the 1980s, remote sensing missions have allowed large-scale characterization of landscape-scale (km scales) vegetation behavior. However, cloud cover sensitivity of widely used optical and thermal satellite-based sensors hinders vegetation sampling periods of less than a week. Therefore, understanding daily landscape-scale plant behavior, especially in the context of pulses, has been limited. Microwave remote sensing missions, insensitive to cloud cover, overcome these limitations with 1-3-day simultaneous retrievals of surface soil and plant water content (Entekhabi et al., 2010;Kerr et al., 2010;Smith et al., 2019). While at a low spatial resolution (∼30 km), these observations provide an opportunity to assess subweekly landscape-scale plant pulse-response dynamics. These data sets show promise in evaluating the 1-3 days evolution of vegetation behavior as a recent study found that plant water content increases following soil moisture pulses, signifying plant water uptake (Feldman et al., 2018). Furthermore, with the streamlining and greater availability of eddy covariance towers, multiyear records of hourly ecosystem-scale (100-m scales) carbon flux measurements are also available across many land cover types (Pastorello, 2020). These measurements also have been used to reveal much about carbon pulse dynamics primarily in drylands (Huxman, Cable, et al., 2004;Kurc & Small, 2007;Potts et al., 2006;Roby et al., 2020). Therefore, these plant water content and carbon flux measurements show promise in providing complementary information to assess plant pulse behavior across climates and biomes, without process-model assumptions.
Here, using satellite plant water content and tower-based carbon flux measurements in the continental United States, we aimed to address two questions: (I) what are the plant water content and carbon flux responses to individual rain pulses across climate gradients (dry to humid)? (II) How sensitive is the plant water content and carbon flux behavior to pulse characteristics, specifically antecedent moisture and pulse magnitude? We evaluated the subweekly responses following individual rain pulses using these spatially integrated vegetation observations. The common expectation is that dryland plants show subweekly water uptake and carbon assimilation responses to rain pulses, due to reliance on near-surface soil moisture . Mesic environment vegetation may instead rely more on seasonal rootzone moisture variations with greater sensitivity to temperature and incoming radiation (Madani et al., 2017;Nemani et al., 2003). Our analysis was conducted during the continental US spring and summer when plants are generally phenologically active with less prevalent low temperature limitations (Kurc & Small, 2007;Loik, 2007;Pockman & Small, 2010;Potts et al., 2006).

Data Sets
Within our study domain of the continental United States, we used all available FLUXNET eddy covariance tower (Tier 1) measurements of net ecosystem exchange and surface soil moisture (0-5 cm depth) (Pastorello, 2020). Net ecosystem exchange was converted to net ecosystem production (NEP) where positive NEP represents land surface carbon uptake. We also used the partitioned gross primary production (GPP) and ecosystem respiration (R e ) products from FLUXNET to assess the role of R e (references in Table 1). Daily NEP values are reported here as averaged daytime measurements between 10 a.m. and 4 p.m. Soil moisture values were averaged over the full day. Sites that are without NEP and soil moisture measurements and those that are irrigated, disturbed by fire, or in wetlands were not included. In total, 36 sites were used in our analysis (Table 1).
Within the same domain, we also used Soil Moisture Active Passive (SMAP) satellite-observed surface soil moisture (∼5 cm) and vegetation optical depth (VOD) from April 2015 to March 2019 posted on a 9-km grid (Entekhabi et al., 2010;Konings et al., 2016). The SMAP satellite makes observations near 6 a.m. local time, reflecting predawn soil and plant water content conditions. Details of the satellite parameter retrieval technique (called the multitemporal dual channel algorithm) and uncertainty can be found in Konings et al. (2016) and Feldman et al. (2018). SMAP satellite soil moisture closely resembles in situ soil moisture dynamics at core validation sites primarily across the study domain (Chan et Figure S1.  , 2018). While robustly estimated, the VOD product has yet to undergo thorough ground validation (Konings et al., 2016). VOD represents a unitless measure of attenuation of microwave emission through a canopy and has been found to be linearly proportional to plant water content (Jackson & Schmugge, 1991;Momen et al., 2017). Direct and equivalent field measurements of integrated plant water content are not available. However, the time-averaged VOD correlates strongly with vegetation density and height (Brandt et al., 2018). Furthermore, VOD temporal dynamics compare to predawn water potential (leaf, xylem, etc.), an independent satellite VOD product, and are complementary to leaf area index (Feldman et al., 2018;Momen et al., 2017;Tian et al., 2018;Zhang et al., 2019). For the primary analysis, we evaluated satellite-based VOD and tower-based estimates of NEP between April 1st and September 30th in available years, representing Northern Hemisphere spring and summer seasons.
We computed mean annual precipitation (MAP) from the satellite Global Precipitation Measurement (GPM) IMERG product (Huffman, 2015) to assess satellite-based vegetation behavior across the dry to humid climate gradient in the United States. Similarly, FLUXNET precipitation was also used to quantify MAP corresponding to flux tower vegetation data across this climate gradient. For four FLUXNET sites that do not have a ground-based MAP estimate available (see Table 1), we used the MAP estimate from GPM IMERG over the SMAP time period. Since MAP was primarily used as a binary indicator of a location existing in dryland or mesic environment, we do not expect use of GPM MAP estimates at these four sites to bias results. We also used International Geosphere-Biosphere Program (IGBP) land cover classification information to broadly assess vegetation types (Kim, 2013). Finally, we used fraction of vegetation cover as converted from MODIS leaf area index to assess vegetation presence in drier climates in our study region (Myneni et al., 2015).

