Variability of Water Use Efficiency of Gmelina arborea Plantations in the Tropical Dry Forest of Colombia

Abstract

Effective forest management strategies to adapt to climate change are essential. Water use efficiency (WUE), which integrates biomass production and water consumption, is a key indicator of forest adaptation. This study evaluated the WUE of Gmelina arborea plantations in the tropical dry forest and identified the main influencing factors, with implications for silviculture and management. Data on total biomass (TB) and total volume (V) were obtained from permanent sample plots in the upper and lower Magdalena River basin in Colombia. WUE was calculated as m³ of V (WUE_V) or kg of TB (WUE_TB) per m³ of evapotranspired water. Significant regional differences were found, with higher WUE in the Caribbean plains (WUE_TB = 1 kg m⁻³ and WUE_V = 0.0018 m³ m⁻³) compared to the inter-Andean valleys (WUE_TB = 0.77 kg m⁻³ and WUE_V = 0.013 m³ m⁻³). Stand variables had the greatest influence on WUE, showing positive associations with site productivity and stand density measures. Soil variables such as texture, available water, and calcium content, along with a drier climate, were crucial for achieving higher WUE. The study underscores the importance of comprehensive site selection and effective silvicultural practices to maximize WUE and productivity, especially in the context of climate change.

1. Introduction

Continued global warming has contributed to the intensification of climate change and its impacts on ecosystems and the resources available for forest growth and productivity [1,2,3,4,5]. Many forest ecosystems are currently under constant water stress and droughts, which are expected to become increasingly recurrent and severe [6], limiting water availability for forests [7].

Water limitations are reflected in forest productivity, growth rates, and resource use efficiency [8,9,10]. Low water availability can limit carbon fixation, significantly reduce biomass production, and, in extreme cases, contribute to tree mortality [11,12]. Estimating water use at the tree and stand level is becoming increasingly important for both forest management and water resource management [13]. Understanding the relationship between water requirements for biomass production is essential for adapting forest management to climate change [14]. In the literature, this relationship is referred to as water use efficiency (WUE) [15,16,17].

In general, WUE is determined by the relationship between biomass produced and water consumed, a balance that can be assessed at different levels, from instantaneous fluxes at the leaf level (e.g., photosynthesis rates vs. transpiration rates), tree level (e.g., biomass produced vs. water transpired), and stand or watershed level (e.g., yield vs. water consumed) [15,16,17,18,19,20].

This relationship between the amount of dry matter produced and the amount of water transpired depends mainly on the characteristics of the species (e.g., its genetics) and those related to the ability to optimize carbon assimilation processes and water evapotranspiration [21,22], as well as to the characteristics of the environment in which the plant grows and develops and forest management practices [23,24,25,26]. Spatial variation in net primary productivity, evapotranspiration, transpiration, and WUE has been attributed to stand variables such as forest type, stand age, stand density [14,27], and site quality [8]. Therefore, it is now important to understand the role of forest management practices, stand characteristics, and site variables with respect to WUE.

In tropical forests, water use has also been reported to depend on variables such as tree size, sapwood area, and leaf area index [13]. Among environmental variables, temperature, soil moisture, and vapor pressure deficit have been shown to have significant relationships with both growth and water use or transpiration [13,27].

Among the tree species planted in Colombia, Gmelina arborea Roxb. stands out as one of the most important. It is native to habitats ranging from the wettest to the driest in India, Bangladesh, Sri Lanka, Myanmar, Thailand, and parts of the Asian continent [28]. It is currently the second fastest-growing hardwood species, surpassed only by some species of the genus Eucalyptus [29]. In Colombia, it occupies about 20,000 ha, mainly in the plains of the Caribbean coast and in the interior of the inter-Andean valleys of the Magdalena River. G. arborea grows in the tropical dry forest, mainly characterized by limited water availability and high evaporative demand, in places where an increase in drought episodes is expected, with the expected negative consequences on productivity and resource use for forest plantations.

The productivity of plantations of this species has shown a strong correlation with climatic factors, with positive relationships with rainfall and negative relationships with temperature, evaporation, and evapotranspiration [30,31]. Evaluations conducted on juvenile and adult G. arborea trees in the Caribbean region of Colombia showed a decrease in transpiration rates, with tree age and changes in physiological traits in response to water availability [32,33]. Additionally, studies conducted under different site conditions in Latin America have linked the wilting of G. arborea apical meristems to increased drought intensity [34,35,36].

