We use 4 years (1 April 2015 to 31 March 2019) of soil moisture and plant water content observations from the Soil Moisture Active Passive (SMAP) satellite (Entekhabi et al., 2010). SMAP measures the low-frequency microwave (1.4 GHz) radiation emitted from Earth's surface. The radiation signal is in units of temperature, or brightness temperature (TB). The radiation is polarized, where the emitted waves' oscillations have distinct horizontal (TBH) and vertical (TBV) orientation. SMAP measures both TBV and TBH. Both TBV and TBH magnitudes alone are sensitive to surface soil moisture (top ∼ 5 cm). Furthermore, the difference between TBV and TBH is sensitive to how much the emitted waves are attenuated when traversing a vegetation canopy. The vegetation attenuation of the microwave radiation is called vegetation optical depth (VOD). More vegetation water content results in higher VOD (Jackson and Schmugge, 1991; Konings et al., 2019). An established radiative transfer equation can partition the TBV and TBH signals into soil moisture and VOD (Mo et al., 1982; Wigneron et al., 2017). We use a recently developed algorithm, called the multi-temporal dual channel algorithm (MT-DCA), to robustly estimate soil moisture and VOD using this radiative transfer equation (Konings et al., 2016, 2017).

The SMAP satellite measurements occur at 06:00 (local time) everywhere on 9 km grids across the globe. The time 06:00 is approximately predawn, when plant water status is assumed to be maximal (due to nighttime plant rehydration). The satellite orbit is such that there is a 1, 2, or 3 d revisit, depending on the day and latitude. Due to the orbit pattern, higher latitudes are measured more frequently. This results in sampling frequencies of 1–2 d at midlatitudes and 2–3 d at the Equator.

Since VOD has been shown to be nearly linearly proportional to total vegetation water content (Jackson and Schmugge, 1991), VOD is proportional to the product of relative water content and aboveground dry biomass (Konings et al., 2019; Momen et al., 2017; Zhang et al., 2019). Therefore, VOD can increase due to either rehydration of cell water storages or biomass growth, as growth provides additional water storage capacity. VOD is expected to be sensitive to rehydration because of near-linear relationships between relative water content and plant water potential, especially for herbaceous species which are primarily investigated in this study (Jones, 2014; Jones and Higgs, 1979; Konings et al., 2019; Nobel and Jordan, 1983). While the low resolution of VOD estimates hinders species-specific or stand-scale assessments, it provides the opportunity to assess integrated, landscape-scale vegetation behavior across global biomes (Feldman et al., 2018; Tian et al., 2018). VOD shows promise for use in monitoring plant water stress, with recent findings showing VOD can monitor time evolution of plant water stress and drought-induced mortality with loss of plant water storage (Feldman et al., 2020; Martínez-Vilalta et al., 2019; Rao et al., 2019).

Soil moisture observations from the MT-DCA algorithm compare closely to other SMAP soil moisture products (which use different algorithms) as well as to in situ observations (Chan et al., 2016; Dadap et al., 2019; Feldman et al., 2018). Direct in situ VOD information is unavailable, although SMAP VOD's mean and dynamics are comparable to another satellite VOD product (Kerr et al., 2010). For further discussion of SMAP VOD estimate performance and comparison with other products, we refer the reader to Konings et al. (2017) and Feldman et al. (2018).

To assist in discriminating VOD changes related to hydraulic or growth activity, we use the daily leaf area index (LAI) product from the Spinning Enhanced Visible and Infrared Imager (SEVIRI) on board EUMETSAT's Meteosat Second Generation (MSG-2) satellite series (Trigo et al., 2011a). These LAI observations serve as an indicator for above-ground biomass independent of VOD. While constrained primarily to Africa, these LAI observations are estimated from 15 min geostationary observations which provide daily LAI fluctuations after cloud contamination mitigation. Both VOD and LAI datasets together are required to determine the occurrence of pulse-driven growth. VOD increases can be linked directly to a specific rain event because of SMAP's more rapid effective sampling (due to no cover contamination), but they are confounded by rehydration. LAI changes over the weekly scales of pulses can detect canopy growth, but because of a non-linear averaging technique (García-Haro and Camacho, 2014), the LAI dataset is partially smoothed over sub-weekly scales and may be less apt to determine whether detected growth over a week is specifically associated with a given rain event. Nevertheless, increasing LAI over a rain event can identify whether VOD increases associated with that storm are due to growth or only rehydration. As such, we are interested in using LAI changes qualitatively to determine whether LAI is increasing or decreasing over more than week-long periods. Therefore, biases in LAI magnitude do not influence the analysis.

