Rainfall Partitioning Dynamics in Xerophytic Shrubs: Interplays Between Self-Organization and Meteorological Drivers

Abstract

Rainfall partitioning, a crucial process in shaping the local hydrological cycle, governs canopy interception and subsequent soil water recharge. While canopy structure and meteorological conditions fundamentally regulate this process, the role of plant self-organization and its interactions with meteorological drivers (non-precipitation variables in particular) remain underexplored. To address this gap, we investigated rainfall partitioning components, including the amount, intensity, efficiency, and temporal dynamics of throughfall and stemflow, in clumped and scattered Vitex negundo L. var. heterophylla (Franch.) Rehder shrubs in the Yangjuangou catchment of the Chinese Loess Plateau during the 2021–2022 rainy seasons. Despite comparable net precipitation (clumped: 83.5% vs. scattered: 84.2% of incident rains), divergent rainfall partitioning strategies emerged. Clumped V. negundo exhibited greater stemflow (8.6% vs. 5.2%), characterized by enhanced intensity, efficiency, and favorable temporal dynamics. Conversely, scattered shrubs favored throughfall generation (79.0% vs. 74.9%). Consistent with previous research, rainfall amount was recognized as the primary control on partitioning rains. Furthermore, our integrated analysis, combining machine learning with variance decomposition, highlighted the critical roles of antecedent canopy wetness (4 h pre-event leaf wetness) and wind speed thresholds (e.g., low wind vs. gust) in regulating partitioning efficiency and temporal dynamics. These findings advance the mechanistic understanding of the interplay between plant self-organization and hydrological processes, demonstrating how morphological adaptations in V. negundo optimize water harvesting in semi-arid ecosystems. This addressed the need to incorporate dynamic interplays between plant structure (specifically, self-organized patterns) and meteorological factors (particularly non-precipitation variables) into ecohydrological models, especially for improved predictions in water-limited regions.

1. Introduction

Structure determines function across multiple scales in the universe. Within the critical zone, the interface between the atmosphere and lithosphere, the exchange of material and energy profoundly sculpts Earth’s surface [1], as well as the structures and functions of ecosystems that depend on it. Among these biogeographical processes, water fluxes are particularly critical, as they not only provide one of the most essential resources for living organisms but also drive the dynamics of material and energy flow [2]. Plant canopies, serving as the primary interface between terrestrial ecosystems and rainfall, play a pivotal role in these processes. The structure of the canopy, combined with meteorological conditions [3], regulates the partitioning of rainfall into distinct pathways [4]: interception loss, which evaporates back into the atmosphere; stemflow, which is channeled down trunks and branches; and throughfall, which reaches the ground through canopy gaps, drips, and splashes [5,6]. By altering the amount, timing, and spatial distribution of water input [7], rainfall partitioning is critical for recharging soil moisture [8] and groundwater [9], enhancing nutrient transport and partitioning [10], and the functioning of rainfed dryland ecosystems [11]. Given the typically isolated and aggregated canopies of scattered and clumped individuals, respectively, Yuan et al. [12] highlighted the importance of vegetation self-organization in regulating the quantity (i.e., volume and the proportion of incident rains) and efficiency (i.e., intensity, funneling ratio, stemflow productivity, and temporal dynamics) of rainfall partitioning (especially stemflow). Building on this, An et al. [4] systematically quantified the temporal dynamics of throughfall using the methodology introduced by Yuan et al. [13], focusing on xerophytic shrub species of Caragana korshinskii Kom. and Salix psammophila C. Wang et Ch Y. Yang. However, the amount, intensity, efficiency, and temporal dynamics of throughfall still remain inadequately understood in relation to self-organized patterns. This knowledge gap hinders a comprehensive understanding of how plant morphological adaptations respond to water-stressed environments.

Meteorological conditions strongly affect rainfall partitioning by vegetation. Rainfall amount is widely recognized as a key determinant of the quantity of stemflow and throughfall for both individual plants and plant communities across diverse ecosystems, including rainforests [14], plantations [15], temperate forests [16], boreal forests [17], and xerophytic shrublands [12], as well as in regional- to global-scale analysis [18,19]. Extensive research has employed methods such as linear regression [20,21], univariate polynomial fitting [22,23], power law regression [24,25], and logarithmic modeling [4,22] to quantify rainfall amount’s influence on the partitioning process. However, the quantitative impacts of other key meteorological characteristics (rainfall duration, rainfall intensity, duration, wind speed, etc.) on the process of rainfall partitioning, including its magnitude, timing, and spatial variability, remain poorly understood. For instance, wind speed strongly affects rainfall portioning by altering its volumetric distribution, spatiotemporal patterns, and kinetic energy [26,27]. Increased wind speed reduces throughfall drop size in both deciduous and coniferous trees under foliated conditions [28], while strong winds enhance the dislodgement of intercepted rainwater from canopies [5]. Despite these mechanistic insights, wind effects are rarely analyzed with respect to speed thresholds (e.g., low vs. high wind regimes) or differentiated by their distinct impacts on stemflow and throughfall. This oversight limits our ability to disentangle how meteorological interactions shape ecohydrological processes in water-stressed ecosystems.