Given these scenarios, it is essential to evaluate the WUE of G. arborea plantations and to identify the main drivers of WUE between stand and site variables. The evaluation of these relationships will allow for the identification and characterization of sites with high WUE and contribute information to the generation of strategies leading to the adaptation of forest management to climate change.

The aims of this study were (i) to evaluate the variability in WUE (in biomass and volume) of G. arborea plantations grown along a gradient of the Magdalena River basin in Colombia, (ii) to investigate the association of stand and site (climatic and soil) variables with WUE (in biomass and volume) of G. arborea plantations, and (iii) to identify the key stand and site variables that exert the greatest influence on the variability in WUE.

2. Materials and Methods

2.1. Study Area

A total of 73 stands of G. arborea were monitored along an altitudinal gradient in the Magdalena River basin, Colombia (Figure 1). These stands are located within the tropical dry forest, as classified by Holdridge’s life zone system [37]. In the lower part of the basin, on the Caribbean plains, there are 54 stands of G. arborea, with an altitude gradient between 13 and 145 m above sea level, an average annual temperature of 28.3 °C, a relative humidity of 78.6%, and an average annual rainfall of 1532 mm yr⁻¹, with a unimodal distribution, with rainfall mainly concentrated in the months of August, September, and October. In the upper part of the basin, in the inter-Andean valleys of the Magdalena River, 19 stands are located; this region is characterized by an elevation ranging from 287 to 388 m above sea level, with an average annual temperature of 27.5 °C, relative humidity of 74%, and an average annual rainfall of 1996 mm yr⁻¹, with a bimodal distribution, with the rainiest months being April and May for the first rainy season and October and November for the second rainy season.

2.2. Permanent Sampling Plots

The study used data from 114 permanent sampling plots (PSPs) established in 73 stands of G. arborea. The PSP network was designed to reflect the distribution of age, stand density, and productivity levels of the plantations. None of the stands showed unusual mortality events, and none were thinned. Circular plots between 500 and 800 m² were used. The age of the stands at the time of plot establishment ranged from 1 to 12 years, with densities between 375 and 1060 trees ha⁻¹. The PSPs were measured on two occasions with an interval of 1 to 4 years during the period between 2007 and 2014. In each plot, diameter (D) measurements were taken for all trees and heights for a sample of trees. Unmeasured heights were estimated using a local height-diameter model for each plot. The dominant height of each plot was calculated as the average height of 100 trees ha⁻¹ with higher diameters.

2.3. Stand Variables

The stand variables used were age, calculated as the difference between the date of measurement and the date of stand establishment, stand density measured through the number of trees (N), basal area (BA), and relative spacing (RS), and productivity measured by dominant height (H), site index (SI), leaf area index (LAI), total volume per hectare (V), total biomass per hectare (TB), and mean annual increment in volume (MAI_V) and biomass (MAI_TB) (

Table 1. Descriptive statistics of stand variables in the last measurement of monitored PSPs of G. arborea in the inter-Andean valleys and the Caribbean plains of the Magdalena River basin.

Stand Variable	Mean	Standard Deviation	Minimum	Maximum
Age (yr)	7.2	3.5	2.2	15.5
Dominant height (H, m)	17.1	5.8	6.3	27.9
Site index (SI, m)	21.9	3.6	13.1	29.0
Density (N, trees ha−1)	660.0	171.4	225.0	1020.0
Basal area (BA, m2 ha−1)	16.7	7.8	2.7	34.2
Relative spacing (RS)	0.3	0.1	0.1	0.6
Leaf area index (LAI, m2 m−2)	6.5	2.7	0.7	11.8
Volume (V, m3 ha−1)	114.3	76.6	6.1	290.8
MAI in volume (MAIV, m3 ha−1 yr−1)	15.6	7.6	1.6	30.8
Total biomass (TB, Mg ha−1)	71.2	39.9	11.1	171.4
MAI in TB (MAITB, Mg ha−1 yr−1)	10.4	4.4	2.8	21.8

). Tree total volume was determined using a taper function fitted by [38] to the species. Site index estimates for a base age of 10 years were made using the model fitted by [39] for G. arborea:

$$\mathrm{S}\mathrm{I}=52.185·{\left(1-{\mathrm{e}}^{-0.038·10}\right)}^{\frac{\ln \left(\frac{\mathrm{H}}{52.185}\right)}{\ln \left(1-{\mathrm{e}}^{-0.038·\mathrm{A}}\right)}}$$

(1)

where SI is the site index (m), H is the dominant height (m), and A is the stand age (yr).

