Characteristics of Four Co-Occurring Tree Species Sap Flow in the Karst Returning Farmland to Forest Area of Southwest China and Their Responses to Environmental Factors

Abstract

Monitoring stem sap flow is essential for understanding plant water-use strategies and eco-physiological processes in the ecologically fragile karst region. In the study, we continuously monitored four co-occurring species—Cryptomeria japonica var. sinensis (LS), Liquidambar formosana (FX), Camptotheca acuminata (XS), and Melia azedarach (KL)—using the thermal dissipation probe method in a karst farmland-to-forest restoration area. We analyzed diurnal and nocturnal sap flow variations across different growth periods and their responses to environmental factors at an hourly scale. The results showed (1) A “high daytime, low nighttime” sap flow pattern during the growing season for all species. (2) The proportion of nocturnal sap flow was significantly lower in the growing than in the non-growing season. (3) Daytime sap flow was primarily driven by photosynthetically active radiation (PAR) and vapor pressure deficit (VPD) during the growing season. In the non-growing season, daytime drivers were species-specific: relative humidity (RH, 39.39%) for LS; air temperature (Ta, 23.14%) for FX; PAR (33.03%) for XS; and soil moisture at a 10 cm depth (SM1, 25.2%) for KL. Nocturnal flow was governed by VPD and RH during the growing season versus soil moisture (SM1 and SM2) and RH in the non-growing season. These findings reveal interspecific differences in water-use strategies and provide a scientific basis for species selection and afforestation management in the karst ecological restoration of this research area.

1. Introduction

The karst region in southwestern China represents one of the world’s largest and most typical karst landscapes, characterized by shallow soils, poor water retention, high susceptibility to water leakage, and heightened ecological sensitivity [1]. The implementation of the Grain for Green Program (GGP) has substantially advanced vegetation restoration and ecological recovery efforts here [2]. The GGP is one of the world’s largest ecological projects, characterized by substantial investment, strong policy support, broad coverage, and high public participation. Its successful implementation has profound implications for global ecological improvement and offers valuable experience for ecological conservation worldwide [3,4]. A critical scientific question arises: in karst habitats where water is the primary limiting factor, how do restored forest ecosystems sustain their water balance? Clarifying the water-use strategies of individual trees is imperative, as it directly influences the stability, health, and long-term sustainability of these restored forests.

Stem sap flow, the core process of plant water transport, is pivotal in plant physiology research [5]. Water is critical for plant life, participating in nutrient uptake, transport, and metabolism while also regulating energy exchange with the environment through transpiration [6,7,8]. As the primary manifestation of transpiration, stem sap flow reflects internal water transport status and environmental adaptability, serving as a key indicator for studying tree water consumption and hydraulic mechanisms [5,9]. The accurate measurement of tree transpiration rates enables a deeper understanding of plant water-use patterns and supports scientific water resource management in planted forests [10].

Advances in monitoring technologies, such as heat dissipation and heat balance methods, have significantly progressed sap flow research [11]. Previous studies show that tree physiological structure—such as leaf area index, diameter at breast height (DBH), tree height, crown width, and stomatal conductance—reflects the potential capacity for sap flow [6,12]. External factors, including meteorological elements like photosynthetically active radiation (PAR), vapor pressure deficit (VPD), air temperature (Ta), and wind speed [13,14,15], as well as soil conditions like soil temperature and moisture [16,17], also significantly affect sap flow dynamics. Furthermore, nocturnal sap flow is closely associated with various physiological and ecological processes, including plant water safety, oxygen supply, and nutrient transport [18,19,20]. While earlier studies have suggested that nocturnal sap flow isrelatively weak, advances in observation techniques have revealed that nearly all plants exhibit it [21]. Therefore, studying sap flow dynamics and their response to environmental factors is key to understanding tree water-use strategies.

Recent years have seen progressive research on sap flow of specific tree species in karst regions. Existing studies primarily focus on sap flow characteristics of individual species or primary forests [22] and their responses to factors such as temperature, solar radiation, vapor pressure deficit and soil moisture—exemplified by species like Toona sinensis [23,24], Quercus glauca [25], Acer wangchii [26], Itea yunnanensis [27], and pear trees [28]. However, there has been limited attention given to coexisting tree species [29]. In naturally restored or artificially facilitated karst forests, these coexisting species are vital components of stand biodiversity and structural complexity. Nevertheless, the similarities and differences in their water-use strategies remain unclear. Moreover, synchronized and systematic comparative studies on sap flow characteristics and environmental responses of co-occurring tree species under identical site conditions are lacking. Such comparisons are fundamental for scientifically evaluating species adaptability and optimizing species composition.

Based on the identified research gaps, this study pioneers in examining four mixed tree species in the karst Grain for Green region of southwestern China. The objectives are to (1) simultaneously and continuously monitor stem sap flow velocity of four co-occurring tree species and quantify their dynamic characteristics; (2) analyze the response patterns of stem sap flow to environmental factors; (3) compare both differences and similarities in sap flow characteristics and environmental response strategies among the four co-occurring species, and investigate their divergent water-use strategies and ecological adaptations. Building on these findings, this research will evaluate the application potential of these four co-occurring species in the karst ecological restoration of this research area from a water-use perspective, providing theoretical guidance and data support for scientific species allocation and adaptive management in regional Grain for Green initiatives.