In addition, water adsorption by branches and leaves is a primary mechanism of rainfall interception [29], influencing the evaporation rate of the rain-wetted forest [30], the spatial heterogeneity of throughfall [31], and the contact time and volume of stemflow [32]. The resultant hydraulic properties of the canopy, including leaf (or canopy) wetness [33] and bark water storage capacity [34], greatly affect the quantity and dynamics of rainfall partitioning. Pre-storm water retention on leaves [35] and the canopy’s surface state (dry or wet) [36] are critical factors impacting interception capacity and, consequently, canopy water balance. However, previous studies have largely produced qualitative rather than quantitative conclusions due to the challenges in obtaining precise in situ measurements of these dynamic processes.

This study comprehensively quantifies rainfall partitioning for clumped vs. scattered V. negundo shrubs in the Chinese Loess Plateau during the 2021–2022 rainy seasons. Complementing our in-depth analysis of the role of plant traits in determining the amount, intensity, efficiency, and temporal dynamics of stemflow [12], we further investigate the meteorological drivers governing these processes for both throughfall and stemflow, focusing on two primary objectives: (1) to compare rainfall harvesting strategies between contrasting self-organized patterns (i.e., clumped vs. scattered) of xerophytic shrubs, and (2) to identify key meteorological drivers beyond rainfall amount that modulate rainfall partitioning. By addressing these objectives, this study advances the mechanistic understanding of rainfall partitioning dynamics and elucidates how plant morphological adaptations enhance water harvesting in semi-arid ecosystems.

2. Materials and Methods

2.1. Study Sites

This study was conducted at the Yangjuangou catchment (36°42′ N, 109°31′ E) in Yan’an city of Shaanxi Province, the central Loess Plateau of China (Figure 1). It is approximately 2.02 km² with 10°–30° slopes and 1050–1295 m elevation. This area experiences a semi-arid temperate monsoon climate, with an annual precipitation of 537 mm and an average temperature of 10 °C (1961–2016) [37]. Rainfall primarily occurs from June to September (i.e., the rainy season), accounting for 60%–70% of the total annual precipitation [38]. This catchment is geomorphologically characterized by loess ridges and gullies. The dominant soil type is loess soil with a 50% porosity and uniform texture, which is characterized by a fragile structure and poor erosion resistance [39].

Native vegetation has been severely degraded in the Yangjuangou catchment due to intense human activities [40]. The shrub species of V. negundo is among the very few that has survived [12], distributed in clumped and scattered patterns. The leaves are oblong-lanceolate, palmately compound with an entire margin or a few coarse serrations on each side [41]. The upper surface is green while the lower surface is densely covered with grayish-white tomentum. The corolla is pale purple and puberulent externally. Vitex negundo flowers from April to June and fruits from July to October [42]. This species is deciduous and perennial, having multiple branches radiating outward from the canopy center. A study plot of V. negundo, covering 400 m² in size at an elevation of 1183 m, with a 282° aspect and a 17° slope, was established to investigate rainfall partitioning between clustered and scattered shrubs [12].

2.2. Meteorological Recordings

A micro-weather station has been installed in the open area at the V. negundo plot to record rainfall characteristics, wind speed (WS, m·s^–1) and direction (WD, °), gust speed (GS, m·s^–1) (Model 03002, R. M. Young Company, Traverse City, MI, USA), air temperature (AT, °C) and humidity (AH, %) (Model HMP 155, Vaisala, Vantaa, Finland), leaf wetness (Model S-LWA-M003, Onset Computer Corporation, Bourne, MA, USA), and solar radiation (SR, KW·m^–2) (Model CNR 4, Kipp & Zonen B. V, Delft, The Netherlands) during the 2021–2022 rainy seasons. The RG3-M tipping bucket rain gauge (Onset Computer Corporation, Bourne, MA, USA) (hereafter “TBRG”) records the rain amount (RA, mm) and timing of incident rains. Discrete events are defined by an RA of 0.2 mm (the resolution limit of the RG3-M rain gauge) and the smallest 4 h gap without rain (the analog period of time to dry canopies from antecedent rains) [20]. Rainfall amount, duration (RD, h), the interval between neighboring events (RI, h), the average rainfall intensity (I, mm·h^–1), and maximum rainfall intensity in 10 min (I₁₀, mm·h^–1) were calculated accordingly. To address the inherent underestimation of TBRG due to the systematic errors of missing the records of inflow during tipping intervals [35], three standard gauges (20 cm diameter) were placed nearby to calibrate the recordings via the linear regression model (R = R_TBRG × 1.32 + 0.16, R² = 0.98, p < 0.01) between automatic recordings (R_TBRG, mm) and manual measurements (R, mm) developed by Yuan et al. [12,13]. Canopy wetness was represented as the average value detected by sensor S-LWA-M003 during the four hours prior to rains (LW₄, %). We computed vapor pressure deficit (VPD, kPa) to quantify the potential evaporation (Equations (1) and (2)) [43].

$$e_s=0.611\times exp\left(17.27\times AT/\left(237.7+AT\right)\right),$$

(1)

$$VPD=e_s\times \left(1-AH\right),$$

(2)

where e_s is the saturated vapor pressure (kPa).

2.3. Plant Trait Measurements

The canopies of clumped and scattered V. negundo that are selected for rainfall partitioning measurements do not overlap with neighboring shrubs. We measured their canopy height (CH, m) at the canopy center and calculated the canopy area (CA, m²) as an ellipse by measuring the east–west and north–south diameters of the canopy. The leaf area index (LAI, m²·m⁻²) of scattered and clumped V. negundo was measured twice a month using the vegetation canopy analyzer (Model Licor 2000, LiCor Environmental, Lincoln, NB, USA). Morphological characteristics were measured at the 224 and 59 branches of clumped and scattered V. negundo, respectively, that were selected for rainfall partitioning measurement, including basal diameter (BD, mm), branch length (BL, cm), and branch angle (BA, °).