The aboveground (AGB) and belowground biomass (BGB) were predicted using the following equations derived from [40,41] and reported by [42]:

$$\mathrm{A}\mathrm{G}\mathrm{B}=-2.6832+\left(\frac{1156.7223+2.6832}{1+{\mathrm{e}}^{\frac{\left(34.8699-\mathrm{D}\right)}{6.4150}}}\right)$$

(2)

$$\mathrm{B}\mathrm{G}\mathrm{B}=\frac{144.4302}{1+{\mathrm{e}}^{-\frac{\left(\mathrm{D}-30.8573\right)}{6.5314}}}$$

(3)

where ABG and BGB are the aboveground and belowground biomass (kg tree⁻¹), respectively, and D is the diameter at breast height (cm). The total biomass (TB) was determined as the sum of AGB and BGB for each tree in the plot. The leaf area index (LAI) of each plot was calculated from the ratio of the sum of predicted tree leaf area, using a model fitted for the species [43], to the plot size.

2.4. Site Variables

2.4.1. Climatic Variables

Climatic variables were studied from 2007 to 2014, using daily data from 109 climatological stations located along the altitudinal gradient of the Magdalena River basin, obtained from the platform of the Colombian Institute of Hydrology, Meteorology, and Environmental Studies (IDEAM) (http://dhime.ideam.gov.co, accessed on 10 January 2024). For each PSP, daily values of climatic variables were obtained using a geometric mean with all stations around the central point of each plot, using the methodology described by [44]. This methodology uses values weighted by the inverse of the distance and elevation differences between the climatological station and the central point of each PSP. The search radius for applying the geometric mean varied according to the availability of information for each variable of interest (e.g., a radius of 25 km was used for rainfall, but a radius of 200 km was used for net solar radiation).

From the daily climatic information generated for each PSP, values of precipitation (P_m), mean (T_m), minimum (T_mn), and maximum (T_mx) air temperature, relative humidity (RH), wind speed (W_s), and net solar radiation (R_n) were determined (

Table 2. Descriptive statistics of climate variables during the period between 2007 and 2014 in the inter-Andean valleys and the Caribbean plains of the Magdalena River basin.

Variable	Mean	Standard Deviation	Minimum	Maximum
Annual precipitation (Pm, mm yr−1)	1566.8	435.7	854.0	3218.1
Reference evapotranspiration (ET0, mm yr−1)	1508.0	134.1	1292.0	2599.4
Actual evapotranspiration (ETa, mm yr−1)	1313.5	116.8	1125.3	2264.1
Aridity index (AI, mm mm−1)	1.06	0.35	0.44	2.28
Net solar radiation (Rn, MJ m−2 yr−1)	3759.9	232.1	3229.4	4217.3
Mean air temperature (Tm, °C)	28.2	0.6	26.7	29.7
Minimum air temperature (Tmn, °C)	22.7	0.6	21.3	24.1
Maximum air temperature (Tmx, °C)	33.7	0.9	31.5	35.6
Relative humidity (RH, %)	76.9	3.0	65.8	83.6
Vapor pressure deficit (VPD, kPa)	1.15	0.14	0.87	1.56
Wind speed (Ws, m s−1)	1.89	0.64	1.19	8.87

). From these variables, the vapor pressure deficit (VPD) and the reference evapotranspiration (ET₀) and actual evapotranspiration (ET_a) were determined. VPD was calculated as [45]

$$\mathrm{V}\mathrm{P}\mathrm{D}=\left(1-\mathrm{R}\mathrm{H}\right)·0.611·{\mathrm{e}}^{\left(\frac{17.502·{\mathrm{T}}_{\mathrm{m}}}{240.97+{\mathrm{T}}_{\mathrm{m}}}\right)}$$

(4)

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

Reference evapotranspiration was determined using the FAO Penman–Monteith method [46]:

$${\mathrm{E}\mathrm{T}}_0=\frac{0.408·∆·\left({\mathrm{R}}_{\mathrm{n}}-\mathrm{G}\right)+\mathsf{\gamma }·\frac{900}{{\mathrm{T}}_{\mathrm{m}}+273}{\mathrm{W}}_{\mathrm{s}}·\left({\mathrm{e}}_{\mathrm{s}}-{\mathrm{e}}_{\mathrm{a}}\right)}{∆+\mathsf{\gamma }·\left(1+0.34·{\mathrm{W}}_{\mathrm{s}}\right)}$$

(5)

where ET₀ is the reference evapotranspiration (mm d⁻¹), R_n is the net radiation (MJ m⁻² d⁻¹), G is the soil heat flux (MJ m⁻² d⁻¹), (e_s − e_a) is the vapor pressure deficit of the air (kPa), ∆ is the slope of the saturation vapor pressure curve (kPa °C⁻¹), γ is the psychrometric constant (kPa °C⁻¹), T_m is the mean temperature (°C), and Ws is the wind speed (m s⁻¹).