2. Materials and Methods

2.1. Study Site and Species

The study site is located in Tianchong Village, Hefeng Township, Kaiyang County, northeast of Guiyang City, Guizhou Province, southwestern China (26°58′29.3″ N, 106°53′36.5″ E, altitude: 1270 m) (Figure 1). Situated 66 km from Guiyang City, this area represents a typical karst region. The climate is humid subtropical monsoon with four distinct seasons, concurrent water and heat conditions, high humidity, abundant summer sunlight, and intense rainfall. The mean annual temperature ranges from 10.6 to 15.3 °C, and the mean annual precipitation is 1120 mm. The terrain is predominantly mountainous, with hills and basins. The soils are classified as Limestone Yellow Earth according to the Chinese Soil Taxonomy (CST), characterized by loess derived from acidic parent materials overlying limestone bedrock. This is a significant and widespread land type in the karst regions of southwestern China, exerting profound influences on local agriculture, ecology, and hydrology.

Four mixed tree species from a 2002 reforestation project were selected. For each species, three individuals with different DBH were selected for long-term continuous sap flow monitoring from 1 January to 31 December 2024. The growth characteristics of the sample trees are as follows.

Cryptomeria japonica var. sinensis Miq. (LS): An evergreen, shallow-rooted conifer with needle-like leaves. It is a non-porous wood species without vessels, exhibiting relatively low water conductivity but high resistance to embolism. It has a small leaf area, is shade- and cold-tolerant, and prefers warm and humid climates. It is an important timber and ecological afforestation species in subtropical regions.

Liquidambar formosana Hance (FX): A deciduous, deep-rooted tree with diffuse-porous wood and paper-like leaves. It has a large leaf area, is light-demanding, drought-tolerant, and tolerant of poor soils, with strong stress resistance and ecological restoration capacity.

Camptotheca acuminata Decne. (XS): A deciduous, deep-rooted tree with diffuse-porous wood and large, thin leaves. It prefers moist and fertile soils. Due to its rapid growth and shading capacity, it holds value in early-stage ecological restoration.

Melia azedarach L. (KL): A deciduous, deep-rooted tree with ring-porous wood. It has large, papery to thinly leathery leaves, is drought- and salt-tolerant, and it can grow in harsh sites. It is an excellent species for ecological restoration in degraded areas (e.g., rocky desertification zones).

Sample trees were selected based on the following: (a) health and the absence of pests/diseases; (b) the DBH within the common range for the species locally; and (c) a complete crown and similar light exposure. Basic information is provided in

Table 1. Sample tree information.

Species	Life Form	Number	Diameter at Breast Height (cm)	Tree Hight (m)	Sapwood Layer Area (cm2)	Crown (m × m)
C. japonica var. sinensis (LS)	Evergreen tree	LS1	21.75	16.5	227.87	3.7 × 3.7
		LS2	32.53	17.5	433.62	4.7 × 4.7
		LS3	27.16	23.0	325.00	4.6 × 4.6
L. formosana (FX)	Deciduous tree	FX1	23.35	22.0	575.67	4.4 × 4.1
		FX2	11.78	17.0	127.17	3.9 × 3.7
		FX3	21.80	18.0	494.69	4.2 × 4.2
C. acuminata (XS)	Deciduous tree	XS1	23.04	25.0	621.21	3.8 × 4.3
		XS2	16.19	22.0	285.41	3.3 × 2.8
		XS3	19.70	22.5	439.86	3.9 × 3.8
M. azedarach (KL)	Deciduous tree	KL1	25.64	17.6	311.19	3.5 × 3.0
		KL2	22.95	18.5	256.73	3.5 × 3.2
		KL3	15.48	15.5	129.61	4.0 × 4.3

.

2.2. Methods

2.2.1. Sap Flow Measurement

Sap flow of the 12 trees was continuously monitored using the STDP plant sap flow sensor (DF-STDP50, ORIENTAL XINHONG, Beijing, China). The operating principle involved two probes (upper and lower) heated to create a temperature gradient. A thermocouple generated an electrical signal based on the temperature difference between them. Sensors were installed 1 m above the ground [30]. After installation, probes were sealed with waterproof glue, insulated with foam, and wrapped with aluminum foil to prevent thermal interference (Figure 2). The temperature difference (ΔT, °C) was measured every 10 s, and data were recorded by a DF-CJ data collector (ORIENTAL XINHONG, Beijing, China) at 10 min intervals.

After the data collection was completed, sap flow rate was calculated. The following formula was used [31]:

$$\mathit{Fd} (\mathrm{sap} \mathrm{flow} \mathrm{rate}, {\mathrm{cm}}^3/({\mathrm{cm}}^2·\mathrm{s})) = 0.0119 \times ({\displaystyle \frac{{\mathit{\Delta }T}_{max}-\mathit{\Delta }T}{\mathit{\Delta }T}{)}^{1.231},}$$

(1)

where ΔT_max is the maximum daily temperature difference between probes when Fd = 0, and ΔT is the temperature difference when Fd > 0.

The following calculation formula for sap flow on an hourly scale was used:

Fs (sap flow density, L/h) = Fdmean × As × 3600/1000,

(2)

where Fdmean represents the average sap flow rate within one hour, and As represents the sapwood area (cm²). The sapwood area was determined by extracting cores from six trees outside the sample plot using an increment borer. After staining, the sapwood thickness was measured and the sapwood area was calculated. Species-specific allometric equations between the sapwood area and DBH were established as follows:

LS: As = 1.6597DBH^1.5983, R² = 0.9593,

(3)

FX: As = 0.55DBH^2.027, R² = 0.9694,

(4)

XS: As = 0.6166DBH^2.0424, R² = 0.913,

(5)

KL: As = 1.1547DBH^1.7358, R² = 0.8923,

(6)

where As represents the sapwood area, and DBH represents the diameter at breast height.

We acknowledge that the Granier (TDP) method may underestimate nocturnal sap flow due to potential inaccuracies in the zero-flow reference (ΔT_max) under low flow conditions [32].