Data Set Representation
While different in their representation, NEP and VOD are complementary whole-canopy, large spatial scale metrics that may be sensitive to rain pulses. Both NEP and VOD are integrated metrics of different components of plant behavior. NEP equals GPP minus R e , where GPP represents photosynthesis via the rate of ecosystem CO 2 uptake and R e represents ecosystem CO 2 effluxes due to both heterotrophic (i.e., soil microbes) and autotrophic processes (Law et al., 2002). VOD, linearly proportional to plant water content, is influenced by both structural growth/decay (creating more/less water storage space) and relative water content (water uptake/loss independent of growth) (Konings et al., 2019;Momen et al., 2017;Tian et al., 2018;Zhang et al., 2019). These metrics are complementary in that NEP and VOD may observe different components of daily vegetation use of rain pulses, such as plant hydraulic (i.e., rehydration) and photosynthetic responses (i.e., carbon assimilation). Note that our primary analysis refrains from partitioning NEP and VOD into more fundamental components to avoid model assumptions that may prescribe (either explicitly or implicitly) pulse behavior we attempt to estimate. Finally, both metrics are integrated over large spatial scales allowing evaluation of behavior beyond individual species. NEP integrates ecosystem carbon flux behavior at 100-m scales, while SMAP VOD observations integrate plant water content at landscape scales of tens of kilometers.
The inherent spatial sampling and time domain differences between NEP and VOD pose a limitation in their joint analysis. Specifically, VOD will observe a larger area of vegetation than that of NEP at a given site. Additionally, their time domains do not overlap, hindering direct comparison of specific events between data sets. It is, therefore, assumed that the pulse responses do not change over the time domain covered here . The validity of this assumption is uncertain given limited characterization of pulse responses. Pulse-response magnitudes and timing may shift with processes such as land cover change, climatic change, and interannual weather and climate variability. For example, free-air carbon dioxide enrichment experiments show that increased ambient carbon dioxide concentrations can alter stomatal responses and thus the magnitude of the NEP response over decade-long durations (Ainsworth & Long, 2005;Bernacchi et al., 2007). These leaf gas exchange magnitudes can be further influenced by increases in vapor pressure deficit (VPD) with temperature (Novick et al., 2016). Further assessment of this stationary pulse-response assumption is beyond the scope of this study. Ultimately, rather than directly compare the two metrics, we intend to assess gradients in NEP and VOD behavior between dryland and mesic environments and their responses to varying pulse characteristics.

Pulse Period Identification
We identified rain pulses using the surface soil moisture time series from both field and satellite observations. Given that rain pulses produce concurrent soil moisture pulses, both terms are used synonymously throughout this study. A conceptual soil moisture pulse is shown in Figure 1. Use of soil moisture rather than precipitation to identify pulses is recommended because soil moisture is more ecologically relevant to plants and it integrates temporal storm details . Furthermore, use of the shallow soil moisture layer to identify pulses and their characteristics is an advantage for this application because this layer will respond to nearly any rain size pulse, unlike deeper soil layers (Kurc & Small, 2007;Schwinning et al., 2003;Scott et al., 2006). Specifically, a pulse period, or soil moisture drydown, is defined here as a detectable signal of volumetric soil moisture increase (>0.01 m 3 /m 3 ) followed by at least four consecutive measurements of soil drying (McColl et al., 2017). Four consecutive measurements were chosen to filter out short drydowns falsely detected due to noise. Due to sampling frequency differences between data sets, this definition results in drydowns of at least 4 days for flux tower observations and at least 6-8 days for satellite observations. The difference in drydown identification length is not expected to impact results because FELDMAN ET AL.
10.1029/2020WR027592 5 of 20 Figure 1. Conceptual soil moisture pulse and expected photosynthetic and hydraulic responses for water-limited and non-water-limited ecosystems. Expected responses are shown for variables used in this study: flux tower observed net ecosystem production (NEP) and satellite-observed vegetation optical depth (VOD, proportional to plant water content). Non-water-limited ecosystem responses represent no detectable NEP or VOD response. Gray shading indicates the rain pulse period when soil moisture is increasing (before day 0). The nonshaded region is the interpulse period when the soil moisture drydown occurs (after day 0). dNEP/dt and dVOD/dt values are reported with the convention that the change rate between 0 and 1 day after pulse is reported between the 2 days or 0.5 days after the pulse.
responses are typically detected here in the first 3 days of the drydown. SMAP 9 km data were pooled within half degree pixels to increase drydown samples. Note that small subsequent rain events may have occurred during the identified drydowns but were ultimately too small in magnitude to discontinue soil drying and were thus likely negligible compared to the initial pulse.