According to the allometric equations developed by Yuan et al. [12], we calculated the total leaf area of individual branches (TLA = 15.281 × BD^1.120, cm²) for clumped and scattered V. negundo, and the biomass of leaves (BML_C = 0.669 × BD^1.520, BML_S = 0.116 × BD^2.069; g) and stems (BMS_C = 1.742 × BD^1.795, BMS_S = 0.283 × BD2.430; g) for clumped and scattered V. negundo, respectively. The specific leaf area (SLA, cm²·g^–1) is calculated as the ratio of TLA to leaf biomass. We estimate the stem surface area of an individual branch in the form of a cylinder (SA = π × BD × BL/20, cm²).

Six scattered V. negundo were selected for stemflow and throughfall measurements, which have similar canopy area (5.6 ± 1.7 m²) and height (2.6 ± 0.5 m). In addition, they had a similar total canopy area (33.8 m²), average canopy height, and LAI (3.02 m²·m⁻²) with clumped V. negundo (31.5 m², 2.2 ± 0.2 m, 3.30 m²·m⁻²) (

Table 1. Plant traits of clumped and scattered V. negundo for rainfall partitioning measurements.

Self-Organized Patterns	Total Canopy Areas (m2)	Average Canopy Height (m)	BD Categories (mm)	Branch Amount	Average BD (mm)	Average BL (cm)	Average BA (°)	Average TLA (cm2)
Clumped	31.5	2.2 ± 0.2	<10	120	6.3 ± 2.0	131.9 ± 49.0	52.6 ± 18.0	846.7 ± 522.5
			10–15	69	12.1 ± 1.4	202.9 ± 72.6	52.6 ± 16.3	3086.6 ± 794.7
			15–20	28	17.7 ± 1.3	225.6 ± 42.3	52.5 ± 12.2	6867.0 ± 1043.6
			>20	7	21.9 ± 1.0	245.2 ± 69.9	51.6 ± 13.1	10,654.8 ± 988.5
Scattered	33.8	2.6 ± 0.5	<10	13	7.4 ± 1.7	158.2 ± 49.3	61.7 ± 15.8	1130.8 ± 460.8
			10–15	13	13.2 ± 1.4	237.6 ± 30.2	50.6 ± 12.4	3665.8 ± 789.2
			15–20	12	17.2 ± 1.7	275.2 ± 44.6	58.4 ± 13.3	6444.7 ± 1378.2
			>20	21	22.5 ± 2.5	266.8 ± 69.0	53.3 ± 20.1	11,425.2 ± 2957.0

Note: BD, BL, and BA are the branch basal diameter, length, and angle. TLA refers to the total leaf area of individual branches.

), thus allowing an effective comparison on rainfall partitioning between differently self-organized V. negundo. For branch morphologies, scattered shrubs had more inclined (63.2° vs. 58.0°) (p > 0.05), significantly larger (16.0 vs. 13.3 mm) (p < 0.05), and longer (260.7 vs. 210.5 cm) (p < 0.05) branches than clumped V. negundo.

2.4. Rainfall Partitioning Quantifications

For throughfall measurement, we labeled the V. negundo canopies into four quadrants, such as northeast, southeast, southwest, and northwest. A TBRG had been placed at the center of each quadrant (Figure 1), with another one and two gauges being randomly placed in each quadrant of scattered and clumped V. negundo, respectively. Throughfall was measured within two hours after the rain ended. If the rain ended in the evening, measurements were conducted on the following morning. By the timely measurements at both fixed and random locations, an accurate quantification of throughfall could thus be guaranteed.

For stemflow measurements, we grouped the experimental branches at clumped and scattered V. negundo into four BD categories, such as the 5–10 mm, 10–15 mm, 15–20 mm, and >20 mm. Seven branches were selected in each BD category to quantify the amount and temporal dynamics of stemflow. Aluminum foil collars were installed to trap stemflow, which were fitted around the branch circumference and sealed with neutral silicone caulking (Figure 1). The PVC hoses with 1 cm diameter channeled stemflow to the covered containers for manual measurements via graduated cylinder or to TBRGs for automatical recording. We returned stemflow to the branch base after measurements and returned throughfall to the soil beneath the gauges, thus minimizing the experiment-introduced drought stress on the measured shrub individuals. We regularly checked these apparatuses against any blockages, aging, or leakage.

We calculated the throughfall amount (TF_d, mm) based on the canopy area for individual shrubs and quantified the stemflow amount as the volume production of individual branches (SF_b, mL) and the equivalent water depth of stemflow (SF_d, mm), as well as the proportion of stemflow (SF%) and throughfall (TF%) to incident rainfall amount (Equations (3)–(5)). Net precipitation depth (NP_d, mm) and percentage (NP%, %) are calculated as the sum of SF_d and TF_d (i.e., NP_d = SF_d + TF_d) and SF% and TF% (i.e., NP% = SF% + TF%), respectively. The average and the 10 min maximum intensities of throughfall (i.e., TFI and TFI₁₀, mm·h^–1) and stemflow (i.e., SFI and SFI₁₀, mm·h^–1) were quantified on the basis of TBRG recordings [13].