2.2.2. Environmental Factors Monitoring

To study sap flow response to environmental factors—including air temperature (Ta), relative humidity (RH), photosynthetically active radiation (PAR), rainfall (P) soil temperature (ST) and soil moisture (SM) at depths of 10, 20, 30, 40, and 50 cm (denoted as ST1–ST5 and SM1–SM5, respectively)—were selected for synchronous monitoring with sap flow by an automatic weather station at the study site. This was designed to holistically capture the water dynamics from the atmosphere through the entire soil profile, thereby reflecting conditions from the surface to the root zone. All data were logged at 1 min intervals using a DF-CJ data collector.

Vapor pressure deficit (VPD) is the difference between the saturated vapor pressure and the actual vapor pressure in the air at the current temperature, representing the synergistic effect of the atmospheric temperature and relative humidity of the air. It is calculated with the following formula [33]:

$$VPD=\left(1-RH\right)\times 0.611 \times \mathrm{e}\mathrm{x}\mathrm{p}(17.502\times Ta/(Ta+240.97\left)\right),$$

(7)

where VPD is vapor pressure deficit (kPa), RH is the relative humidity (%), and Ta represents the air temperature (°C).

2.3. Data Processing

2.3.1. Data Quality Control

Data were quality-controlled to exclude outliers caused by power failure or sensor malfunction. The following methods were applied [34]: (1) visual inspection of the plotted data, and (2) removal of values beyond ±3 standard deviations from the mean.

2.3.2. Data Arrangement

Based on radial growth patterns, growing seasons were defined as follows: April–July for C. japonica var. sinensis, June–September for L. formosana, June–August for C. acuminata, and April–September for M. azedarach. The remaining months were classified as non-growing seasons. Daytime and nighttime were distinguished using a photosynthetically active radiation threshold of 5 W·m⁻² [20].

2.3.3. Statistical Analysis

Due to the non-normal distribution of sap flow data, the Wilcoxon test method was used to examine the differences between day/night and growing/non-growing seasons. Wilcoxon rank-sum test (Mann–Whitney U test), introduced by Frank Wilcoxon in 1945 [35], is a non-parametric test used to assess the significance of differences in the central tendency between two independent groups that do not meet parametric assumptions. The correction method used was Holm Correction and the α level was 0.005.

To classify nocturnal sap flow, this study employed an ARMAX (p, q, r) model to establish a fitting relationship between nocturnal sap flow and two environmental drivers: VPD and SM1 [36]. The Nash–Sutcliffe efficiency (NSE) and the ratio of the root mean square error to the standard deviation of observations (RSR) were used as model performance metrics to quantify nocturnal transpiration and stem rehydration. The results were classified into four grades: Grades I and II (0.75 < NSE ≤ 1 and 0 ≤ RSR ≤ 0.5; 0.5 < NSE ≤ 0.75 and 0.5 < RSR ≤ 0.7) indicated a good fit between nocturnal sap flow and environmental drivers, which was attributed to nocturnal transpiration. Grades III (0 < NSE ≤ 0.5 and 0.7 < RSR < 1) and IV (NSE = 0 and RSR = 1) indicated a poor fit with environmental drivers, which was attributed to stem rehydration [36,37].

To evaluate the relative importance of the environmental variables, a random forest (RF) regression model was employed. The main procedures were as follows: (1) Data preparation: compiling hourly scale sap flow and environmental data; (2) preprocessing: removing duplicates and outliers, then randomly splitting data into a training set (70%) and testing set (30%); (3) model establishment: outputting the optimal model after hyperparameter tuning via 5-fold cross-validation; (4) model validation and evaluation: the model was evaluated using performance metrics (KGE, R², RMSE, and MAE); and (5) the feature importance results were obtained. Random forest regression, an ensemble algorithm proposed by Leo Breiman [38], generates predictions by averaging the results of multiple decision trees. It is robust against missing data and outliers, handles high-dimensional data without scaling, and provides estimates of feature importance. In Random forest regression, feature importance is measured by increase in MSE—the percentage increase in the mean squared error after permuting a feature, where a higher value denotes greater importance. Model performance is evaluated using KGE (Kling–Gupta efficiency), R² (variance explained), RMSE (standard deviation of residuals), and MAE (average absolute error).

A Linear Mixed-Effects Model (LME) [39] was applied to analyze the data with hierarchical structure (three individuals per species). Environmental variables were treated as fixed effects. The Variance Inflation Factor (VIF) was calculated to remove variables with multicollinearity. The model fitting results are represented by Marginal R², Random R², and Conditional R². Marginal R² is the proportion of variance explained by fixed effects alone. Random R² is the proportion of variance explained by random effects alone. Conditional R² is the proportion of variance explained by both fixed effects and random effects together and reflects the overall fit of the model. Variance partitioning decomposed the contribution of random effects (month and tree individual).

Analyses were conducted using Excel 2019 for data processing and R version 4.2.0 for statistical analysis and visualization.

3. Results

3.1. Dynamic Variation in Environmental Factors

3.1.1. Meteorological Factors

During the observation period, the key meteorological drivers of plant transpiration—air temperature (Ta), relative humidity (RH), vapor pressure deficit (VPD), photosynthetically active radiation (PAR), and rainfall (P)—exhibited distinct seasonal dynamics (Figure 3). These patterns collectively define the varying atmospheric demand for water and the energy available for sap flow.

Monthly trends showed that Ta was unimodal, RH and P peaked in June, while PAR and VPD shared a bimodal pattern with a June trough due to heavy rainfall. Specifically, the monthly mean Ta followed a unimodal trend, reaching its annual peak in July (23.91 °C), which corresponded to the period of maximum potential for photosynthetic and transpirational activity. The lowest monthly mean Ta was recorded in December (4.09 °C), indicating the cold constraints of the winter season. In contrast, RH and rainfall peaked in June (95.28% and 174.20 mm, respectively), creating a period of high water availability and low atmospheric demand.