Pulse-Response Assessment
Figure 1 conceptually displays our expectations of NEP and VOD pulse responses for the driest and most humid ecosystems. Specifically, ecosystems with sufficiently available water may nominally exhibit no tendency to change NEP or VOD following a rain pulse with variability instead due to temperature, light, and nutrients ("non-water-limited" case in Figure 1). Conversely, a rain pulse may cause distinct hydraulic and photosynthetic responses as seen by NEP and VOD in dryland ecosystems ("water-limited" case in Figure 1). Increases in NEP can indicate primary production responses where carbon uptake accelerates above that of prestorm conditions. This may follow an immediate NEP decrease due to upregulation of respiration. This respiration response is known as the "Birch effect" where rewetting of previously dry conditions activates soil microbes, releasing soil carbon through decomposition processes (Jarvis et al., 2007). VOD can detect water uptake and pulse-driven growth responses, with the time scale of water uptake depending on the plant's capacitance and resistances (Phillips et al., 2004). In the absence of dry biomass changes on these time scales, 6 a.m. observed VOD may represent predawn water potential (i.e., xylem and leaf). Since soil and plant (xylem, leaf, etc.) water potentials are commonly thought to approach equilibrium overnight (Donovan et al., 2001), nominal rapid rehydration responses following rainfall may result in VOD closely covarying with soil moisture. Beyond nominal rehydration responses, we suspect that the driest ecosystems may frequently show multiday VOD responses to rain pulses due to pulse-driven growth and/or slow rehydration due to hydraulic recovery upon rewetting (such as recovering soil-root interface) (Blackman et al., 2009;Carminati et al., 2017;Noy-Meir, 1973). Therefore, both NEP and VOD upregulation responses in drylands are expected to occur over multiple days after rewetting of drier soil conditions. We anticipate that both NEP and VOD will eventually decrease with drying soil with decelerating carbon uptake and plant water loss. Our procedure here evaluates these expectations.
NEP and VOD immediately preceding and during these drydown periods (before and during the storm) were recorded. With the goal of isolating plant responses due to the rain event alone, the seasonal cycles were subtracted from the NEP and VOD time series to remove confounding effects (i.e., seasonal phenology) according to Feldman et al. (2019). Additionally, the first difference (derivative) was taken of the NEP and VOD time series to remove dependence on absolute NEP and VOD values, highly coupled to climatology. The resulting dNEP/dt and dVOD/dt metrics allow closer evaluation of daily to weekly plant responses to individual rain pulses. A conceptual time series of NEP and VOD with conversion from absolute units to the first difference for both NEP and VOD is shown in Figure 1. dNEP/dt becomes an acceleration which measures the ability for the ecosystem to upregulate, as was used previously to evaluate soil respiration (Potts et al., 2014). dNEP/dt and dVOD/dt values from all drydowns at each (later specified) site were binned based on the day after the start of the drydown. Note the caveat that sites with more identified drydowns, due in part to longer records, will influence the aggregated NEP behavior more (Table 1).
To evaluate the frequency of upregulation responses as observed by both data sets, the fraction of pulses with NEP and VOD increases on the first day following the pulse was computed at each location. This metric assesses how frequently NEP and VOD increase rather than decrease on the day following the rain pulse. It serves as an indicator for frequency of poststorm lagged hydraulic, photosynthetic, and growth responses, suggesting upregulation from previously dry soil conditions. This comparison serves to set the stage for more detailed analyses that follow.

Pulse Characteristic Assessment
To evaluate the effects of pulse characteristics, the pulse magnitude and antecedent soil moisture were recorded for each drydown period. The pulse magnitude is defined as soil moisture at the beginning of a drydown minus soil moisture before the drydown. The antecedent soil moisture is defined as the soil moisture value before the drydown beginning. Within a site, each drydown was binned into small/large pulse mag-nitude and dry/wet antecedent moisture conditions based on whether the pulse characteristics are above or below the median of the respective condition. These classifications result in four combinations of pulse characteristics (small or large pulse magnitude and dry or wet antecedent moisture). Results are insensitive to bin thresholds; nearly identical results are obtained when binning below the 40th percentile and above the 60th percentile of the respective pulse characteristic. Surface soil moisture may not be characteristic of the rootzone conditions if deeper rooting distributions exist, which may impact validity of antecedent moisture definitions here. However, daily-scale surface soil moisture has been shown to be correlated with deeper soil layer moisture variations, especially in mesic environments .
Kruskal-Wallis nonparametric tests were performed to determine whether there are significant differences (p < 0.05) in medians of the four pulse scenario distributions (combinations of small/large pulse magnitude and dry/wet antecedent moisture) on each day after the pulse. This was completed for both NEP and VOD.

Response Uncertainty Assessment
We created a null model to determine whether the dVOD/dt or dNEP/dt values are significantly different from that of a random noise series that is purposefully independent of soil moisture (where mean variable change rates are zero). Specifically, the null model tests whether the NEP and VOD behavior is different from that of the "non-water-limited" scenario in Figure 1. For each location, we generated a random NEP time series with the same mean, variance, and first-lag autoregressive coefficient as the original NEP time series. One hundred NEP time series replicates were created. These synthetic dNEP/dt series were also binned based on the day after the pulse coinciding with the observed soil moisture drydowns. Mann-Whitney U tests were performed between the synthetic and observed dNEP/dt distributions to assess whether the dNEP/dt values on that day after the pulse are significantly different from random noise with median zero (p < 0.05). Significance indicates dNEP/dt values depend on soil moisture on the respective day. This procedure was repeated for VOD.