Use of SEVIRI LAI for this application is preferred due to SEVIRI's frequent sampling and filtering techniques that provide better resolution of the seasonal growth and senescence stages, especially during the wet season, than other available satellites (García-Haro et al., 2013; Gessner et al., 2013). SEVIRI ultimately provides 3–5 d effective sampling during cloud-contaminated periods, which is typically 4 times less than effective sampling with low-Earth-orbit satellites (i.e., MODIS) (Fensholt et al., 2006). Therefore, despite being global, low-Earth-orbit satellites are not used because they sample too coarsely in time for the applications here. Furthermore, SEVIRI LAI retrievals in the herbaceous biomes evaluated in Africa have the lowest retrieval errors (García-Haro et al., 2013). Therefore, SEVIRI LAI is likely to detect increasing biomass over 1- to 2-week periods. While no other adequate satellites exist for direct comparison with results here, the analysis was repeated with the SEVIRI fraction of absorbed photosynthetically active radiation (FAPAR) observations, derived from different measurement frequencies than LAI, and similar results were obtained (Fig. S1).

Ancillary data are used to evaluate climate and biome dependencies of the findings. Specifically, we compute mean annual precipitation using the Global Precipitation Measurement IMERG product (Huffman, 2015) and tree cover from the Moderate Resolution Imaging Spectroradiometer (MODIS) (Dimiceli et al., 2015) to evaluate VOD behavior across climate gradients. International Geosphere-Biosphere Programme (IGBP) land cover maps are used to remove frozen and bare ground (Kim, 2013). Tree cover is also used to remove densely vegetated forests where soil moisture and VOD satellite estimation are uncertain.

VOD behavior and response timescales are evaluated during soil moisture drydown periods that occur after rainfall (Fig. 1). Drydown periods, or soil moisture pulses, are defined as an increase in soil moisture of at least 0.01 m3/m3 followed by a drying period of at least four consecutive measurements (approximately 8–12 d). This approach is nearly identical to previous approaches (McColl et al., 2017; Shellito et al., 2018). To remove seasonal drydowns less associated with an individual rainfall event, drying periods of longer than 20 d are not included. Seasonal trends (periodic climatology) are removed from soil moisture and VOD time series during drydowns, while preserving the magnitude of initial conditions (Feldman et al., 2019). This procedure removes seasonal VOD growth trends to isolate short-term increases associated with a given storm that are due to pulse-driven growth and/or slow rehydration. Note that differences persist in the literature on whether the rain event or plant response is defined as the pulse (Reynolds et al., 2004). Both soil and plant responses are discussed as pulses here.

For infiltrating water to ultimately reach the leaf, it must typically percolate from the soil surface to the roots, pass through the root endodermis, and move up the xylem through the shoot to leaves (Mackay et al., 2015; Sperry et al., 1998, 2016). Given an identified soil moisture drydown period, the VOD response timescale, defined here as time to peak (tp), is estimated as the time from the beginning of the soil moisture drydown to the first local maximum value of VOD (Fig. 1). After a period of water storage, plant water content loss occurs due to surface and atmospheric drying and warming, creating a peak in plant water content conditions that tp attempts to quantify (Feldman et al., 2020). tp captures the aggregate rehydration and growth timescale during this soil–plant water transport process. The tp estimation relies on consecutive VOD increases, which provide more robust estimates than the global maximum during the drydown. To increase sample size, we conduct the analysis on all pixels contained within a 0.5∘×0.5∘ domain (includes ∼ 30 SMAP pixels). Densely forested regions (> 40 % tree cover; such as the Congo and Amazon basins) are masked because soil moisture and VOD estimates are less certain from radiative transfer limitations in dense canopies (Feldman et al., 2018; Konings et al., 2017).

We compute the median tp over all drydowns within each 0.5∘×0.5∘ pixel. The tp probability mass function within a given pixel typically has a mixed distribution with many zeros, resembling a zero-inflated Poisson distribution. The median tp is chosen to describe this distribution because it not only provides a typical timescale of VOD increase, but also indicates whether or not the majority of pulses resulted in consecutive, multi-day VOD increases (as opposed to the mean, which can be greater than zero even if a majority of pulses resulted in a tp of zero). Several tests are performed to determine the effects of SMAP's irregular, above-daily sampling period, the algorithm, and measurement noise on tp estimates for a given pulse (see Sect. 2.4).