Crucially for sap flow drive, VPD and PAR both exhibited a bimodal pattern with a pronounced trough in June. This depression was directly attributable to the concurrent peaks in RH and rainfall, which reduced evaporative demand. VPD, which directly reflects the atmosphere’s capacity to draw water from the leaves, reached its highest monthly mean in August (0.57 kPa), suggesting a period of high transpirational pull. PAR, the energy source for photosynthesis, showed a similar trend, with its highest monthly mean also in August (44.24 W·m⁻²), implying a late-summer peak in photosynthetic potential. The concentration of rainfall from April to September demarcated the core growing season, during which soil water recharge events were frequent and water stress was likely alleviated.

3.1.2. Soil Factors

Soil temperature and moisture, as critical factors influencing root water uptake and sap flow, displayed distinct seasonal variations (Figure 4). The peak monthly mean soil temperatures occurred in August (up to 22.17 °C at the shallowest depth), coinciding with the period of maximum plant physiological activity, while the minimum temperatures in February (as low as 7.02 °C) reflected the dormant season. Interestingly, the thermal regime inverted with depth between seasons: the shallowest soil was warmest in summer, whereas the deepest soil was warmest in winter (8.20 °C), indicating differential heat transfer and storage.

Soil moisture patterns were consistently ranked across depths (SM2 > SM1 > SM4 > SM3 > SM5) during both wet and dry periods. The temporal shifts in these values, from highs (e.g., 33.13% at SM2) to lows (e.g., 12.93% at SM5), demarcated the transition between water-sufficient and water-stressed conditions. The persistent high moisture at SM2 (16.27–33.13%) suggests it may represent a key water-holding horizon for the root system.

3.2. Dynamic Variation in Sap Flow

Hourly sap flow density varied among species and months (Figure 5). LS and XS exhibited an inverted U-shaped diurnal pattern (higher daytime but lower nighttime flow) from January to November. FX followed this typical pattern from April to November and KL from May to October. During the remaining months, FX and KL displayed a U-shaped curve, with higher sap flow at night. From April to September, daily sap flow of XS and FX was significantly higher than that of LS and KL. Peak sap flow in LS generally occurred between 14:00 and 16:00, except in December (03:00). For FX, peaks generally occurred between 12:00 and 15:00, except in January and December (05:00), February (00:00), and March (23:00). For XS, peaks generally occurred between 12:00 and 15:00, except in February (05:00). For KL, peaks occurred between 10:00 and 14:00 from May to October, but at 23:00 in January, 07:00 in February, 05:00 in March and April, 03:00 in November, and 06:00 in December.

3.3. Comparison of Daytine and Nighttime Sap Flow and Division of Nighttime Sap Flow

Mean daytime sap flow during the growing season exceeded that in the non-growing season for all species. Except for KL, nighttime sap flow in the non-growing season was greater than in the growing season (

Table 2. The results of Wilcoxon test on sap flow density.

Tree	Group1			Group2			n1	n2	Statistic	p Value	Effect Size
LS	non-growing-day	Median	0.088	growing-day	Median	0.073	8108	3392	14,769,898	<0.001	0.009
		Q1	0.023		Q1	0.023
		Q3	0.324		Q3	0.183
		IQR	0.301		IQR	0.161
	non-growing-night	Median	0.083	growing-night	Median	0.087	9612	3170	14,978,645	0.155	−0.002
		Q1	0.045		Q1	0.052
		Q3	0.13		Q3	0.12
		IQR	0.085		IQR	0.068
	growing-night	-		growing-day	-		3170	3392	5,626,023.5	<0.001	0.006
	non-growing-night	-		non-growing-day	-		9612	8108	36,219,140.5	<0.001	−0.011
FX	non-growing-day	Median	0.12	growing-day	Median	0.659	6421	4952	8,262,302	<0.001	−0.418
		Q1	0.031		Q1	0.176
		Q3	0.409		Q3	1.861
		IQR	0.378		IQR	1.685
	non-growing-night	Median	0.104	growing-night	Median	0.141	7983	4682	17,863,996.5	<0.001	−0.008
		Q1	0.033		Q1	0.033
		Q3	0.313		Q3	0.384
		IQR	0.28		IQR	0.352
	growing-night	-		growing-day	-		4682	4952	5,599,562.5	<0.001	−0.458
	non-growing-night	-		non-growing-day	-		7983	6421	24,589,170	<0.001	−0.007
XS	non-growing-day	Median	0.145	growing-day	Q3	1.642	8491	2956	6,228,613	<0.001	−1.051
		Q1	0.042		IQR	0.264
		Q3	0.6		Q3	5.654
		IQR	0.558		IQR	5.39
	non-growing-night	Median	0.079	growing-night	Median	0.083	10,323	2477	12,506,170.5	0.091	−0.002
		Q1	0.027		Q1	0.017
		Q3	0.174		Q3	0.235
		IQR	0.147		IQR	0.217
	growing-night	-		growing-day	-		2477	2956	1,081,011.5	<0.001	−1.478
	non-growing-night	-		non-growing-day	-		10,323	8491	32,081,770.5	<0.001	−0.058
KL	non-growing-day	Median	0.1	growing-day	Median	0.556	5585	5747	7,768,638.5	<0.001	−0.34
		Q1	0.019		Q1	0.245
		Q3	0.4		Q3	1.27
		IQR	0.381		IQR	1.024
	non-growing-night	Median	0.112	growing-night	Median	0.139	7232	5239	19,533,593.5	0.003	0.005
		Q1	0.033		Q1	0.035
		Q3	0.509		Q3	0.37
		IQR	0.476		IQR	0.335
	growing-night	-		growing-day	-		5239	5747	7,062,419	<0.001	−0.351
	non-growing-night	-		non-growing-day	-		7232	5585	22,070,092.5	<0.001	0.014

Notes: Q1: first quartile, Q3: third quartile, IQR = Q3 − Q1 (a measure of statistical dispersion). The short line (-) indicates that the content at this location was repetitive and thus omitted. A minus sign before a value in the Effect Size column only indicates direction. The background color is used to make it easier to distinguish the content of each row in the table.

