The study site, the Charles H. Herty Pines Nature Preserve, is a forest fragment in Statesboro, Georgia, USA (Fig. 2a), at Georgia Southern University's main campus (32.430∘ N, 81.784∘ W; 65 m a.s.l.). The climate is subtropical (Köppen Cfa) with mean monthly temperatures (1925–2014) in July that range from 21 to 33 ∘C and generally mild winter months, i.e., the lowest mean January temperature is 3.5 ∘C (University of Georgia, 2019). Mean annual precipitation is 1170 mm yr-1, and precipitation occurs almost exclusively as rain, which is relatively evenly spread over the year. The overstory is dominated by Pinus palustris (longleaf pine), and overstory rainfall partitioning for this site has been reported (Mesta et al., 2017; Van Stan et al., 2018; Yankine et al., 2017). The trunk diameter at breast height (DBH) was relatively consistent across all trees in the study plot: 49.7 cm (mean) with an interquartile range of 36.2–55.7 cm. The mean tree height was 30.4±4.5 m and was derived from terrestrial lidar (terrestrial lidar methods identical to Van Stan et al., 2017a). The stand density was 223 trees ha-1 with 50.4 m2 ha-1 of basal area. Dogfennel, our study plant, was particularly dominant along the forest edge. Dogfennel is a forb of the Asteraceae family that is native to (and widespread across) North America (Van Deelen, 1991; Wunderlin and Hansen, 2003). Although dogfennel behaves as an annual plant throughout much of its North American range, it can behave as a perennial in the southern US by overwintering as a rosette, typically from January to March, before regrowing from a taproot in the spring, typically in April (Macdonald et al., 1992, 1994). Dogfennel can be abundant in disturbed forest understories, particularly pine forests (Brockway et al., 1998) and pastures (Fig. 2b). In the study pine forest, the dogfennel stem density was 56 770 stems ha-1 along the stand edge. In pasturelands, dogfennel can reach this stem density within a single season and, if left unmanaged, dogfennel densities have been measured as high as 74 stems m-2, or ∼740000 stems ha-1 (Dias et al., 2018). The growth habit of dogfennel results in “clumps” of stems. The dogfennel density was estimated in ten 10 m ×10 m plots by counting the stems per clump for three randomly selected clumps in each plot. For each plot, the mean stems per clump were multiplied by the number of clumps per plot. Finally, all stems per plot were summed and scaled to 1 ha. Three dogfennel clumps were randomly selected for throughfall and stemflow monitoring. Within these three clumps, 30 individual dogfennel stems were randomly selected for stemflow monitoring. Individual plant attributes – canopy radius (cm), stem radius (cm), leaf angle at the stem (degrees from vertical) at various canopy heights (1.00, 1.25, 1.50, 1.75, 2.00 m), and the relative location within the clump, interior (I), middle (M) or exterior (E) – were measured for each stemflow-instrumented plant (Table 1). Canopy and stem radii were determined manually with a tape measure, where canopy radii were the mean of measurements from eight directions (N, NE, E, SE, S, SW, W and NW) and stem radius was determined by a single manual measurement at the stem base. The leaf angle at the stem was determined for two leaves at each height using the Protractor™ app for iPhone (2013, Phoenix Solutions) which logs an angle after the leveling of the iPhone camera (see Fig. S1 in the Supplement for an example).