The tp definition evaluates continuous post-rainfall VOD increases and potentially neglects the duration of plant water content increases during the period of soil moisture increase (between the observations before the drydown beginning and at the drydown beginning). We do not attempt to estimate the duration of VOD increase during the soil moisture increase period because it is not possible to resolve when plant water content increases initiated due to the 1–3 d satellite sampling frequency. Instead, the VOD behavior preceding the drydown is categorically evaluated by determining the frequency of plant water content increases during the rain pulse. This allows evaluations of tp of zero, which can result from either a rapid rehydration response during the rainstorm (on the order of hours) or no rehydration response throughout the pulse (no VOD increase).

For each soil moisture pulse within a pixel, tp is estimated along with the LAI change from beginning to end of the drydown (ΔLAI), antecedent surface soil moisture (soil moisture value before drydown beginning), soil moisture pulse magnitude (difference between initially pulsed and antecedent surface soil moisture), and antecedent VOD. Antecedent is defined here as the observation just preceding the peak soil moisture observation beginning the drydown. Each variable is binned into rapid VOD response (tp=0), short VOD increase (1≤tp≤3 d), and long VOD increase (tp>3 d) groups because they provide partitions consistent with the satellite sampling and because uncertainty analyses reveal that while a tp estimate for a given drydown is uncertain, there is more confidence in whether it exists within a given bin (see Sect. 3.4). The groups of three different tp lengths are then compared for each respective metric of ΔLAI, antecedent surface soil moisture, soil moisture pulse magnitude, and antecedent VOD. Due to non-normality of groups based on Jarque–Bera normality tests (Jarque and Bera, 1980), Kruskal–Wallis non-parametric tests are performed to determine significance of difference in medians between the tp groups for each respective metric. Also, correlation coefficients are computed between tp and ΔLAI, antecedent moisture, and pulse magnitude to augment the categorical analyses.

The seasonal timing of rapid, short, and long tp values is assessed relative to peak seasonal moisture, or the proximity to the wet season. The peak seasonal soil moisture is determined by smoothing the soil moisture times series using a 90 d moving-average window. This only provides a zero-order seasonal moisture peak approximation as many locations have intermittent rainfall or bimodal precipitation distributions.

Several tests were conducted to evaluate the robustness of tp estimates given uncertainties due to a 1–3 d satellite sampling frequency, the soil moisture–VOD retrieval algorithm, and random instrument noise on the order of that of the SMAP radiometer (Piepmeier et al., 2017). A stochastic rainfall generator was used to simulate soil moisture and consequent drydowns. A range of “true” VOD behavior was considered such as perfect correlation with soil moisture (true tp of zero) and multi-day VOD increases during drying (true tp greater than zero). Analyses were conducted directly on these simulated time series, including converting these time series to TB measurements for implementation in the algorithm and comparing the original true VOD time series to the algorithm-estimated VOD time series as in Zwieback et al. (2019). For tests with the 1–3 d satellite sampling frequency, the effect of randomly removing observations every 1–2 d on tp was assessed. To test the effect of the soil moisture–VOD retrieval algorithm on tp, tp was estimated after inputting true TB measurements into the retrieval algorithm. Finally, to assess the effect of instrument noise on tp estimates, this aforementioned process was repeated by adding normally distributed random error to TB measurements.

To investigate the underlying mechanisms that alter plant rehydration timescales, we evaluate plant hydraulic storage timescales under varying conditions after a surface soil moisture pulse using a plant hydraulic model. We specifically choose a one-dimensional soil–plant–atmosphere continuum (SPAC) model, assessed in previous studies (Carlson and Lynn, 1991; Hartzell et al., 2017; Lhomme et al., 2001; Zhuang et al., 2014). Note that assimilating satellite VOD into a SPAC model is beyond the scope of this study and is hindered by the large number of unknown plant hydraulic parameters at global scales. SPAC simulations are repeated and randomized using a Monte Carlo approach, drawing from parameter distributions based on previous field measurements. More details can be found about the SPAC model in the SI.