$$TF\%=\left(T{F}_d∕RA\right)\times 100\%,$$

(3)

$${SF}_d=10\times {\Sigma }_{i=1}^{n}S{F}_{bi}∕CA,$$

(4)

$$SF\%=\left(S{F}_d∕RA\right)\times 100\%,$$

(5)

$$TFI={TF}_d/\mathrm{R}\mathrm{D},$$

(6)

$$TF{I}_{10}=6000\times TF{V}_{10}/S_{RG},$$

(7)

$$SFI=1000\times \mathrm{S}\mathrm{F}\mathrm{V}/(\mathrm{B}\mathrm{B}\mathrm{A}\times \mathrm{R}\mathrm{D}),$$

(8)

$$SF{I}_{10}=6000\times SF{V}_{10}/BBA,$$

(9)

where SF_bi refers to the stemflow volume of branch i (mL); n is the total number of branches; CA is the canopy area (cm²); RA is rainfall amount (mm); S_RG is the base area of rain gauges; SFV and SFV₁₀ (mL) are the total volume and the 10 min maximum volume of stemflow at individual branches within an incident rain; TFV₁₀ (mL) is the 10 min maximum volume of throughfall; and BBA is branch basal area (mm²).

Stemflow efficiency has been calculated using a variant of the funneling ratio (FR, unitless) introduced by Herwitz [44] (Equation (10)). The efficiencies of rainfall partitioning are further quantified as the productivities of stemflow (SFP, mm·kg⁻¹) [20] and throughfall (TFP, mm·kg⁻¹) (Equations (11) and (12)), representing the plot-scaled stemflow depth (SF_d, mm) of unit aboveground biomass (AGB, kg) and throughfall depth (TF_d, mm) of unit leaf biomass (BML, kg), respectively. Rainfall partitioning productivities describe the efficiencies of collecting water in the forms of stemflow and throughfall with unit biomass investment [20]. SFP and TFP are conducive to explaining the plant’s drought tolerance by quantifying the ability to convert water resources into carbon accumulation and growth [45]. The rainfall partitioning dynamics have been systematically quantified for clumped and scattered V. negundo [13]. They described the time lags to generate, maximize, and end throughfall (TLG_TF, TLM_TF, and TLE_TF, min) and stemflow (TLG_SF, TLM_SF, and TLE_SF, min) in response to rains and their duration (TFD and SFD, h) within incident rains.

$$FR=10\times SF\mathrm{V}/(\mathrm{R}\mathrm{A}\times \mathrm{B}\mathrm{B}\mathrm{A}),$$

(10)

$$SFP={SF}_d/AGB,$$

(11)

$$TFP={TF}_d/BML$$

(12)

2.5. Data Analysis

Boruta algorithm method, a wrapper approach constructed based on the random forest algorithm, is characterized by a good performance in handling complex relationships, including non-linearities and interactions, and robustness to noise and outliers. This approach has been performed to evaluate the effects of meteorological characteristics (i.e., RA, RD, RI, I, I₁₀, WS, GS, LW₄, and VPD) affecting throughfall and stemflow. This approach, by setting the p-value of 0.05, the maxRun parameter of 600, and the doTrace parameter of 2, computes the Z-scores and effectively identifies the most influential indicators on the quantities and efficiencies of rainfall partitioning [46]. These indicators are then fed into the multiple regression models by using the “MuMIn” (v1.48.4) package [47]; thus, fitting an optimal model in terms of the Akaike information criterion with a maximum likelihood in R. Their contributions are quantified by performing the variance decomposition with the “relaimpo” (v2.2.7) package in R. We use a Kruskal–Wallis rank sum test to examine the significant differences in rainfall partitioning amount (i.e., TF_d, SF_d, and NP_d; mm), intensities (i.e., TFI, TFI₁₀, SFI, and SFI₁₀; mm·h^–1), and temporal dynamics (i.e., TLG_TF, TLM_TF, TLE_TF, TLG_SF, TLM_SF, and TLE_SF; min; TFD and SFD; h) between clumped and scattered V. negundo, since those data are not normally distributed in terms of the Shapiro–Wilk test (p < 0.05).

Statistical analyses and graphing in this study have been conducted within the R statistical framework (version 4.4.2, R Core Team, 2024, Vienna, Austria) and OriginPro 2022 (OriginLab Corporation, Northampton, MA, USA). The R packages applied in this study are listed as follows: “Boruta” (v8.0.0) [46], “ggplot2”(v3.5.1) [48], “MuMIn” (v1.48.4) [47], “ggthemes” (v4.2.0) [49], “relaimpo”(v2.2.7) [50], and “tidyverse” (v2.0.0) [51]. The average values are expressed as the mean ± standard deviation.

3. Results

3.1. Meteorological Conditions

This study analyzed 33 rainfall events (12 in 2021, 21 in 2022) in the Yangjuangou catchment during the 2021 and 2022 rainy seasons. Key event characteristics include the following mean ± standard deviation values: RA = 29.3 ± 36.7 mm, RD = 8.7 ± 8.9 h, RI = 98.2 ± 99.3 h, I = 4.9 ± 8.6 mm·h^–1, I₁₀ = 26.7 ± 30.3 mm·h^–1, VPD = 0.4 ± 0.5 kPa, LW₄ = 79.2% ± 25.4, WS = 0.7 ± 0.7 m·s^–1, and GS = 3.1 ± 1.9 m·s^–1 (Figure 2). Events were categorized by RA into four groups: <10 mm (11 events, total RA = 64.4 mm), 10–20 mm (7 events, 93.6 mm), 20–30 mm (6 events, 146.6 mm), and >30 mm (9 events, 535.0 mm). Rainfall intensity distributions included <2 mm h^–1 (21 events, mean = 1.1 mm·h^–1), 2–5 mm h^–1 (6 events, 3.8 mm·h^–1), and >5 mm h^–1 (6 events, 19.5 mm·h^–1). Rainfall duration categories comprised <5 h (14 events), 5–10 h (11 events), and >10 h (8 events). In general, the data revealed a predominance of low-magnitude, short-duration rainfall events with moderate intensity in the study area.