Moisture Threshold Detection
To determine if soil moisture state-dependence behavior exists, we assessed dNEP/dt and dVOD/dt behavior versus soil moisture at each site. We specifically divided the dNEP/dt values into two bins above and below a given soil moisture threshold, based on the soil moisture values that the dNEP/dt values cooccurred with. We then tested whether the difference in the two dNEP/dt distributions is statistically significant. A tested soil moisture value was labeled a significant threshold if the dNEP/dt median above this moisture value is both (1) significantly greater than zero (Mann-Whitney U test between dNEP/dt observations and dNEP/dt null model distribution; p < 0.05) and (2) significantly greater than the dNEP/dt distribution median below this moisture value (Mann-Whitney U test between dNEP/dt medians at high and low soil moisture; p < 0.05). Together, these criteria determine whether wetter soils show stronger increases in carbon uptake than do drier soils. We tested soil moisture values in 0.01 m 3 /m 3 intervals between the maximum and minimum soil moisture value at each site and computed the fraction of successful detections out of the total number of soil moisture values tested. A higher value of this metric will suggest nonlinear NEP and VOD behavior indicative of moisture threshold behavior at the respective site. This same process was repeated for dVOD/dt to determine whether wetter soils result in greater dVOD/dt. These criteria are robust to nonlinear and seasonal relationships with soil moisture. Based on 5 cm surface soil moisture holding information about moisture from deeper soil layers, we anticipate that the detected threshold represents nonlinear soil moisture behavior from deeper than 5 cm (Short Gianotti, Salvucci, et al., 2019).

Gradient of Pulse Responses
We find that vegetation in drier environments tends to show more frequent poststorm upregulation responses, where NEP and VOD tend to increase on the first day after the rain pulse ( Figure 2). This FELDMAN ET AL.

10.1029/2020WR027592
is shown with negative correlations between increase frequencies (for both NEP and VOD) and MAP (p < 0.05; Figure 2b). For NEP, this negative correlation with the rainfall gradient is more apparent considering only the sites across the Southwest US to Midwestern US grasslands with convective rainfall regimes (ρ = −0.87; p < 0.01). While NEP tends to increase across the United States after rainfall (fractions > 0.5), increases are more prevalent in drylands. VOD tends to increase after rainfall across the Southwest US drylands but tends to decrease across the mesic Midwestern and Eastern United States. These results are for the continental US spring and summer months. The gradient of behavior for NEP disappears in the winter months due to reduced pulse responses in drylands and increased responses in Mediterranean climates along the US west coast ( Figure S2). We focus only on spring and summer months for the remainder of our analysis.
NEP variations likely reflect GPP rather than R e variations here. NEP (or GPP-R e ) could increase due to GPP increases, R e decreases, and/or GPP increasing more than R e is. Results repeated with FLUX-NET-partitioned GPP show nearly identical results (not shown). Furthermore, previous field studies show that R e tends to increase which would instead decrease NEP immediately after a rain event due to the "Birch effect" (Chen et al., 2009;Huxman, Cable, et al., 2004;Jenerette et al., 2008;Roby et al., 2020;Williams et al., 2009;Xu et al., 2004). Therefore, we interpret the frequent NEP increases as GPP upregulation. Fractions greater than 0.5 are thus interpreted as photosynthetic activity consistently accelerating above prestorm levels after rainfall. Partitioning GPP and R e is discussed further in Section 3.2.
FELDMAN ET AL.