The VOD data show that more arid regions, with lower annual rainfall and tree cover (Fig. 2b and c), exhibit multi-day vegetation water content increases following moisture pulses (tp≥1 d, blue regions in Fig. 2a). That is, after soil moisture increases following a storm, vegetation water content increases for multiple days even while surface soil moisture begins to dry. Furthermore, in regions with tp≥1 d, VOD typically begins increasing during the rain pulse period instead of with a lag after soil moisture drying begins (occurs in 77 % of the pixels). Aggregated example time series of this nonzero tp behavior can be seen in drylands in the Sahel and southwest United States (Fig. 3a and b). In the regions with multi-day VOD increases, the spatial median tp is 2 d. Note that various responses are spatially aggregated together to produce the post-rainfall responses in Fig. 3. In subsequent sections, we evaluate and partition the mechanisms underlying these multi-day plant water content increases primarily in drylands (blue regions in Fig. 2a).

By contrast, more humid ecosystems with more woody plant coverage typically do not exhibit multi-day plant water content increases (tp=0; Fig. 2). They instead exhibit water loss following the pulse during soil drying (see average behavior illustrated in Fig. 3c and d). In 83 % of regions with tp of zero (red regions in Fig. 2a), the plant drying responses are typically preceded by an initial VOD increase, showing rapid water uptake during the storm period (Fig. S2). In contrast, a minority of these regions typically show no VOD increases, suggesting plant water content continuously dries throughout the pulse with no discernable hydraulic response (Fig. S2). We do not investigate regions with median tp of zero further here because their exact sub-daily timescales are unresolvable, but within expectations (see Discussion).

A positive correlation between LAI rates of change and plant water content increase timescales is found in 72 % of African pixels with median tp≥1 (p<0.05). Therefore, longer tp values are associated with increasing biomass within a given pixel (Fig. 4). Calculating the LAI rates of change for the rapid VOD response (tp=0), short VOD increase (1≤tp≤3), and long VOD increase (tp>3) groups reveals that growth tends to occur alongside plant water content increases longer than 3 d (Fig. 5a, c, and d). These longer plant water content uptake timescales average 7 d and continue beyond a week 40 % of the time. This growth influence means that rehydration alone cannot explain longer plant water content increase durations. Note that VOD increases during growth still demonstrate increased aboveground plant water content because more aboveground biomass requires water uptake to hydrate a greater volumetric plant storage capacity. There are some pixels that show declining biomass during longer tp (Fig. 5d). We attribute these cases to detection of longer tp during senescence in regions where senescence of leaf area is differentially more rapid than growth. Ultimately, we interpret overall spatial patterns and avoid interpreting individual pixels, acknowledging noisy tp estimates in some cases (see Sect. 3.4).

In general, growth does not influence shorter plant water content increase timescales; LAI is often decreasing when tp is 1–3 d (Fig. 5a). Therefore, plant water content increases over less than 3 d are mostly due to rehydration. Furthermore, when VOD increases do not extend beyond a day (tp=0), growth is also less frequently occurring.

The reoccurrence of growth-influenced, multi-day VOD increases consistently following soil moisture pulses means that rainfall intermittently triggers growth throughout a year. tp values greater than 3 d are linked to pulse-driven growth because they coincide with increasing daily LAI (Fig. 5), consistently co-occur with a soil moisture pulse, and are separated from seasonal growth patterns. Our seasonal detrending of VOD isolates these pulsed plant growth responses from seasonal growth cycles. These isolated sub-weekly VOD responses closely link to the timing of moisture pulses, suggesting a cause–effect of rain pulse followed by plant water content response.

Although this daily LAI dataset is limited to Africa only, Africa contains one-third of the world's regions with median tp≥1 d (blue regions in Fig. 2a), and we expect similar results for the rest of the globe. Note that these results are not sensitive to the 3 d threshold choice between long and short VOD increase groups; they are nearly identical if choosing a threshold of 2, 4, or 5 d. Furthermore, results repeated with FAPAR are qualitatively the same (Fig. S1; see Sect. 2.1).

On average, the short and long VOD increase bins occur approximately with equal frequency, both with seasonal variations (Fig. 5b). Longer-duration VOD increases influenced by growth (Fig. 5a) appear to occur more frequently during times of the year when soil moisture is higher (Fig. 5b). In contrast, short VOD increases, associated more with rehydration, occur more often during drier times of the year (Fig. 5b). Furthermore, rapid rehydration responses occur 40 %–50 % of the time throughout the year amongst the multi-day VOD increases.