Maximum water storage capacity, Su (mm), was estimated for the dogfennel canopy and stem, both as volume (L) per unit surface area (m2). All field leaf and stem samples were collected during an inter-storm dry period (>24 h after any rainfall). For the canopy, 50 leaves representing the median size of the site dogfennel plants were sampled (broken off at the base of the leaf), taken back to the lab, their “field-dry” mass (g) was determined on a bench scale and then the broken ends of their leaf-stems were sealed with silicon to prevent water exchange from an area that was not previously exposed in its natural state. Sampling for the stems was similar; however, as dogfennel heights reach (and can exceed) 2 m, the stems were cut into 5 cm sections. Just as with the leaves, 50 representative samples of these stem sections were weighed in the lab and then sealed with silicon on both ends. Next, all leaf samples and stem sections were submerged in water for 3 d until maximum saturation was achieved (per Van Stan et al., 2015), whereupon the maximum saturation mass (g) was recorded. For comparison with the field-dry mass, all samples were oven-dried until their mass no longer changed (mass recorded every 3 h), whereupon the oven-dried mass (g) was recorded. No leaf or stem samples were oven dried longer than 15 h. The gravity convection oven (Isotemp, Fisher Scientific) was set to 40 ∘C (confirmed with a standard thermometer). The maximum volume of all samples' water storage capacity is the difference between the saturation and oven-dried masses. The oven-dried leaves and stems did not visually appear to be damaged (aside from the sampling cuts, obviously), and care was taken to ensure the plant samples were not damaged. It is likely that internal (not externally intercepted) water was exchanged during this process; however, this is not entirely problematic as plant surfaces are known to permit interaction between externally intercepted water and internal water (Berry et al., 2019). Moreover, we explicitly acknowledge that although these submersion methods are commonly used, they produce the “maximum” possible water storage capacity (hence, our objective to estimate maximum water storage capacity), as multiple intrinsic and extrinsic factors of plant surfaces could reduce the available water storage capacity in situ (Klamerus-Iwan et al., 2020).

Specific water storage capacity, SL (mL cm-2), for the leaves and stems was determined by dividing the lab-derived maximum volume (mL) by the samples' surface area (cm2). For leaves, after sampling, leveled photos of each sample were taken on a grid system (every block representing 2.5 cm × 2.5 cm for scale), and the leaf images were then vectorized and processed for 2-D projected surface area using the “Measure Path” extension in Inkscape (v. 0.92, Inkscape.org). An example vectorized image of leaf area is provided in the Supplement (Fig. S3). Error in this vector-based leaf surface area estimate was estimated by repeating the process five times for each leaf. Stem surface area for all samples was estimated from their radii and height. SL estimates for the stem (0.436 mL cm-2) and leaves (0.195 mL cm-2) were then scaled to SU (mm as L m-2) using stem and leaf surface area estimates per plant (A=171.9 and 807.5 cm2 per plant, respectively), and multiplied by the site plant density (D=5.68 plants  m-2) before being divided by 1000: 2SU=SLstem×Astem×D/1000+SLleaf×Aleaf×D/1000. Plant stem and leaf surface area estimates were determined from five representative plants that were cut from the site and separated into leaves and stems, and the sums of the leaf and stem areas (determined as mentioned earlier in the paragraph) were then divided by 5. Total leaf surface area compares well to values reported from ∼1 m tall dogfennel plants (212 cm2 per plant; Carlisle et al., 1980) considering that our plants were much taller (∼2 m).

Descriptive statistics were compiled for all variables presented and regression analyses were performed to relate plant canopy and hydrologic variables. All statistical analyses were done using Statistica 12 (StatSoft, Tulsa, OK, USA). Throughfall volumes (L) from all gauges were summed and converted to yields (mm) by dividing by the total gauge area (m2). Stemflow yield (mm) for an individual plant was determined by dividing its volume (L) by the projected canopy area (m2). To compare stemflow production across plants, two metrics were computed per plant for each storm: normalized stemflow, P‾S,i (–), and the funneling ratio, F (–). P‾S,i was computed per Keim et al. (2005): 3P‾S,i=PS,i-P‾SsS, where PS,i is stemflow volume (mL) from each individual plant in a single storm, P‾S is the mean stemflow for all plants in a single storm and sS is the standard deviation of stemflow for all plants in a single storm. F values for individual plants in each storm were computed per (Herwitz, 1986): 4F=PS,iBiP, where Bi is the basal area (cm2) at the base of an individual plant, and P will be either Pg or PT,o′ (this will be explicitly indicated in the results). There are an increasing number of F metrics (Carlyle-Moses et al., 2018; Levia and Germer, 2015); however, the selected method is the most common F metric applied to stemflow data to date. Moreover, in situ observations of non-collared dogfennel plants during rainfall confirmed that dogfennel PS rates did not produce visible runoff areas.