3.2. Quantity and Efficiency of Rainfall Partitioning

Clumped and scattered V. negundo exhibited comparable net precipitation, as quantified by both depth (NP_d: 25.7 ± 33.3 mm vs. 25.4 ± 32.6 mm, p > 0.05) and proportion of incident rainfall (NP%: 83.5% ± 11.2% vs. 84.2% ± 8.2%, p > 0.05) (

Table 2. Comparisons on rainfall partitioning between differently self-organized Vitex negundo.

Rainfall Partitioning Components	Indicators	Clumped V. negundo	Scattered V. negundo
Net precipitation	Depth (NPd, mm)	25.7 ± 33.3 a	25.4 ± 32.6 a
	Percentage (NP%, %)	83.5 ± 11.2 a	84.2 ± 8.2 a
Stemflow	Depth (SFd, mm)	2.7 ± 3.6 a	1.7 ± 2.4 b
	Percentage (SF%, %)	8.6 ± 2.0 a	4.7 ± 1.2 b
	Average intensity (SFI, mm·h–1)	651.0 ± 962.1 a	346.9 ± 649.1 b
	10 min maximum intensity (SFI10, mm·h–1)	2386.6 ± 2990.3 a	2623.9 ± 3135.8 a
	Funneling ratio (FR, unitless)	154.4 ± 51.7 a	134.8 ± 46.1 b
	Productivity (SFP, mm·kg–1)	3.3 ± 4.8 a	2.2 ± 2.1 b
	Time lag for generation (TLGSF, min)	27.2 ± 29.8 a	23.8 ± 26.1 a
	Time lag for maximization (TLMSF, min)	119.3 ± 221.9 a	92.3 ± 161.0 a
	Time lag for ending (TLESF, min)	161.0 ± 137.0 a	137.0 ± 112.6 a
	Duration (SFD, h)	11.9 ± 8.0 a	11.6 ± 8.0 a
Throughfall	Depth (TFd, mm)	23.0 ± 29.3 a	23.8 ± 29.8 a
	Percentage (TF%, %)	74.9 ± 10.2 a	79.0 ± 7.9 b
	Average intensity (TFI, mm·h–1)	5.0 ± 8.0 a	5.8 ± 10.5 a
	10 min maximum intensity (TFI10, mm·h–1)	20.3 ± 19.2 a	16.4 ± 16.3 b
	Productivity (TFP, mm·kg–1)	6.1 ± 7.6 a	9.4 ± 12.2 b
	Time lag for generation (TLGTF, min)	17.8 ± 18.0 a	17.6 ± 23.7 a
	Time lag for maximization (TLMTF, min)	61.5 ± 101.2 a	138.4 ± 142.5 b
	Time lag for ending (TLETF, min)	102.2 ± 218.1 a	216.1 ± 270.7 b
	Duration (TFD, h)	11.1 ± 14.7 a	14.2 ± 9.4 b

Note: Different letters indicate significant differences in rainfall partitioning between differently self-organized V. negundo (p < 0.05). The presented values are means ± standard deviation.

). This indicated that the total amount of water transferred beneath canopies, available for soil moisture recharging, did not differ significantly between the two self-organized shrub patterns. However, marked differences were observed in the ways of partitioning the intercepted rains. Clumped V. negundo yielded more stemflow (SF_d: 2.7 ± 3.6 vs. 1.7 ± 2.4, mm; SF%: 8.6% ± 2.0% vs. 4.7% ± 1.2%) (p < 0.05), while exhibiting significantly reduced throughfall proportions (TF%: 74.9% ± 10.2% vs. 79.0% ± 7.9%, p < 0.05) than those of scattered V. negundo. Despite these proportional differences, throughfall depth remained non-significantly different between them (TF_d: 23.0 ± 29.3 mm vs. 23.8 ± 29.8 mm, p > 0.05) (Table 2). These contrasting patterns underscored how stand structure modulates rainfall partitioning.

Clumped V. negundo exhibited greater stemflow collection efficiency. Specifically, clumped plants demonstrated significantly higher stemflow intensity (SFI: 651.0 ± 962.1 vs. 346.9 ± 649.1, mm·h^–1; p < 0.05), stemflow production per unit biomass (SFP: 3.3 ± 4.8 vs. 2.2 ± 2.1, mm·kg^–1; p < 0.05), and funneling ratio (FR: 154.9 ± 51.7 vs. 142.5 ± 46.1; p < 0.05), though peak 10 min stemflow intensity (SFI₁₀: 2386.6 ± 2990.3 vs. 2623.9 ± 3135.8, mm·h^–1; p > 0.05) showed no significant difference (Table 2). On the contrary, scattered V. negundo displayed superior throughfall collection efficiency with lower TFI (5.0 ± 8.2 vs. 5.8 ± 8.7, mm·h^–1; p > 0.05) and reduced TFP (6.3 ± 8.1 vs. 9.7 ± 12.4, mm·kg^–1; p < 0.05), but significantly higher TFI₁₀ (20.3 ± 19.5 vs. 16.0 ± 17.1, mm·h^–1; p < 0.05).