LAI growth rates average 0.005 m2/m2 per day for these long VOD increases. On a mean percent change basis, this translates to a 15 % LAI increase on average over the course of a week after a pulse. Note that LAI may not detect additional branch–stem biomass growth that VOD may detect. Ultimately, we are more interested in qualitatively increasing trends in LAI rather than the magnitudes of LAI rates of change which are less certain.

Variations in VOD increase timescales across space and time likely occur as a result of differences in vegetation traits, edaphic and topographic properties affecting soil moisture infiltration, and climatic properties. While an evaluation of all of these factors is beyond the scope of this paper, we focus here on climatic drivers. To evaluate the climatic drivers of VOD increase timescales in regions with median tp≥1 d (blue regions in Fig. 2a), we assess how tp relates to rain pulse conditions: antecedent surface soil moisture, soil moisture pulse magnitude, and antecedent VOD. Growth-influenced VOD increases of longer duration are associated with initially wetter surface soil (Fig. 6a) as well as with larger pulse magnitudes (Fig. 6b). This suggests that the surface must be sufficiently wet initially, and a large enough pulse must occur to elicit a growth response. Conversely, shorter-duration VOD increases associated primarily with rehydration frequently occur under drier initial soil conditions with smaller rewetting pulses (Fig. 6). This is consistent with short increase durations becoming more prevalent during drier periods and long increase durations becoming more prevalent in wet periods (Fig. 5b). Note that while these results are shown globally, they are nearly identical when calculated for only Africa (not shown), and therefore they can be consistently compared with the growth assessment results and timescale bins (Sect. 3.2; Fig. 5).

In assessing what differentiates rapid responses (tp=0 d) and short VOD increases (tp=1–3 d) that appear driven by only rehydration, we find short VOD increases have slightly larger pulse magnitudes (Fig. 6b) and drier antecedent soil moisture than rapid responses (Fig. 6a). Also, drier initial plant water status for short VOD increases (Fig. 6c) independently suggests a slightly drier root zone initially than for rapid responses (Fig. S13). Note that mean differences are small between these metrics, even though they show statistical significance (likely effect of large sample size deflating p values). Nevertheless, cases of vegetation water content increase on the order of 1–3 d, due primarily to rehydration, occur under dry soil conditions with small to moderate rewetting pulses.

Satellite tp estimates appear robust with effects of satellite sampling frequency, algorithmic estimation error, and measurement noise increasing tp variance, but not introducing discernable biases. The SMAP sampling period of 1–3 d results in greater variance but no mean biases for tp estimates below the Nyquist frequency of 4–6 d (Figs. S4 and S5). One can combine low-frequency microwave measurements from similar satellites (Kerr et al., 2010) to increase the sampling frequency and reduce uncertainty in tp estimates here. This is not attempted due to complications in combining the datasets. The MT-DCA algorithm used here reduces sensitivity to noise within the simultaneous soil moisture–VOD estimation (Konings et al., 2015, 2016; Zwieback et al., 2019). We found that use of a traditional algorithm biases tp towards zero (Fig. S7) because its greater sensitivity to noise will tend to spuriously induce positive correlation between soil moisture and VOD within the estimation procedure (Konings et al., 2016). Therefore, increases in VOD during soil drying and thus positive tp values are not a result of algorithmic artifacts from the MT-DCA algorithm used here (Feldman et al., 2018). It is also unlikely that algorithmic noise is driving spatial patterns as both algorithms produce the same tp spatial patterns. Note that the MT-DCA algorithm can slightly artificially increase tp, though measurement noise may cancel this effect (Fig. S4). Finally, measurement noise primarily increases the variance of tp (Fig. S4).

Ultimately, while identifying precise tp values for a given drydown may be hindered by these sources of uncertainty, median tp values for a pixel are likely not biased, and more confidence is exhibited in whether tp is zero or non-zero (Fig. S6). This uncertainty analysis provides confidence in the global patterns of median tp and results based on binned tp where zero, short, and long tp can be confidently partitioned.