Discrete rain events, as measured above the forest canopy, ranged in magnitude from 0.1 mm (during dewfall) to 101.3 mm (Table 1). The distribution of storm magnitudes was skewed, such that the mean, 16.5 mm, was many times greater than the median, 6.6 mm (Table 1). Estimated overstory throughfall (PT,o′), as shown in Fig. 3, ranged from 0 (again, during dewfall) to 72.2 mm, with a median of 3.5 mm (Table 1). Thirty of the plants in the selected dogfennel clusters – those being monitored for stemflow – had an average canopy radius of 18.3 cm (±4.5 cm standard deviation), which was nearly identical to the median canopy radius (Table 1). The stem radii of all measured dogfennel plants ranged from 0.1 to 0.7 cm, with a mean radius of 0.6 cm (Table 1). The resulting ratio of canopy : stem radii was also normally distributed, with a mean and median of ∼36 (dimensionless), but ranged from 24 to 50 (Table 1). For all plants, the mean leaf angle decreased from 54 to 32∘ from vertical with increasing canopy height, i.e., the higher in the dogfennel canopy, the closer the leaf angle was to vertical (Table 1). This trend appears consistent across each individual study plant regardless of which clump the plants' were located in, as the standard deviation across all elevations are low, 1.8–3.1∘ from vertical, and do not overlap (Table 1).

Note that PT,o′ is an event-scale estimate derived from past observations, limiting its utility in examining fine-scale PT and individual-plant scale PS. The sum of data from all storms throughout the study period resulted in PT, PS and I of 71 %, 8 % and 21 % as a portion of PT,o′, respectively, beneath dogfennel plants at our site. Water storage capacity achieved by dogfennel leaves in the lab was 0.90±0.04 mm, whereas dogfennel stems stored a capacity of 0.43±0.02 mm (Fig. 4). This resulted in the total SU of dogfennel plants in the understory of this study site being approximately 1.3 mm. This SU estimate agrees with the reductions of PT,o′ below dogfennel plants; for example, mean PT:PT,o′ was 76.6 % for rain-only storms (Table 2), or a mean yield of PT=12.9 mm which exceeds a 1.3 mm reduction (due to SU and evaporation) in the estimated mean PT,o′ yield, 16.5 mm (from Table 1). A large portion of the rainwater captured on dogfennel canopies was able to overcome the stem water storage capacity and generate PS. Dogfennel PS data were highly skewed, producing a mean relative PS (PS:PT,o′) of 36.8 % but a median of 7.6 % within a narrow interquartile range, 2.8 %–27.2 % (Table 2). For events including occult precipitation, both maximum PS:PT,o′ and PT:PT,o′ exceeded 100 %: PT:PT,o′ during mixed storms reached a maximum at 192 %, whereas the maximum for PS:PT,o′ was just over 900 % (Table 2). Note that dew in the understory was not measured by the above-canopy rainfall gauges, and PT,o′ was only increased by an assumed maximum dew contribution equal to SU (1.33 mm); thus, dew accumulation allows PT and PS to exceed 100 % of Pg and PT,o′ (Table 2). When compared to rainfall above the overstory (Pg), the medians are much smaller: PT:Pg values are 45 % and 58 % for rain-only storms and mixed storms, respectively, and PS:Pg values are 4.1 % and 14.7 %, respectively (Table 2).