). For LS, median daytime sap flow was 0.073 (Q1: 0.023, Q3: 0.183) L·h⁻¹ in the growing season and 0.088 (Q1: 0.023, Q3: 0.324) L·h⁻¹ in the non-growing season; and the nighttime values were 0.087 (Q1: 0.052, Q3: 0.120) L·h⁻¹ and 0.083 (Q1: 0.045, Q3: 0.130) L·h⁻¹, respectively, with no significant nocturnal difference between seasons (Wilcoxon test: p = 0.155, effect size: −0.002, and the minus sign represents direction). For FX, mean daytime sap flow was (Q1: 0.176, Q3: 1.861) L·h⁻¹ (growing) and 0.12 (Q1: 0.031, Q3: 0.409) L·h⁻¹ (non-growing); and the nighttime values were 0.141 (Q1: 0.033, Q3: 0.384) L·h⁻¹ and 0.104 (Q1: 0.033, Q3: 0.313) L·h⁻¹. For XS, mean daytime sap flow was 1.642 (Q1: 0.264, Q3: 5.654) L·h⁻¹ (growing) and 0.145 (Q1: 0.042, Q3: 0.6) L·h⁻¹ (non-growing); nighttime values were 0.083 (Q1: 0.017, Q3: 0.235) L·h⁻¹ and 0.079 (Q1: 0.027, Q3: 0.174) L·h⁻¹, with no significant nocturnal difference (p = 0.091). For KL, mean daytime sap flow was 0.556 (Q1: 0.245, Q3: 1.270) L·h⁻¹ (growing) and 0.1 (Q1: 0.019, Q3: 0.4) L·h⁻¹ (non-growing); nighttime values were 0.139 (Q1: 0.035, Q3: 0.370) L·h⁻¹ and 0.112 (Q1: 0.033, Q3: 0.509) L·h⁻¹ (Table 2). FX and KL showed significant differences in both daytime and nighttime sap flow between the growing season and non-growing season, whereas LS and XS exhibited a significant difference in daytime sap flow between the growing season and non-growing season but showed no significant difference in nighttime sap flow during non-growing season.

The proportions of daytime and nighttime sap flow are shown in Figure 6. Nocturnal sap flow exceeded 50% in LS and XS in December, in FX from January to April and December, and in KL from January to April and November to December. During the growing season, nocturnal sap flow accounted for 14.53–30.19% in LS, 3.90–17.50% in FX, 2.65–11.02% in XS, and 17.75–61.84% in KL. During the non-growing season, proportions were 17.89–58.19% for LS, 15.47–62.77% for FX, 6.25–52.23% for XS, and 33.43–71.50% for KL. These results indicate distinct seasonal variations in nocturnal sap flow contribution across species.

The classification results of nocturnal sap flow for each tree species are presented in Figure 7. During the growing season, nocturnal transpiration exceeded stem rehydration in FX and XS, with nocturnal transpiration accounting for 55.2% and 55.9%, and recharge accounting for 44.8% and 44.1%, respectively. In contrast, nocturnal transpiration was lower than stem rehydration in LS and KL, with nocturnal transpiration accounting for 33.5% and 36.6%, and recharge accounting for 66.5% and 63.4%, respectively. During the non-growing season, the proportion of nocturnal transpiration was lower than that of stem rehydration across all species. Specifically, the nocturnal transpiration proportions for LS, FX, XS, and KL were 17.2%, 49.0%, 32.8%, and 37.2%, respectively, while the corresponding recharge proportions were 82.8%, 51.0%, 67.2%, and 62.8%, respectively. The results indicate that during the growing season, nocturnal sap flow in FX and XS was predominantly allocated to nocturnal transpiration, while in LS and KL, it was mainly utilized for stem rehydration. During the non-growing season, nocturnal sap flow in all tree species was primarily directed toward stem rehydration.

3.4. Response of Sap Flow Density to Environmental Factors

3.4.1. Results of the Random Forest Regression Model

The prediction accuracy results of the random forest regression models are shown in

Table 3. Random forest regression model prediction accuracy indicators.

Season	Species	Day					Night
		KGE	R2	RMSE	MAE	N	KEG	R2	RMSE	MAE	N
Growing	LS	0.74	0.80	0.16	0.29	1030	0.59	0.54	0.28	0.41	2441
	FX	−0.22	0.64	0.34	0.43	1530	0.29	0.35	0.39	0.45	1916
	XS	0.84	0.90	0.10	0.21	898	0.81	0.53	0.26	0.31	2536
	KL	0.48	0.26	0.65	0.60	1738	0.43	0.52	0.27	0.35	1679
Non-growing	LS	0.57	0.85	0.13	0.21	960	0.53	0.63	0.20	0.35	2886
	FX	0.56	0.16	0.41	0.36	1404	0.38	0.15	0.69	0.60	2428
	XS	0.67	0.63	0.14	0.23	747	0.49	0.19	0.44	0.48	3077
	KL	0.56	0.33	0.49	0.44	1560	0.5	0.25	0.38	0.38	2180

. The results indicate that the prediction accuracy for LS and XS was relatively good, while it was comparatively poor for FX and KL. XS achieved the highest prediction accuracy during the growing season daytime (KGE = 0.84, R² = 0.90, RMSE = 0.10, MAE = 0.21) and growing season nighttime (KGE = 0.81, R² = 0.53, RMSE = 0.26, MAE = 0.31). LS exhibited the highest prediction accuracy during the non-growing season daytime (KGE = 0.57, R² = 0.85, RMSE = 0.13, MAE = 0.21) and non-growing season nighttime (KGE = 0.53, R² = 0.63, RMSE = 0.20, MAE = 0.35). In general, except for KL, prediction accuracy was higher during the daytime than at night, and higher during the growing season than during the non-growing season.