We observe a continuum of plant water uptake timescales from humid to dryland environments, with mainly drylands showing frequent multi-day plant water content increases after rainfall before water loss occurs (Fig. 2). Given that plant hydraulic capacitance increases at least 3 orders of magnitude from grasses in drylands to trees in humid regions (Carlson and Lynn, 1991; Hunt et al., 1991), one might expect, if at all, occurrence of multi-day responses in wooded regions. However, humid, wooded regions broadly exhibit peak plant water content during rather than after the storm event, before soil drying begins (Figs. 2 and S2). Plant water loss occurs thereafter (Fig. 3c and d), likely due to simultaneous soil and plant drying where plant rehydration becomes progressively restricted with drying soil (Feldman et al., 2020). The initial VOD increase can be due to plant water uptake where pre-dawn water potential approaches equilibrium with soil moisture and/or due to plant interception of rainfall droplets. In some cases, no discernible VOD increase occurs before or after the pulse, which may indicate sufficiently well-watered conditions (Fig. S2). Even in drylands, pulse water utilization for plant rehydration decreases if the plant–soil system is initially sufficiently wet (Ehleringer et al., 1991; Gebauer et al., 2002; Ignace et al., 2007). Nevertheless, due to the 1–3 d satellite sampling, we are unable to resolve more specific plant water content timescales and underlying mechanisms for these well-watered, wooded regions.

The consistent trend of multi-day plant water content increases, which are found broadly across dry regions (Fig. 2), is unexpected, at least in the context of nominal RC time constants (plant water uptake and storage timescales). Field-based estimates of plant water uptake timescales (via RC plant hydraulic time constants) typically do not exceed a day, regardless of species (Huang et al., 2017; Nobel and Jordan, 1983; Phillips et al., 1997, 2004; Ward et al., 2013). This is in part because plant capacitance and resistance tend to trade off with changes in plant architecture and moisture conditions (i.e., capacitance increases and resistance decreases generally from grass to tree species) (Hunt et al., 1991; Phillips et al., 1997; Richards et al., 2014; Ward et al., 2013). We find both the influence of growth and slow plant rehydration contribute to these observed multi-day VOD increases. We discuss these growth and plant rehydration mechanisms observed in drylands further below.

As is evident in independent satellite LAI observations, growth increases the duration of plant water content increases (Fig. 4) and appears to occur primarily for plant water content increases of more than 3 d in dryland regions (Fig. 5). These week-long consecutive plant water content increases occur when the soil is initially wetter and pulses are larger (Fig. 6). These results are based on 1–2-week increasing trends in LAI coinciding with VOD increases of more than 3 d. Confidence is exhibited in these sub-monthly LAI trends because of SEVIRI's ability to resolve the seasonal growth stages during the wet season, lower LAI uncertainty in Africa's biomes with herbaceous vegetation, and SEVIRI's filtering of LAI noise. Therefore, plant rehydration alone cannot explain these longer-duration VOD increases. We further suspect rehydration is rapid under these well-watered conditions. While pulsed growth is expected to occur with a lag of 1–5 d (Ogle and Reynolds, 2004), these lags may be obscured in the sampling of VOD and initial VOD increases due to rehydration. Furthermore, these pulsed plant water content increases due to growth may continue for longer than detected here (beyond 2 weeks). However, continued water loss and VOD decreases through transpiration may eventually dominate over VOD increases due to growth, curtailing the peak VOD (resulting in behavior like that shown schematically in Fig. 1). VOD ultimately shows sub-weekly growth temporal dynamics beyond those resolved from optical instruments.

These results indicate that large soil moisture pulses on initially wetter soils trigger dryland vegetation growth responses after storm events, as hypothesized under the pulse reserve paradigm (Collins et al., 2014; Noy-Meir, 1973). This weekly variability, at least in part, drives seasonal growth in these locations (Reynolds et al., 1999) where the seasonal growth cycles appear to be made up of sub-weekly intermittent growth dynamics as modeled in Ogle and Reynolds (2004). The growth occurrences under wetter conditions are expected given that cell turgor must be high for cell expansion and rapid growth to occur (Kramer and Boyer, 1995). Furthermore, a recent study showed that larger pulses during the growing season resulted in 1–2 weeks of increasing leaf and stalk density in a semi-arid grassland, consistent with results here (Post and Knapp, 2019). Additionally, larger pulses have previously been shown to elicit greater plant photosynthetic responses (Chen et al., 2009; Dougherty et al., 1996; Schwinning and Sala, 2004). In a similar study, these longer satellite-based plant water uptake responses coincided with larger and longer carbon uptake responses at dryland flux tower sites following larger moisture pulses on initially wet soils (Feldman et al., 2021). Therefore, detection of pulse-triggered growth on timescales of drydowns here is consistent with previous results, although it is the first to show how widespread the pulse-triggered growth dynamics are in drylands. Additionally, the seasonal occurrence of growth-driven, longer tp (Fig. 5b) supports the fact that pulses will trigger growth primarily in the season when species are phenologically active and able to invest in aboveground biomass (Post and Knapp, 2019; Reynolds et al., 1999; Schwinning and Sala, 2004).