Yield values (mm) were estimated for dogfennel PT and PS across storms, and both event-level PT and PS yields linearly correlated with estimated event-level PT,o′ (Fig. 5a, b). Regarding PT, as the catchment area (canopy area above the gauge) is equal to the input area (soil area below the gauge), the PT yield from the canopy and the PT supply to the surface are equal; therefore, the term “yield” will be applied for both. Median PT yield beneath dogfennel for the measured storms was 4.4 mm with an interquartile range of 1.1 to 11.3 mm (Fig. 5c). The maximum PT yield approached 50 mm during a large-magnitude rain storm (where Pg=101.3 mm). As the canopy area that generates stemflow is many times greater than the surface area around plant stems that receive stemflow (see Table 1), PS yield and F will differ. F values are typically used to represent PS supply to soils. Yields of PS from dogfennel were as high as 24 mm, but the median was 0.4 mm and the interquartile range was narrow, 0.1–1.3 mm (Fig. 5c).

Coefficients of variability (CV) and quartile variability (CQV) were computed for both PS and PT, relative to Pg and PT,o′ (Table 2), and storm-normalized temporal stability plots were generated for PS yield only (Fig. 6). Storm-normalized temporal stability plots were not generated for PT yields because the experimental design accounts for the spatial variability of PT via the deployment of large gauge areas (compared with the dogfennel canopy area); this permits estimates of variability across a few large-area gauges (Table 2), but it limits the observable variability. CV and CQV for relative PT ranged from 22 % to 90 % and were generally lower for rain-only storms, <40 %, than for mixed storms, >60 % (Table 2). Variability in relative PS across study plants, ranging from 77 % to 257 %, was always greater than observed for relative PT for the monitored storms (Table 2). Due to the greater skew in the relative PS data compared with relative PT, the CV was many times greater than CQV for relative PS (Table 2). CV and CQV for PS:PT,o′ was similar for rain and the mixed storms; however, the CV for PS:Pg was greater for rain-only storms compared with mixed storms.

Temporal stability of normalized stemflow, P‾S,i (Fig. 6) indicates that there were only a few plants that captured most of the PT,o drained as stemflow (three plants' mean P‾S,i≫1). Thus, most of the studied dogfennel plants captured similar amounts of PT,o′ as stemflow – having P‾S,i between -1 and 1 (y=0 represents the central tendency of P‾S,i data). Funneling ratios (F based on PT,o′) show that all plants concentrated PS yields to the surface around their stem bases (Fig. 6). Mean F across all plants was 87, and for the 27 plants whose mean P‾S,i fell between -1 and 1, median F ranged from 18 to 200 (Fig. 6). However, for the three plants with the highest P‾S,i, their mean F values were 287, 476 and 484 (Fig. 6). These voluminous stemflow-generating plants single-handedly account for one-third of total PS volume (8734 from 27 870 mL). To evaluate possible canopy structural influences on PS variability, various directly measured structural metrics were compared: radii of canopies and stems and the vertical variability in leaf angle (see Fig. S4). No clear visible or statistical correlations or correspondences were found between these structural variables and P‾S,i across plants (Fig. S4). In fact, variability in the measured canopy structural variables was low (Table 1) compared with the variability observed for dogfennel PS and P‾S,i (Fig. 6).