Random forest modeling regression analysis was conducted on the data during the observation period to analyze the order of variable importance results, as shown in Figure 8. The results show that during the growing season, the most important factor affecting LS and KL was VPD, with importance of 31.88% and 25.28%, respectively. The top three most important factors affecting FX and XS were PAR, Ta, and VPD in sequence, with importance of PAR (FX: 40.44%; XS: 35.07%), Ta (FX: 28.04%; XS: 25.36%), and VPD (FX: 27.33%; XS: 24.36%). In contrast, during the non-growing season, the most important factors affecting each tree species during the day were different. The factors and their importance were, respectively, as follows: LS (RH: 39.39%), FX (Ta: 23.14%), XS (PAR: 33.03%), and KL (SM1: 25.2%). Meanwhile during the growing season at night, RH was among the top three influencing factors for each tree species. The importance of RH was as follows: LS (39.9%), FX (27.31%), XS (29.36%), and KL (39.11%). In addition, except for XS, the VPD of the other tree species was also among the top three influencing factors, with their importance being LS (38.16%), FX (27.66%) and KL (37.61%), respectively. On the contrary, during the night of the non-growing season, SM1 was the most significant influencing factor for LS and KL, with importance values of 51.19% and 31.76%, respectively. RH was the most crucial factor for FX and XS, with importance values of 28.22% and 36.01%, respectively. The second and third most influential factors for LS were RH (50.02%) and SM2 (43.17%), while for KL, they were SM2 (31.64%) and P (31.05%). For FX, the second and third most influential factors were SM1 (27.87%) and SM2 (27.63%), while for XS, they were VPD (34.29%) and SM2 (28.23%).

The results demonstrate distinct environmental drivers of sap flow across seasons and diel periods. During the growing season, PAR and VPD emerged as the dominant daytime controls, whereas VPD and RH were primary nocturnal drivers. In contrast, non-growing season dynamics exhibited species-specific regulatory patterns, but with VPD serving as the principal daytime controlling factor, and soil moisture at 10 cm and 20 cm depth (SM1, SM2) and RH for nighttime.

3.4.2. Results of the Linear Mixed-Effects Model

The mixed-effects model analysis of the relationship between sap flow and environmental drivers for different tree species revealed significant interspecific, seasonal, and diurnal variations in the model fit and driver responses (

Table 4. The goodness-of-fit of the linear mixed-effects model.

Season	Species	Day			Night
		Marginal R2	Random R2	Conditional R2	Marginal R2	Random R2	Conditional R2
Growing	LS	0.463	0.179	0.642	0.072	0.073	0.145
	FX	0.188	0.488	0.676	0.134	0.283	0.417
	XS	0.599	0.183	0.782	0.284	0.130	0.414
	KL	0.094	0.194	0.288	0.040	0.325	0.365
Non-growing	LS	0.508	0.131	0.639	0.085	0.090	0.175
	FX	0.102	0.203	0.305	0.050	0.406	0.456
	XS	0.231	0.329	0.560	0.138	0.202	0.340
	KL	0.067	0.196	0.263	0.051	0.499	0.550

). Analysis of the model explanatory power showed a marked difference between marginal R² (R2m) and conditional R² (R2c), indicating that random effects (e.g., month and individual tree differences) collectively accounted for a substantial portion of the total variation. Specifically, LS and XS exhibited relatively high R2m during the growing season daytime (0.463 and 0.599, respectively), suggesting their sap flow was primarily driven by the measured environmental factors. In contrast, FX and KL showed low R2m under most conditions (e.g., 0.102 for FX during the non-growing season daytime), indicating that their variation stemmed largely from random effects such as individual tree differences. R2m generally decreased during nighttime and non-growing season conditions, indicating a weaker or more complex relationship between sap flow and environment under these scenarios.

The variance partitioning results directly explained the above differences (

Table 5. The variance component of the random effect in the linear mixed-effects model.

Season	Species	Day			Night
		Month	Tree Id	Residual	Month	Tree Id	Residual
Growing	LS	0.052	0.394	0.561	0.110	0.146	0.623
	FX	0.049	1.074	0.877	0.127	0.362	0.551
	XS	0.566	0.107	0.629	0.369	0.427	1.200
	KL	0.098	0.453	0.889	0.328	0.381	0.703
Non-growing	LS	0.160	0.335	0.616	0.245	0.239	1.036
	FX	0.134	0.163	0.390	0.243	0.736	0.898
	XS	0.342	0.096	0.411	0.338	0.205	0.713
	KL	0.172	0.371	0.792	0.450	0.799	0.871

). Tree individual differences (Tree id) were the primary source of random variation, excluding residuals, with their standard deviation generally far greater than the variation caused by months (Month). The residual standard deviation was generally comparable to or larger than the tree individual standard deviation, suggesting that the unexplained random variation still constituted a significant proportion. For XS during the growing season daytime, the variation was predominantly driven by the month effect, aligning with its high R2m. Conversely, variation for FX and KL was mainly attributed to individual tree differences (Tree id) and residuals, resulting in limited explanatory power from the fixed effects. LS showed a more balanced composition of the variance sources.