Over half of the moisture pulses, primarily in global drylands, result in multi-day satellite-observed plant water content increases (Fig. 2). These multi-day VOD increases are often only due to rehydration, especially the shorter VOD increases (1–3 d) following small to moderate pulses on initially dry soils (Figs. 5 and 6). They can occur even when biomass is decreasing (Fig. 5c; such as leaf off), where the relative water content increases are larger than what the VOD increase signal suggests. For dryland ecosystems that include grass and shrub species with isolated forests, multi-day rehydration is generally unexpected with nominal RC time constants on the order of an hour (Carlson and Lynn, 1991; Hunt et al., 1991). However, previous field studies often show 1–4 d rehydration of grasses and shrubs upon rewetting following dry conditions, especially in the southwestern United States, where multi-day VOD increases are observed (Briones et al., 1998; Fravolini et al., 2005; Huxman et al., 2004; Ignace et al., 2007; West et al., 2007).

To better understand the physiological drivers of multi-day rewetting, we assessed the potential hydrologic and physiological mechanisms driving slow rehydration using a plant hydraulic (SPAC) model and parameters within known bounds for semi-arid species (Figs. S8 to S14 and Table S1). We find that the sufficient conditions for multi-day plant rehydration determined here include initially high soil–plant resistances decreasing over multiple days following a storm. These time-varying resistances can occur either in the soil, plant, or both (Figs. 7, S9, and S11). The possibility of multi-day rehydration due to these conditions suggests that RC timescales can greatly deviate from nominal conditions (Scholz et al., 2011), especially under drought scenarios where resistances are both higher and changing.

After uncoupling effects of soil and plant resistances in the SPAC model, we suspect that multi-day rehydration as seen by VOD is dominated by plant resistance limitations rather than soil resistance limitations. This is because high soil resistances reduce infiltration rates and result in a phase-lagged delay in plant rehydration (Fig. 7b), which is not observed in the satellite VOD behavior here. In the slow rehydration cases (tp=1–3 d), VOD increases begin immediately during the storm and not with a phase-lagged delay (Fig. S3). This behavior more closely resembles slow plant rehydration dominated by plant resistance limitations rather than those dominated by soil resistance limitations. For example, 1–3 d uptake timescales based on satellite VOD observations appear like that in Fig. 3a and b, which more closely resemble SPAC model simulations in Fig. 7a than in Fig. 7b. Note that both conditions may be present within a coarse-resolution pixel because the pixel spatially averages plant water content behavior over the landscape. As a result, a combination of behaviors like those in Fig. 7 aggregate into the spatially averaged behavior, like that shown in Fig. 3a and b. Therefore, while plant resistance limitations may dominate most landscapes that show 1–3 d VOD increases based on the above discussion, slow infiltration responses may still be spatially prevalent, with a potential dependence on sub-pixel antecedent moisture variability.

The initially high, decreasing resistances, as determined from the SPAC model and likely influencing landscape-scale plant water content behavior, are likely due to drought recovery of the soil–root interface and xylem architecture. Initially high, decreasing plant resistances have been observed in the field, where after rewetting of dry soil conditions, soil–root interface and xylem resistances can decrease by 1 to 3 orders of magnitude over a few days (Carminati et al., 2017; North and Nobel, 1995; Trifilò et al., 2004; West et al., 2007). Under prolonged dry conditions, a disconnect between soil and root interface can occur, and after rewetting, the soil–root and radial root hydraulic conductivity progressively increase (Carminati et al., 2009; North and Nobel, 1997). Similarly, xylem cavitation and embolism from drying lead to increased xylem resistance that can regain conductance and refill after rewetting (Martorell et al., 2014), though noting controversies with existence of xylem repair and refilling (Charrier et al., 2016; Lamarque et al., 2018; Venturas et al., 2017). Recent evidence suggests that whole-root resistance (i.e., soil–root interface, radial) rather than xylem resistance (from cavitation) dominates the whole-plant resistance during these drying and rewetting cycles (Rodriguez-Dominguez and Brodribb, 2020). Finally, fine root growth can occur after rewetting, which can contribute to decreasing root resistances, though these effects may occur over longer, weekly scales (Eissenstat et al., 1999).