Partitioning of overstory throughfall by this example dominant understory and pasture forb resulted in hydrologically relevant losses of rainwater to the surface at our site (Table 2). As the maximum water storage capacity is a major driver of rainfall interception (Klaassen et al., 1998), the magnitude of dogfennel's overstory throughfall interception may be attributed to its canopy being able to store a sizable magnitude of rainwater per unit area, 1.33 mm (Fig. 4). Although mass changes of dried and submerged vegetation samples are discrepant from the processes and temporal scales of natural rainfall interception, it is a common method with well-known and long-discussed limitations that was selected to estimate water storage capacity as more direct water storage capacity estimation methods are still currently under development – see discussions in reviews by Friesen et al. (2015) and Klamerus-Iwan et al. (2020). Methodological limitations withstanding, the SU estimates in this study fit within the range of water storage capacities of other herbaceous plants synthesized by Breuer et al. (2003). This synthesis is focused on the leaves of herbaceous plants (alongside other plant types) (Breuer et al., 2003), but less research has estimated the stem component (or reported a total including the stem component) of the water storage capacity for short vegetation (Bradley et al., 2003; Wang et al., 2016; Wohlfahrt et al., 2006; Yu et al., 2012). The stems of herbaceous plants, even thick smooth stems (>1 cm in diameter) can store nearly 0.5 mm, e.g., Taraxacum officinale (dandelion) (Wohlfahrt et al., 2006). Even thin (<1 cm radius) herbaceous stems with epidermal outgrowths, like hairs, can store large amounts of rainwater, e.g., 0.25 mm for Achillea millefolium (yarrow) and 0.20 mm for Trifolium pretense (red clover) (Wohlfahrt et al., 2006). In the case of dogfennel stem water storage capacity at our site, the 0.43 mm estimate is within this range, and its magnitude is likely a result of two principal factors: (1) dense stem coverage by desiccated leaves (photo in Fig. 4) and (2) the fact that this species can achieve large densities, up to 700 000 stems ha-1 (Dias et al., 2018) – 56 770 stems ha-1 at our study site. We note that, to our knowledge, stem water storage capacities for herbaceous plants with spines, thorns and other such physical structures have not been evaluated.

Overstory throughfall was also redistributed into a highly spatially variable (Table 2) but temporally persistent pattern beneath dogfennel canopies (where CV or CQV was approximately 20 %–40 % for PT and 80 %–250 % for PS; Table 2), despite all measured canopy structures – such as branch angle, stem size and canopy size – being similar (Table 1). As our sampling plan measured PT over a large area of the dogfennel canopy (rather than at numerous localized points), this discussion point will focus on the intraspecific PS observations. The high spatial variability and temporal persistence of PS across plants despite canopy structural similarity raises the following question: what caused the intraspecific PS patterns observed in this study? A likely explanation may be that, in this case, access to precipitation for stemflow production is related to overstory throughfall patterns (which, we reiterate, were not able to be measured without removing or disrupting PT and PS). Overstory throughfall patterns are known to be spatially variable but temporally persistent across forest types (Van Stan et al., 2020). Specifically, individual dogfennel plants that persistently generated greater PS than other plants may have just received greater overstory throughfall from persistent overstory drip points. If the overstory throughfall pattern is a major driver of intraspecific variability in PS in this study, then the funneling ratios computed from mean overstory throughfall (per Fig. 3) would be incorrect (in Fig. 6). In this case, funneling ratios (computed from the localized overstory throughfall above each plant) could be similar across the monitored dogfennel plants. Testing this hypothesized relationship between dogfennel PS patterns and overstory throughfall patterns was not possible in the field, as sampling overstory throughfall would prevent PS from being generated by the plant. Future work to test this hypothesis could, however, make use of rainfall simulators.

The large diversion of rainwater and dew to their stem base may be partially responsible for dogfennel survival during extended periods of drought (or improved invasion efficacy during droughts; Loveless, 1959; Forthman, 1973), and may also explain why this species tends to be one of the most problematic in improved grazing systems located in Florida (Sellers et al., 2009). Rainfall patterns in central and south Florida may also intersect with dogfennel's canopy water balance to “tip the scales” in its favor. Specifically, rainfall in our study region is often limited from January through May, with the bulk of the rainfall occurring from June through October, and the water storage capacity of burgeoning dogfennel plants during early spring may enhance the chances of individual plant survival (resulting in large infestations as referenced previously).

The dominant understory plant at our study site, dogfennel, intercepted similar amounts of modeled overstory throughfall, with an interquartile range of 11 %–59 % per storm (Table 2), compared to the gross rainfall interception by their overstory pine canopy, which had an interquartile range of 19 %–60 % per storm (Van Stan et al., 2017b). Similar rainwater interception between dogfennel and the pine overstory may be due to dogfennel's maximum water storage capacity comparing favorably to that of overstory tree species, 0.07–4.30 mm (Klamerus-Iwan et al., 2020). Even the maximum stem water storage capacity is of a similar magnitude to values reported by past work on woody plants, 0.2–5.9 mm (Klamerus-Iwan et al., 2020), albeit at the lower end of the range. Most current research on stem water storage has focused on intrinsic factors of woody plant stems, like bark thickness, porosity, microrelief or roughness (Ilek et al., 2017; Levia and Herwitz, 2005; Levia and Wubbena, 2006; Sioma et al., 2018; Van Stan et al., 2016; Van Stan and Levia, 2010); however, other stem structures besides bark may be capable of storing substantial water, e.g., the desiccated leaves of our study plant.