Distinct temporal dynamics of rainfall partitioning were observed between varied self-organized shrubs. Clumped V. negundo required a longer time to initiate (TLG_SF: 27.2 ± 29.8 vs. 23.8 ± 26.1, min; p > 0.05), peak (TLM_SF: 119.3 ± 221.9 vs. 92.3 ± 161.0, min; p > 0.05), cease (TLE_SF: 161.0 ± 137.0 vs. 137.0 ± 112.6, min; p > 0.05), and produce stemflow (SFD: 11.9 ± 8.0 vs. 11.6 ± 8.0, h; p > 0.05) than scattered V. negundo (Table 2). In contrast, clumped shrubs demonstrated significantly faster throughfall dynamics, reaching peak intensity earlier (TLM_TF: 61.5 ± 101.2 vs. 138.4 ± 142.5, min; p < 0.05), ceasing sooner (TLE_TF: 102.2 ± 218.1 vs. 216.1 ± 270.7, min; p < 0.05), and producing throughfall (TFD: 11.1 ± 14.7 vs. 14.2 ± 9.4, h; p < 0.05).

3.3. Meteorological Influences on Rainfall Partitioning

By comparing the importance of the original features and the randomly generated shadow features, the Boruta algorithm addressed the significant influences of RA, RD, LW₄, I₁₀, and I on the amount (e.g., SF_d and TF_d) and productivity (e.g., SFP and TFP) of rainfall partitioning of clumped and scattered V. negundo (Figure 3). The dominant influence of RA had further been examined by the variance decomposition analysis with >60% contribution, followed by RD (~15%), I₁₀ (~10%), I (<5%), and LW₄ (<5%) (Figure 4). FR was significantly influenced by I (60.5%) and I₁₀ (39.5%) at clumped V. negundo, but by RD (69.6%, p < 0.05), VPD (21.7%), and I (8.7%) at scattered V. negundo (Figure 3 and Figure 4).

I, I₁₀, and RD addressed by the Boruta algorithm significantly affected the SFI and TFI of clumped and scattered V. negundo (Figure 3). The variance decomposition analysis further identified I (p < 0.05) as the most influential indicator with ~60% contribution, followed by I₁₀ (~30%) and RD (~5%) (Figure 4 and Figure 5). Additionally, I₁₀, I, and RA were highlighted for significantly affecting TFI₁₀ of clumped and scattered V. negundo (Figure 3). The dominant influence of I₁₀ had been finally located with ~50% contributions, followed by I (~30%) and RA (~20%) (Figure 4 and Figure 5).

With regard to stemflow temporal dynamics of clumped and scattered V. negundo, I₁₀ and GS, examined by the Boruta algorithm, were the only meteorological characteristics having significant impacts on TLG_SF and TLE_SF (Figure 3). TLM_SF was significantly affected by RD (with contributions of 47.6% and 87.7%, p < 0.05), followed by RA (46.9% and 10.7%, p < 0.05), VPD (2.6% and 1.1%, p > 0.05), and LW₄ (2.9% and 0.6%, p > 0.05) (Figure 3 and Figure 6). SFD was majorly explained by RD (with contributions of 70.3% and 63.3%, p < 0.05), followed by RA (21.9% and 24.2%, p > 0.05), I (4.7% and 8.2%, p > 0.05), VPD (2.0% and 3.6%, p > 0.05), and LW₄ (1.0% and 0.8%, p > 0.05) (Figure 6). While for throughfall temporal dynamics, TLG_TF was only influenced by RI for clumped V. negundo and by WS for scattered V. negundo (Figure 3). TLE_TF was significantly affected by LW₄ at clumped V. negundo (p < 0.05), but determined by the interplay of RD (with a contribution of 42.3%, p > 0.05), GS (23.7%, p > 0.05), LW₄ (18.3%, p > 0.05), VPD (8.9%, p > 0.05), and RA (6.8%, p > 0.05) (Figure 6). TLM_TF was subject to the combined influence of RD (with a contribution of 67.6%, p < 0.05), LW₄ (19.9%, p > 0.05), and VPD (12.2%, p > 0.05) at clumped V. negundo, but significantly affected by VPD at scattered V. negundo (p < 0.05) (Figure 6). As for TFD, RD (with a contribution of 63.5%, p < 0.05), VPD (15.5%, p > 0.05), RA (12.1%, p > 0.05), LW₄ (5.1%, p < 0.05), and I (3.7%, p > 0.05) had a combined effect at clumped V. negundo. Whereas the combined influence of RD (with a contribution of 60.2%, p < 0.05), LW₄ (20.0%, p < 0.05), RA (15.6%, p > 0.05), and VPD (4.2%, p > 0.05) had been noted at scattered V. negundo (Figure 6).