10.1029/2020WR027592
8 of 20 VOD tends to increase after the pulse in the Southwest US drylands indicating canopy upregulation through growth and/or slow rehydration. Based on VOD's proportionality to vegetation water content, positive dVOD/dt after rain pulses can represent a growth response and/or multiday predawn water potential rehydration due to hydraulic recovery from dry conditions. Consistent with our findings, growth responses on daily to weekly scales (Angert et al., 2007;Miao & Bazzaz, 1990;Novoplansky & Goldberg, 2001;Post & Knapp, 2019;Sher et al., 2004) as well as multiday plant water storage increases under dry initial conditions have been observed across drylands previously (Blackman et al., 2009;Brodribb & Cochard, 2009;Fravolini et al., 2005;Ignace et al., 2007;West et al., 2007).
Vegetation in the more humid US regions still shows discernable sensitivity to rain pulses despite showing less frequent upregulation. NEP in the mostly mesic Midwestern and Eastern US regions shows a subdued tendency to increase with values over 0.5. VOD shows more frequent decreases after rain pulses in these regions, suggesting frequent water loss, indicated as values below 0.5 (red shading in Figure 2a). We suspect that the decreased pulse-response frequency is primarily because these mesic regions are limited by other resources (such as light, temperature, and nutrients) where pulsed water inputs only marginally increase ecosystem photosynthesis from its current state (Short Gianotti, Rigden, et al., 2019). Additionally, frequent VOD decreases likely indicate maintenance of predawn water potential equilibrium during soil drying where upregulation is less frequent (see Section 3.2). Nevertheless, rain pulses still appear to influence these more humid ecosystems.
It is unclear why only dryland vegetation in the Southwest United States exhibits multiday increases in plant water content after rain pulses. Portions of the Great Plains grasslands, while also receiving low amounts of rainfall (<500 mm/year), tend to instead show rapid decreases in plant water content during soil drying after rainfall (Figure 2b, fractions < 0.5 in drier environments). These regions also tend to show infrequent VOD increases during the pulse indicating that no rehydration is occurring (not shown). IGBP land cover classifications show that these two regions are divided by broadly differing vegetation types with shrublands in the Southwest United States and grasslands across the Great Plains. These differing responses suggest more opportunistic pulse water usage of the summer convective events for Southwest US vegetation as opposed to usage of summer events across the Great Plains, potentially related to the regions' differing phenologies.
Despite low vegetation coverage in the Southwest United States, VOD observations are likely detecting vegetation variability not due to artifacts. Both fraction of vegetation cover and mean VOD, an indicator of aboveground dry biomass (Brandt et al., 2018), show that the Southwest US shrublands have a detectable vegetation presence similarly to the Midwest US grasslands ( Figure S3). These values are statistically significantly different than that of baseline nonvegetated surfaces (with the Sahara desert used as a test) which expectedly show near-zero mean VOD and do not return vegetation fraction values (due to soil contamination in the visible and near-infrared spectrums).

Dryland and Mesic Environment Comparison
The summary metric in Figure 2 provides information only on VOD and NEP changes on the first day after the rain pulse ignoring behavior during the pulse and possible further lagged behavior Williams et al., 2009). This limitation motivates evaluating the full-time evolution of behavior during drydowns in each region. Due to a potential trend in NEP and VOD behavior from dryland to mesic regions, there may also be differences in pulse behavior between these environments. Additionally, far less is known about pulse behavior in mesic than dryland environments. We therefore aggregate mesic and dryland sites separately for comparison. These regions are partitioned based on a 500-mm MAP threshold below which regions are typically classified as arid to semiarid (Noy-Meir, 1973). Note that results are not qualitatively sensitive to this arbitrary MAP threshold selection. We analogously evaluate SMAP VOD for the pixels that contain each tower site.
In drylands, we find that the frequent poststorm VOD and NEP increases observed in Figure 2 continue typically for 2-3 days into soil moisture drydowns with weaker increases after the first day (Figures 3a  and 3c). Dryland NEP exhibits significant, frequent NEP increases over the first 2 days of the drydown after a 1-day delay and decreases after 5 days of soil drying (Figure 3a). VOD shows similar dynamics albeit with increases beginning during the rain pulse (Figure 3c). This time evolution suggests that the NEP and VOD peaks generally occur between 2 and 4 days after the pulse (Figure 1). The NEP behavior shows photosynthetic rates consistently increasing above prestorm levels and eventual decreases given sufficient drying between storms. This behavior is broadly consistent with previously measured NEP behavior at similar dryland sites (Chen et al., 2009;Q. Guo et al., 2016;Y. Hao et al., 2010;Loik, 2007;Scott et al., 2006;Tang et al., 2018). Note that longer durations of NEP and GPP increases have been observed previously (Kurc & Small, 2007;Williams et al., 2009). The VOD behavior shows that slow rehydration and/or pulse-initiated growth begins during the rain pulse and typically continues for multiple days. Partitioning these aforementioned VOD phenomena is beyond the scope of this study, but we conjecture about them in Section 3.3.1.
FELDMAN ET AL.