There were differences in how gross rainfall was redistributed by the overstory canopy compared with how modeled overstory throughfall was redistributed by the dogfennel understory. Stemflow from the overstory, P. palustris was negligible at this site, 0.2 % of gross rainfall (Yankine et al., 2017), but median dogfennel PS was 7.6 % of modeled overstory throughfall (with an interquartile range of 2.8 %–27.2 %) (Table 2). Annual relative PS (and PT) estimates from trees and herbaceous plants reported by previous work indicates that herbaceous plants are generally greater stemflow producers than woody plants (Sadeghi et al., 2020). Although relative PT beneath dogfennel was similar to observations of relative overstory throughfall beneath P. palustris at this site (Mesta et al., 2017), throughfall has been found to be generally lower beneath herbaceous plant canopies than for woody plant canopies (Sadeghi et al., 2020). This seems reasonable because if interception is similar between herbaceous plants and woody plants, an increase in relative stemflow would necessitate a decrease in relative throughfall. The results of this study support statements by several past studies suggesting that plants in the understory and overstory interact differently with rainfall. Thus, we repeat the long-standing calls for increased research on understory precipitation partitioning, particularly stemflow (Price et al., 1997; Price and Watters, 1989; Verry and Timmons, 1977; Yarie, 1980).

For a few storms (n=5), dew contributed significantly to PT and PS by the studied dogfennel plants. The median PT generated from dew beneath dogfennel plants at our site was 0.74 mm per plant with an interquartile range of 0.47–0.99 mm per plant, resulting in a total dew-related contribution to PT of 17.1 mm over the study period. Volumes of stemflow under dewfall totaled 558 mL for all study plants, with individuals supplementing the dew-related PT with up to 61 mL per plant (yielding an additional ∼0.6 mm). Dew contributions to net precipitation below plant canopies have rarely been studied. The earliest quantity for dew drainage was 0.08 mm from a single event on a single tree in Johanniskreuz, Germany (Ney, 1893). Since then, to our knowledge, only one other study has examined dew-related drainage from plants, focusing on stemflow from the herbaceous Ambrosia artemisiifolia, or common ragweed (Shure and Lewis, 1973). They estimated that the drainage of dew via PS resulted in an additional input of 1.1 L per month during the growing season and hypothesized that this process may “play a vital role in governing the density, diversity, and distribution of plant species within field ecosystems” (Shure and Lewis, 1973). Dew drainage from plant canopies and down stems may, in addition to being a valuable water source, influence plant–soil interactions by transporting leached or dry-deposited materials to the soils – something also discussed by Shure and Lewis (1973). Globally, dew contributes a small percentage to the annual precipitation (Baier, 1966); however, in semiarid and arid (Baier, 1966; Hao et al., 2012), as well as summer-dry climates (Tuller and Chilton, 1973), dew can form a significant water input. In such ecologic settings as these, it is, therefore, reasonable to suppose that any factor that doubles the frequency of plant-moisture availability, even though the amounts be small, must materially affect the plant growing condition. Thus, further research is needed to assess dew (and mixed storm) drainage in arid and semiarid climates, with days on which dew occurs being ≥70 % yr-1 (Hao et al., 2012). The global importance of occult precipitation and resulting wet canopy conditions has recently been reviewed and described as a critical future research direction for plant sciences (Dawson and Goldsmith, 2018). Given these scant but ecologically relevant findings, further research on the influence of condensation events on plant–soil interactions via throughfall and stemflow may be merited.