4. Discussion

4.1. Distinct Rainfall Harvesting Strategies Explained by Plant Traits

Rainfall partitioning determines the amount of net precipitation, thus strongly affecting soil water recharge [52,53], even the primary production in drylands [54]. In this study, the amount of net precipitation of differently self-organized V. negundo (averagely 25.7 mm and 87.7% of incident rain) agrees with findings of previous studies on xerophytic shrub species, such as S. psammophila (85.9%) [4], Spiraea pubescens Turcz. (80.8%) [55], C. korshinskii (86.4%) [23], Larrea divaricate Cav. (90.6%) [56], Haloxylon ammodendron (C. A. May.) Bunge (77.7%) [25], Amygdalus pedunculata Pall. (79.2%), and Amorpha fruticosa L. (83.1%) [57]. Although clumped and scattered V. negundo harvested similar net precipitation (86.8% and 84.2%), they display distinct patterns of partitioning rain. Compared to scattered V. negundo, clumped V. negundo significantly collected more stemflow (SF%: 8.6% vs. 5.2%; p < 0.05) in a significantly more efficient manner (FR: 154.8 vs. 134.8, unitless; SFP: 3.3 vs. 2.2, mm·kg⁻¹; p < 0.05), but less throughfall (TF%: 74.9% vs. 79.0%; p > 0.05) with less efficiency (TFP: 6.1 vs. 9.4, mm·kg⁻¹; p < 0.05). The differences lead to distinct responses of soil moisture in terms of the amount and timing [12]. Being the representative self-organized pattern at the plot scale [58], it can be inferred that clumped and scattered V. negundo possess distinct strategies for harvesting rains via canopy interception. This quantitative advantage, i.e., larger net precipitation and throughfall yields, likely contributes to a larger soil water budget in scattered V. negundo [59]. Conversely, the efficiency advantage, i.e., greater stemflow yield and collection efficiency, potentially facilitates clumped V. negundo with preferential paths to channel intercepted rains to the root zone, thereby minimizing the evaporation loss and enhancing soil water availability [12].

Plant traits partly explain the distinct rainfall harvesting strategies between differently self-organized shrubs. Compared to scattered V. negundo, clumped V. negundo exhibited a larger total leaf area (TLA: 58.1 vs. 38.0, m²) and bigger leaf biomass (BML: 3648.1 vs. 2439.6, g), which led to greater interception loss but lower throughfall by the aggregated canopies [60,61]. Despite the smaller stemflow volume of individual branches (372.6 vs. 579.8 mL) at clumped V. negundo, their much denser branches (6.9 vs. 1.7 individual branches per·m⁻²) yielded a greater amount of plot-based stemflow (2.7 mm vs. 1.7 mm; 8.6% vs. 4.7%) than those of scattered V. negundo. This is consistent with the positive relation between stemflow and forest density in Japanese coniferous forests [62] and Aleppo pine plantations in Spain [9]. On the contrary, scattered V. negundo had a sparser canopy (e.g., LAI: 3.02 vs. 3.30, m²·m⁻²; AGB: 20.6 vs. 34.6, kg) and consequently less shading from neighboring shrubs, which facilitates throughfall production [12] but reduces stemflow yield.

Structure determines function. Canopy structure has been repeatedly addressed for significantly affecting the function of rainfall harvest of woody plants. Yuan et al. [20] identified a beneficial branch architecture for stemflow yield of S. psammophila, including the relatively smaller-sized branches, larger leaf biomass, and bigger branch angles. Magliano et al. [56] found stemflow of L. divaricate increased with branch angle but decreased in plants that had a large canopy area. Tree size (e.g., the diameter of the tree trunk or shrub branch) was frequently highlighted as a dominant indicator controlling stemflow production [19,55,63,64,65,66]. Therefore, plant traits well explain the distinct strategies of partitioning rains between clumped and scattered V. negundo. In addition, meteorological conditions greatly affected rainfall partitioning [27]. The plants with unique structures, such as distinct self-organizations, might respond differently to the same meteorological characteristics. Yet, the meteorological impacts have not been clearly understood.

4.2. Meteorological Impacts on Rainfall Partitioning Strategy

Rainfall amount determines the quantity of rainfall partitioning. In this study, the Boruta algorithm and the variance decomposition analysis addressed the dominant influence of rainfall amount, having greater than 60% contributions, on the yield of stemflow and throughfall at clumped and scattered V. negundo. This is consistent with the majority of studies that have focused on the meteorological influences. They generally highlighted the dominant influence of rainfall amount on stemflow or throughfall yield of tropical trees [14], temperate trees [67], boreal trees [17], plantations [15], xerophytic shrublands [12], etc. Significant relations between rainfall intensity and stemflow intensity at C. korshinskii, such as I and SFI, I₁₀ and SFI₁₀, have been observed in our previous study [13]. The significant influence of rainfall intensity (I and I₁₀) has been verified on stemflow (SFI and SFI₁₀) of V. negundo with distinct self-organizations, as well as on those of throughfall (TFI and TFI₁₀). Therefore, the important but different influences of rainfall amount and intensity could be concluded, given that they significantly affect the event- and process-based rainfall partitioning, respectively [13].

Adsorbing water by branches and leaves was a primary mechanism of rainfall interception [29]. The significant impact of the water storage on leaves before rains [35] and the surface state (dry or wet) of the canopy [36] have been emphasized on the interception capacity and the consequent water balance of the canopy. In this study, we used the indicator of LW₄, representing the water adsorbed on the surface of the branches and leaves within four hours prior to raining, to quantify the hydrologic property of the canopy. Canopy wetness has significant influences on the interception and consequent partitioning of rains. In theory, throughfall and stemflow would not generate until the canopy becomes saturated [25,68]. Elevated antecedent canopy wetness accelerates canopy saturation, consequently lowering the threshold for both stemflow and throughfall initiation and, to some extent, extending their duration, thereby enhancing overall throughfall and stemflow yields [3,7,23]. Our analysis indicated that LW₄ was the dominant factor affecting throughfall TLE at clumped V. negundo, exerting significant impacts on throughfall TLE at scattered V. negundo (contribution rate 18.3%), throughfall TLM at clumped V. negundo (19.9%), and TFD at both clumped and scattered V. negundo (5.1% and 4.2%, respectively) (Figs. 3 and 6). By providing the quantitative evidence, our study firmly supported the hydraulic properties of the canopy surface for great influences on the process-based rainfall partitioning.