10.1029/2020WR027592
10 of 20  Table 1 on each day after rain pulses. (a) NEP drylands. (b) NEP mesic environments. (c) VOD drylands. (d) VOD mesic environments. Soil moisture increases during the pulse in the shaded gray region (before day 0) and decreases thereafter (see Figure 1). For consistency, only VOD pixels nearest to each FLUXNET site were used. dNEP/dt and dVOD/dt values are reported with the convention that the change rate between 0 and 1 day after pulse is reported between the 2 days or 0.5 days after the pulse (see Figure 1). N is the total number of drydown samples from all sites in the respective panel. Since pulse samples are disproportionate across FLUXNET sites (Table 1), VOD distributions are weighted using a resampling technique to ensure that a given site contributes proportionally to both NEP and VOD distributions while preserving the original sample size. y axis limits are adjusted based on NEP and VOD standard deviations to proportionally compare dryland and mesic environment change rates (right and left columns). The red asterisk indicates that the change rate has a statistically significant response to the rain pulse on the respective day (Mann-Whitney U test between null model and distribution; p < 0.05). The frequency metric in the top panel is the same as that reported in Figure 2. Refer to Table 1 and Figure S1 for site information. NEP, net ecosystem production; VOD, vegetation optical depth.
Contrary to our expectations, mesic US vegetation is sensitive to pulses with a brief period of primary production increases and plant water content behavior indicative of rapid rehydration and drying under predawn equilibrium. NEP increases in mesic environments, though with less frequency and with no significant changes persisting for more than a day (Figures 2 and 3b). Since drying continues uninterrupted for at least 4 days in the selected drydown periods, it is unlikely that shorter NEP responses are an artifact of more frequent rainfall in mesic environments. VOD frequently increases rapidly during the pulse period, followed by sustained frequent drying in step with soil moisture during the drydown (Figure 3d). These 6 a.m. observed VOD dynamics likely represent predawn water potential equilibration with soil water potential, with plants rehydrating during the rain pulse then drying in step with soil moisture during the drydown. This interpretation is based on the assumption of minimal dry biomass changes during the drydown and widespread near-linear, monotonic relationships between relative water content and plant water potential (Konings et al., 2019). Interception may also contribute to the initial VOD increase. We suspect that the consistent VOD decreases during drydowns are due to water content loss (i.e., transpiration) rather than structural decay (i.e., leaf-off and mortality). Despite consistent plant water content and soil drying, NEP does not decrease (or downregulate), suggesting carbon uptake dynamics do not typically reach a water-limited state at least initially under a week of continued drying (Figures 3b and 3d). This NEP and VOD behavior ultimately further supports that mesic environment vegetation responds to pulses, though with more subdued behavior compared to dryland environments. It is possible that the most humid environments (such as tropical rainforests), which receive more than twice the rainfall of the humid environments in the United States, may show no responses as in Figure 1.
Poststorm NEP increases are likely due primarily to increases in GPP with only small changes in R e . Two lines of evidence using FLUXNET-modeled separation of NEP into GPP and R e and nighttime carbon fluxes suggest that R e fluctuations are comparatively smaller on this time scale in our study. First, GPP and R e -partitioned estimates show that R e changes are comparably smaller (∼15% of GPP) and therefore NEP primarily represents GPP behavior (Figure 4). Second, nighttime (12 a.m.-6 a.m.) NEP behavior also suggests that R e changes are relatively smaller (∼30% of daytime NEP; Figure S4). Overnight carbon fluxes are thought to reflect R e behavior (Jenerette et al., 2008;Kurc & Small, 2007), though see Fisher et al. (2007) for uncertainties due to active nighttime transpiration. The two methods show different timing of R e increases FELDMAN ET AL.

10.1029/2020WR027592
11 of 20 with nighttime NEP suggesting they begin during the storm and FLUXNET-partitioned estimates showing R e increases on the day after the storm (Figures 4 and S4). Nevertheless, previous studies instead show larger R e increases within the day after rainfall that may overwhelm the initial GPP response (Huxman, Cable, et al., 2004;Jenerette et al., 2008;Kurc & Small, 2007;Potts et al., 2006). We suspect that the smaller R e changes here are due to evaluating pulse behavior during spring and summer months when soil microbes have already largely upregulated following dormancy during the winter months. We ultimately interpret the NEP increases to be due to primary production as seen in previous studies (Roby et al., 2020), though with caution considering uncertainties in partitioning GPP and R e and using nighttime carbon fluxes.

Dependence of Pulse Responses on Pulse Characteristics
We find that varying pulse characteristics, specifically antecedent moisture conditions and pulse magnitudes, results in significantly different NEP and VOD responses in drylands but has little effect on responses in mesic environments (Figures 5 and 6). Linear regressions between pulse characteristics and NEP and VOD rates of change agree with the direction of tendencies in Figures 5 and 6 (see Figures S5 and S6). These results are thus robust to the binning methods applied in Figures 5 and 6. Therefore, given that previous studies identified pulse moisture thresholds and effects of antecedent soil moisture, plotting all pulse FELDMAN ET AL.

10.1029/2020WR027592
12 of 20 for reference though noting satellite soil moisture was used to bin the VOD responses. The "overall median" is reproduced from Figure 3 as a gray line and is the same across all bins. The "bin median" is the median for its respective bin shown in the bold line. Colored shading indicates that the bin median is above (green) or below (red) the overall median from Figure 3. N indicates the number of pulses in each bin. A black asterisk indicates that the Kruskal-Wallis test shows significantly different distributions in rates of change between the four pulse cases for a given day (p < 0.05). The red asterisk indicates that the change rate has a statistically significant response to the rain pulse on the respective day (Mann-Whitney U test between null model and distribution; p < 0.05). Refer to Table 1 and Figure S1 for site information. NEP, net ecosystem production; VOD, vapor pressure deficit. behavior in Figure 3 appears to obscure dependence on these characteristics at least in drier environments (Feldman et al., 2018;Potts et al., 2006;Reynolds et al., 2004;.