The analysis of fixed effects further revealed species-specific responses to environmental drivers (Figure 9). VPD was a widespread and strong positive driver, although KL showed a negative response to VPD during the daytime. PAR had a positive effect on most species during the growing season daytime but a consistently negative effect on LS, suggesting potential photoinhibition. Deep soil moisture (SM5) showed a stable positive effect in most models. In contrast, the effects of shallow soil temperature (ST1) and moisture (SM1) varied depending on the species and condition. Air temperature (Ta) exhibited negative effects on some species during the nighttime.

4. Discussion

4.1. Dynamic Variation in Sap Flow

This study revealed interspecific differences in sap flow dynamics. The diurnal patterns of LS and XS, characterized by a typical inverted U-shaped curve with higher daytime and lower nighttime flow, remained stable throughout most of the year. In contrast, FX and KL only followed this pattern during their respective growing seasons (FX: April–November; KL: May–October). These differences were closely related to species-specific phenological traits [40]. As an evergreen species, LS maintains photosynthetic organs year-round, supporting relatively continuous physiological activity across all seasons [41,42]. In contrast, sap flow activity of the deciduous species FX and KL was strictly regulated by leaf phenology—leaf expansion and abscission. During the non-growing season, leaf loss led to a significant reduction in transpirational pull, fundamentally altering their sap flow dynamics. Consequently, the proportion of nocturnal sap flow increased markedly across all species, reflecting a functional shift from “supporting growth” to “sustaining survival” [43]. This study quantified the allocation of nocturnal sap flow during both the growing and non-growing seasons. In FX and XS, nocturnal sap flow during the growing season was predominantly used for nocturnal transpiration, indicating that their stomatal conductance remained relatively high at night [44]. Conversely, in LS and KL, nocturnal sap flow was primarily allocated to stem rehydration, reflecting their primary nighttime role in repairing xylem embolisms and restoring water potential caused by daytime transpiration. During the non-growing season, nocturnal sap flow in all studied tree species was mainly directed toward stem rehydration [45,46,47]. This pattern suggests that under conditions of limited water availability or reduced growth activity, trees generally enhance their mechanisms for maintaining xylem water potential to prevent hydraulic failure due to xylem embolism.

4.2. Response of Sap Flow Density to Environmental Factors

This study employed a combination of random forest regression and linear mixed-effects models to systematically reveal the response mechanisms of sap flow in four tree species to environmental drivers. The results show that the predictability of sap flow significantly differs among species. Their environmental regulation patterns undergo fundamental shifts across seasons and between day and night. Importantly, inter-individual variation emerged as a key factor driving differences in model explanatory power.

The goodness-of-fit of the random forest and linear mixed-effects models generally followed the patterns of “growing season > non-growing season” and “daytime > nighttime”. This can be attributed to the distinct physiological processes governing sap flow during these periods. During the daytime in the growing season, sap flow was predominantly driven by transpiration, a process tightly coupled with meteorological factors such as PAR and VPD. The well-defined physical mechanisms of transpiration make it relatively straightforward for the model to capture [48,49,50]. In contrast, sap flow during the non-growing season and at night was influenced more by processes like stem rehydration, tissue maintenance, and root hydraulic redistribution. These processes are not only weakly driven by instantaneous environmental variables but may also involve significant time-lag effects and complex internal plant physiological rhythms. Consequently, the model’s explanatory power generally decreases under these conditions [22,51]. From a species-specific perspective, the models for LS and XS demonstrated better performance, indicating successful capture of their substantial sap flow responses to environmental drivers. In contrast, for species FX and KL, the proportion of sap flow variation explained by fixed effects (measured environmental factors) was low. However, their conditional R² values indicated that the models still captured most of the overall variation (Table 4). Variance partitioning results (Table 5) directly revealed the source of this apparent contradiction: for FX and KL, inter-individual differences (Tree id) and unmeasured factors (residuals) jointly constituted the major components of their sap flow variation [52]. In particular, during daytime in the growing season, for FX, the random forest regression model metrics (KGE = −0.22, R² = 0.64, RMSE = 0.34, and MAE = 0.43) indicate that the relatively high R² value (0.64) suggests that the model successfully captured the main temporal trends of FX xylem sap flow. However, decomposition of the KGE (r = 0.82, β = 2.07, γ = 0.45) revealed an overestimation of the significant bias ratio (β). This was corroborated by the results of decomposition of the variance of random effects in the mixed-effects model (Tree id: 1.074, Residual: 0.877). High residuals indicate that important physiological or environmental variables not measured in this study may have dominated their xylem flow processes, such as soil–plant hydraulic conductances and soil water potentials, etc. [53]. Future research needs to integrate these traits to reveal the underlying mechanisms of xylem flow regulation [54].

Previous studies have established that PAR and VPD are the primary meteorological factors governing transpiration, as they determine atmospheric evaporative demand and sap flow driving forces [55,56]. PAR reflects the photon-driven force of transpiration, directly regulating stomatal conductance, photosynthesis, and leaf temperature. Increases in PAR and concurrent rises in VPD and Ta enhance leaf–atmosphere water exchange and transpirational pull, thereby increasing sap flow [57,58,59]. The results of this study show that during the growing season, the most important factors affecting LS and KL are VPD. For FX and XS, the most important factors are PAR, and the most important influencing factors are positively correlated with sap flow (Figure 8). FX and XS exhibited patterns typical of photosynthesis-driven sap flow, characterized by high synchrony with PAR. This reflects a carbon-acquisition priority strategy adopted by these species in the study area. During periods when water remains available, they maintain high stomatal conductance to fully utilize light and thermal resources, thereby supporting their rapid growth [24,26,60]. In contrast, LS and KL demonstrated a different adaptive pathway. Their sap flow was primarily governed by VPD, indicating high sensitivity to atmospheric aridity. In the environment of the study area, this VPD-driven response mechanism helps effectively prevent xylem embolism, representing a typical hydraulic-safety-first strategy [15].