Wind speed has been commonly recognized for strongly influencing rainfall portioning, particularly in its magnitude, spatiotemporal attributes, and kinetic energy [26,27]. For instance, the incident angle of wind-driven precipitation induces spatial heterogeneity of throughfall. Wind enhanced throughfall and stemflow yield by promoting the expulsion of stored water from leaves and twigs [68,69,70] and impacting the perturbation of canopy water storage [3,26]. However, previous studies have rarely differentiated the wind in terms of speed and discussed their different effects on rainfall partitioning. This study investigated the influences of the speed of strong wind within a short time (i.e., the gust speed, GS) and the average wind speed of incident rains (WS), respectively. Wind can dislodge stored water from the canopy while also promoting rainwater along surface flow pathways for extended durations. This supported our results that GS is the key factor significantly affecting stemflow TLE of scattered V. negundo and significantly impacted throughfall TLE of scattered V. negundo with a contribution of 23.7%. Nevertheless, WS only demonstrated significant influence on throughfall TLG of both clumped and scattered V. negundo. This might relate to more rains intercepted on the surface and within the canopy under more inclined rains triggered by the gust [26]. Thus, the gust might contribute to beneficial rainfall partitioning dynamics by influencing the start and end of stemflow and throughfall.

4.3. Limitations and Future Research

This study assessed the meteorological impacts on rainfall partitioning of V. negundo and their relations with plant self-organization. Beyond the commonly recognized dominant role of rainfall amount [22,25,71,72], this study particularly highlighted the significant effects of non-precipitation factors, such as gust speed and antecedent canopy wetness, on rainfall partitioning processes. These findings advanced our understanding of the mechanisms governing rainfall partitioning, thereby contributing to more accurate estimations of water budgets in dryland ecosystems. Integrating the biotic influences elucidated by Yuan et al. [12] with the meteorological impacts on rainfall partitioning observed in clumped and scattered shrubs in this study further enhanced our knowledge of the plant’s drought resistance. This was achieved by the plant’s self-organized patterns (clumped vs. scattered), which optimized the harvest of critical water resources, ultimately facilitating ecological restoration efforts in degraded ecosystems. Despite the statistically significant findings regarding effective rainfall events and a comparable sample size to other regional studies in arid and semi-arid zones [21,25,73,74], this study exhibited limitations in its spatial and temporal scopes. Specifically, the investigation was confined to a small watershed, the Yangjuangou catchment on the Loess Plateau of China, focusing on V. negundo at the individual plant and hillslope scales over a two-year rainy season period. To address the potential for elevated standard deviations in data analysis and to bolster the robustness of research results, future investigations are highly encouraged to encompass more plant species, to broaden spatial scale to larger areas, such as the catchments or regions, and to extend the duration of field observations. Furthermore, the increased use of automated monitoring equipment during data collection was advised to acquire more detailed and precise process-based data, concurrently reducing labor and resource demands. Moreover, we advocate for the integration of isotopic methods and mathematical models to thoroughly examine the impacts of rainfall redistribution on critical ecological processes, including vegetation root distribution, soil water and nutrient cycling, and plant transpiration.

5. Conclusions

Rainfall partitioning is a complex ecohydrological process governed by the intricate interplay between plant traits and meteorological characteristics. Our previous study [12] elucidated the significant role of plant traits in shaping the divergent rainfall partitioning strategies of clumped and scattered xerophytic shrubs, particularly concerning stemflow quantity and efficiency. Nevertheless, the interactions between plant self-organization patterns and meteorological drivers, especially non-precipitation variables, remained largely unexplored. To address this gap, we employed the Boruta random forest algorithm to thoroughly investigate the impact of meteorological conditions on the amount, intensity, efficiency, and temporal dynamics of both throughfall and stemflow in clumped and scattered V. negundo shrubs. Our findings revealed that (1) despite comparable net precipitation amounts, clumped V. negundo exhibited greater and more efficient stemflow collection, while scattered shrubs demonstrated a higher capacity for throughfall generation. (2) Expanding on the established understanding of the dominant role of rainfall amount, we further demonstrated that canopy surface hydraulic properties, represented by LW₄, significantly influenced the temporal dynamics of rainfall partitioning, including throughfall TLE in scattered shrubs, throughfall TLM in clumped shrubs, and throughfall TFD in both clumped and scattered V. negundo. (3) Furthermore, we found that wind speed, while commonly acknowledged as an influential factor, exerts differential effects on distinct rainfall partitioning processes. Short-duration strong wind (represented by gust speed, GS) was identified as the key determinant of stemflow TLE in scattered shrubs and also significantly impacted throughfall TLE in the same self-organized pattern. In contrast, event-based average wind speed (WS) primarily influenced throughfall TLG in both clumped and scattered V. negundo. These findings deepen mechanistic understanding of rainfall harvesting processes by shrubs, offering critical insights into plant morphological adaptations and drought resilience strategies in arid and semi-arid ecosystems.