Drylands
Following smaller pulses under initially dry soil, no consistent carbon flux change occurs and plant water content consistently dries along with soil moisture (Figure 5a). Initial increases in VOD occur during the pulse suggesting a rehydration response. However, VOD dries consistently with soil moisture loss and no changes in carbon uptake are exhibited. These responses may indicate that the soil rewetting event was insufficient to create a significant photosynthetic response. We discuss this scenario further in the context of thresholds in Section 3.3.3.
Following larger pulses on initially dry soil, NEP increases occur with a lag beginning during the first day of soil drying, due likely to a delay in GPP upregulation (Figure 5b). The initial decrease in NEP may be because of a combination of increased respiration from rewetted soil microbes (Jenerette et al., 2008;Xu et al., 2004) and continued GPP loss during the previous interpulse period as supported by nighttime carbon fluxes and partitioned flux estimates ( Figure S7b). The 1-day lagged NEP increase appears to be due to a GPP increase, though its magnitude and duration of increase is less than when the soil is initially wetter as is consistent with previous findings (Potts et al., 2006). Additionally, VOD shows consecutive increases during and following the pulse with a similar delayed larger increase the day following the pulse ( Figure 5b). With drier initial soil conditions, these consecutive plant water content increases may in part represent slow rehydration due to previously dry conditions that may have created initially high soil-plant resistances FELDMAN ET AL.

10.1029/2020WR027592
13 of 20 Figure 6. Mesic ecosystem (MAP ≥ 500 mm/year) carbon flux (NEP) and plant water content (VOD) behavior following pulses binned based on pulse size and antecedent soil moisture conditions. The notation and format are the same as Figure 5. NEP, net ecosystem production; VOD, vapor pressure deficit.
Following smaller pulses on initially wet soil, NEP increases are subdued (with insignificant increases; p < 0.05) but occur immediately during the storm rather than with a day delay (Figure 5c). Based on partitioned estimates and nighttime fluxes, no respiration response occurs perhaps because of initially wetter conditions ( Figure S7c). Similarly, plant water content increases are weaker (Figure 5c). This similar initial VOD and NEP behavior indicates that wetter initial moisture conditions create quicker responses due to previous upregulation, but more intense rainfall may be required for a discernable response to be exhibited.
Following larger pulses on initially wet soils, dryland NEP and VOD exhibit the longest and greatest increases (Figure 5d; bin median is above overall median). Larger pulses enhance VOD and NEP increases beyond the second day of the drydown ( Figures S5 and S6). These longer VOD increases are potentially detecting rainfall-triggered dry biomass growth based on several lines of reasoning. First, the gradual VOD increases are not likely due to slow rehydration because hydraulic conductivities throughout the soil-plant continuum (including soil-root interface and xylem) are typically highest under initially wet conditions (Lo Gullo et al., 1998;Martorell et al., 2014;North & Nobel, 1997;Trifilò et al., 2004). Second, rapid growth requires high cell turgor with the help of wet soil as is the case under these conditions (Kramer & Boyer, 1995). Third, the longest duration carbon flux increases (likely from increases in GPP; Figure 4) are concurrent with these consecutive VOD increases reflecting rapidly upregulating and continued photosynthetic processes ( Figure 5d). Week-long growth responses to larger pulses have been detected in drylands previously (Post & Knapp, 2019). Furthermore, studies also found that primary production is of greater magnitude and longer duration following larger pulses on wetter antecedent moisture (Chen et al., 2009;Loik, 2007;Plaut et al., 2013;Tang et al., 2018). These observations therefore provide evidence for pulsed growth as hypothesized under the pulse-reserve paradigm (Noy-Meir, 1973).

Mesic Environments
While plants in mesic environments on average tend to respond to pulses, different pulse conditions greatly influence neither their carbon flux responses nor their plant water content responses (Figures 3 and 6). There are no significant differences between pulse characteristics in carbon flux behavior ( Figure 6, middle row; p < 0.05). VOD changes similarly show nearly identical magnitudes and signs of responses across all bins of pulse characteristics ( Figure 6, bottom row). Statistical significance is likely inflated by the large VOD sample size here.
It is well known from remote sensing and field studies that vegetation in humid regions, especially in the temperate Eastern United States, shows greater sensitivity to temperature and light than soil moisture variability (Madani et al., 2017;Nemani et al., 2003). This is in part because these more humid regions exist in an energy-limited state where soil water availability has less influence on evapotranspiration and GPP Koster et al., 2009;Short Gianotti, Rigden, et al., 2019). Given that there is still a vegetation response to pulses, but no response sensitivity to soil moisture pulse characteristics, pulse-response variability may be a greater function of other variables' dynamics. Temperature, VPD, and incoming solar radiation can increase on the scale of drydowns (Feldman et al., 2019). It has also been shown that VPD influences leaf gas exchange more in mesic environments (Novick et al., 2016). Additionally, nutrient availability during wetting and drying cycles can be variable based on soil microbe behavior (Gessler et al., 2017;Jarvis et al., 2007). We suspect that initial conditions of these aforementioned variables and their dynamics over a drydown may instead control differences in mesic vegetation pulse responses. Evaluation of these other environmental factors is beyond the scope of this work.