However, during the non-growing season, the most important factors affecting each tree species are different, namely, RH for LS, Ta for FX, PAR for XS, and SM1 for KL. The possible reason is that during the non-growing season when environmental pressure is relatively low, LS respond to the low VPD (high RH) by actively closing their stomata, thereby reducing transpiration and sap flow [28]. This is a conservative and risk-avoiding strategy. FX does not completely enter dormancy in winter. Its stem still needs to undergo weak respiration, enzymatic reactions, and other life activities. The rates of these processes are mainly controlled by Ta, and as temperature rises, the rate of biochemical reactions accelerates, slightly increasing the demand for water and energy, thereby driving weak sap flow. At low temperatures, the physiological activities of XS mainly rely on light energy to drive, and it may carry out weak photosynthesis, with PAR becoming the key limiting factor for sap flow [61]. As a deeply rooted tree species, KL ensures its drought tolerance. During the non-growing season when water demand is low, its weak life activities rely on and respond to changes in the shallow soil environmental conditions (SM1). Otherwise, it will enter a dormant state. The differences in the influencing factors of sap flow during the non-growing season reflect the differentiated survival strategies of various tree species under adverse conditions.

It has been reported that the Granier method employed may underestimate actual sap flow under low sap flow conditions. Nevertheless, underestimation is relative, not absolute. Significant and considerable nocturnal sap flow values were observed across all four tree species in this study, especially during the non-growing season. Previous studies indicate that nocturnal sap flow is influenced by VPD and wind speed [62,63,64]. They control the occurrence of nocturnal transpiration by affecting the closure of stomata in plants at night. Soil moisture is the fundamental factor influencing transpiration intensity [21]. In this study, RH ranked among the top three factors for all species during growing season nights, and VPD was important for all except XS [65]. After stomatal closure, RH may regulate non-stomatal transpiration, while VPD reflects potential water loss risk. During non-growing season nights, RH (for FX and XS) and SM1 (for LS and KL) were the most important factors, but their explanatory power was low, indicating complex multi-factor regulation.

This study demonstrates that random forest models and linear mixed-effects models possess powerful complementarity in ecological mechanism research. The random forest model excels at identifying key drivers with complex, nonlinear relationships and delivers high predictive accuracy, thereby guiding focused mechanistic inquiry. In contrast, the linear mixed-effects model quantifies the direction and magnitude of environmental effects while partitioning variance into fixed and random components, revealing the critical role of individual differences in explaining species-specific responses. This combined approach, integrating “data-driven prediction” with “mechanistic statistical explanation,” effectively avoids the limitations of using a single model. For instance, while the random forest indicated low prediction accuracy for FX and KL, the mixed-effects model explicitly identified individual random effects as the primary cause. This provides a clear direction for future research: to combine the search for novel environmental factors with the measurement of individual traits to jointly explain sap flow variation.

4.3. Limitations

While this study, through year-round monitoring of four key tree species, provides important preliminary insights into tree water-use dynamics in a karst Grain for Green region, certain limitations in the research design must be acknowledged. These limitations also indicate directions for future investigations.

Firstly, the data originated from a single karst hillslope site. While this approach provided detailed phenological and hydrological data for this specific location, the high intrinsic heterogeneity of karst landscapes—such as spatial variations in the epikarst structure, soil depth, and moisture distribution—makes it difficult to directly extrapolate the findings to areas with different geological backgrounds or topographic conditions. This limits the ability to distinguish between regional patterns and local, site-specific phenomena. Furthermore, this study predominantly focused on tree transpirational water use, with insufficient attention given to a critical karst hydrological process—the exchange within the rock–soil–groundwater system. The soil moisture monitoring did not capture the rapid movement of water through fissures or the potential contribution of deep soil water or groundwater. Consequently, this gap constrains our capacity to model water fluxes within the entire soil–plant–atmosphere continuum holistically.

Despite these limitations, this study undoubtedly highlights the importance of considering nocturnal sap flow and interspecific differences in karst ecohydrological research. The dataset we established serves as a valuable benchmark case for the field. Future research should strive to establish replicated observation sites across a broader range of karst regions and integrate more direct measurements of physiological and hydrogeological parameters. Through such multi-site synthesis, we can progressively build a more comprehensive and predictive vegetation-hydrology model.

5. Conclusions

Sap flow monitoring of four tree species in a karst Grain for Green area revealed distinct water-use strategies. LS exhibited a “conservative” strategy with stable diurnal patterns. FX and XS demonstrated “photosynthesis-driven” strategies, with daytime sap flow regulated by PAR and Ta. KL displayed a unique “deep-rooted drought-avoidance” strategy, showing reversed diurnal patterns seasonally while maintaining high nocturnal sap flow. Environmental drivers varied diurnally and seasonally. During the growing season, daytime sap flow was primarily controlled by PAR and VPD across all species. In contrast, non-growing season drivers were species-specific. Nocturnal sap flow during the growing season was mainly driven by VPD and RH, while shallow soil moisture (SM1, SM2) and relative humidity were key determinants in the non-growing season. These complementary water-use strategies enable species coexistence in the karst ecosystems. For restoration initiatives, we recommend employing FX and XS for rapid vegetation cover, while the drought-resistant LS with the compromise KL. To advance our understanding, future studies must expand monitoring sites in karst Grain for Green areas and to synthesize diverse datasets. This effort is key to building a process-based framework capable of reliably predicting ecosystem dynamics, thereby providing critical support for targeted restoration strategies.