Soil Moisture Threshold Detection
We find that satellite VOD and flux tower NEP show soil moisture threshold behavior in drylands but less so in mesic regions (Figure 7). This difference in part explains why dryland vegetation shows no NEP changes and consistent plant drying under small pulse and dry initial conditions, while NEP and VOD increase under well-watered conditions ( Figure 5). Water-limitation may induce dormancy conditions, perhaps even intermittently between rainfall within a rainy season, where plants on average upregulate above basal photosynthetic activity only when rootzone moisture is increased to sufficiently wet conditions . Conversely, moisture thresholds appear less prevalent in mesic environments, where carbon and water uptake responses occur following both small and large pulses regardless of antecedent moisture conditions ( Figure 6). Moisture thresholds may still exist in mesic environments, but the land surface may only infrequently dry to states where moisture pulses are no longer ecologically important. Instead, temperature, humidity, nutrient, or radiation thresholds may more commonly control these environments (Jarvis, 1976).
Threshold behavior has been detected previously in semiarid environments, though primarily using rain depth (Chen et al., 2009;Y. Hao et al., 2010;Sala & Lauenroth, 1982;Tang et al., 2018). Use of soil moisture is preferred here because it holds more information about initial moisture conditions . Consistent with results here, soil moisture thresholds for vegetation function have been previously detected at the landscape scale (Bassiouni et al., 2020;Feldman et al., 2018).

Implications
Recent work showed that annual NEP is sensitive to interannual variations in rainfall totals and their projected changes (Gherardi & Sala, 2019;Maurer et al., 2020;Sala et al., 2015). However, reduced frequency and higher intensity rain events, apart from mean annual rainfall changes, are broadly expected under climate change Sala et al., 2015). Our findings show that vegetation across the US gradient of moisture availability shows significant carbon flux and water storage responses to pulses (Figures 2 and  3). Vegetation in drier environments is further sensitive to rain pulse characteristics, such as antecedent soil moisture and pulse magnitude ( Figure 5). Therefore, changes in rain pulse characteristics under climate change can conceivably impact regional-to-global water, energy, and carbon balances.
It is ultimately unclear whether the projected less frequent, more intense rainfall will increase or decrease annual carbon sequestration. Some rainfall manipulation field experiments shifting rain intensity and frequency while holding total rainfall constant showed increases in primary production (Heisler-White et al., 2008;Knapp et al., 2002;Thomey et al., 2011). Others with a similar manipulation showed no change or decreases in primary production (Y. B. Hao et al., 2017;Harper et al., 2005;Liu et al., 2017). These differing experimental results may be due to competing physical processes. Larger rain events create deeper FELDMAN ET AL.

10.1029/2020WR027592
15 of 20 Higher fraction values suggest more confidence in threshold detection. At a given site, a successful moisture threshold detection requires the dNEP/dt (or dVOD/dt) median at high soil moisture to be significantly greater than zero and greater than the median at low soil moisture (p < 0.05; see Section 2.7). NEP, net ecosystem production; VOD, vapor pressure deficit.
infiltration (shifting soil water availability deeper, Sala et al., 2015) and induce longer plant productivity responses, but longer interevent drying periods reduce overall productivity and result in lagged responses after the next event Plaut et al., 2013). We suspect that evaluating fundamental plant hydraulic and leaf gas exchange responses to pulses may provide a framework to evaluate these competing processes and potentially reconcile these experiments upon temporally upscaling to annual scales.

Summary and Conclusions
We use satellite-based landscape-scale plant water content and tower-based ecosystem-scale carbon flux measurements to evaluate vegetation responses to rain pulses across the continental United States. Our results are entirely observation driven, with observed pulse behavior not confounded by process-model assumptions. We find that plant water content and carbon fluxes show significant changes following moisture pulses across the US dry to humid climate gradient-not only in drylands where pulse field experiments have been primarily carried out. However, after aggregating sites into dryland and mesic environment groups, drylands show more frequent, longer plant water content and carbon flux responses following rain pulses. After further partitioning behavior based on pulse characteristics, only dryland vegetation tends to show discernable sensitivity to antecedent moisture and pulse magnitude. Specifically, large pulses and wet initial conditions result in the largest and longest dryland plant water content and carbon flux increase responses, evidence for pulsed growth responses. In contrast, small pulses and dry initial conditions show no carbon flux change and immediate plant water content loss along with soil moisture. These differences are in part due to soil moisture thresholds, below which plants are unable to upregulate, as found here primarily in drylands.
Ultimately, this study extends the assessment of vegetation pulse responses to a landscape scale across a dry to humid climatic gradient. It identifies similarities in plant responsiveness to pulse characteristics observed in independent carbon flux and plant water content metrics. Our results, therefore, show that novel satellite VOD observations hold meaningful vegetation information at daily to weekly time scales similarly to observed signatures in tower carbon flux measurements. Additionally, the study shows further evidence for soil moisture thresholds in both satellite and tower observations that a moisture pulse needs to exceed in order for primary production and water uptake responses to occur. These results have implications for how ecosystem plant behavior scales up to describe seasonal and annual primary production and vegetation sensitivity to rainfall characteristics.

Data Availability Statement
SMAP L1C brightness temperature used to retrieve soil moisture and vegetation optical depth are available from the National Snow and Ice Data Center (NSIDC) (https://nsidc.org/data/SPL1CTB_E). This work used eddy covariance data acquired and shared by the FLUXNET community.