The actual evapotranspiration (ET_a) was calculated as ${\mathrm{E}\mathrm{T}}_{\mathrm{a}}={\mathrm{E}\mathrm{T}}_0·{\mathrm{K}}_{\mathrm{c}}$ [46]. This method accounts for the relationship between reference evapotranspiration (ET₀) and a crop coefficient (K_c) specific to the crop. In this study, a K_c = 0.871 was used, derived from transpiration measurements on trees of the same species obtained via sapflow techniques and ET₀ data. Additionally, an aridity index (AI) was calculated as the ratio between P_m and ET₀.

2.4.2. Soil Variables

Soil samples were collected during plot establishment and analyzed from the top layer to a depth of 30 cm for each of the monitored stands. The soil variables analyzed were texture (Tex), available water capacity (AWC), soil pH, organic matter (OM), phosphorus content (P), potassium content (K), cation exchange capacity (CEC), calcium content (Ca), magnesium content (Mg), and sodium content (Na) (

Table 3. Descriptive statistics of soil variables monitored in G. arborea stands in the inter-Andean valleys and the Caribbean plains of the Magdalena River basin.

Variables	Mean	Standard Deviation	Minimum	Maximum
Soil texture (Tex)	6.0	4.1	1.0	11.0
Available water content (AWC, cm3 cm−3)	0.12	0.02	0.07	0.16
Soil pH	5.93	0.93	4.12	7.60
Organic matter (OM, %)	2.25	1.43	0.02	7.21
Phosphorus (P, ppm)	23.0	19.4	1.7	86.5
Potassium (K, meq 100 g−1)	0.20	0.17	0.03	0.70
Cation exchange capacity (CEC, meq 100 g−1)	23.4	15.1	4.6	60.0
Calcium (Ca, meq 100 g−1)	14.3	9.7	1.0	38.2
Magnesium (Mg, meq 100 g−1)	8.1	8.0	0.2	34.4
Sodium (Na, meq 100 g−1)	0.40	0.50	0.05	2.67

). For statistical analysis, textural classes were classified according to their available water capacity in the following order: 1 = clay; 2 = loamy clay; 3 = clay loam; 4 = sandy clay; 5 = clay loamy sand; 6 = loam; 7 = loamy sand; 8 = sandy clay loam; 9 = silt loam; 10 = sandy loam; 11 = loamy sand; 12 = sand [47].

2.5. Water Use Efficiency

WUE was expressed as the amount of biomass or stem wood produced per unit of water evapotranspired [15,16,17,18,19,20]. To provide a comprehensive understanding of WUE, we calculated it using two distinct measures: WUE_TB, which represents the efficiency in terms of total biomass production, and WUE_V, which represents the efficiency in terms of stem wood volume production. These two measures capture different aspects of forest productivity and resource use. WUE_TB (kg m⁻³ H₂O) reflects the overall growth efficiency, including leaves, branches, stem, and roots, thus providing a holistic view of the ecosystem’s WUE. In contrast, WUE_V (m³ m⁻³ H₂O) focuses specifically on the economically valuable component of the forest, the stem wood, which is critical for timber production and forest management. By evaluating both measures, we can better understand the trade-offs and synergies between overall biomass production and stem wood volume in relation to water use, thereby informing more sustainable forest management practices. WUE was calculated for each PSP using the following relationships:

$${\mathrm{W}\mathrm{U}\mathrm{E}}_{\mathrm{T}\mathrm{B}}=\frac{\mathrm{i}\mathrm{T}\mathrm{B}}{\sum {\mathrm{E}\mathrm{T}}_{\mathrm{a}}}$$

(6)

$${\mathrm{W}\mathrm{U}\mathrm{E}}_{\mathrm{V}}=\frac{\mathrm{i}\mathrm{V}}{\sum {\mathrm{E}\mathrm{T}}_{\mathrm{a}}}$$

(7)

where WUE_TB (kg m⁻³ H₂O, i.e., kg of biomass per m³ of water evapotranspired) and WUE_V (m³ m⁻³ H₂O, i.e., m³ of solid wood per m³ of water evapotranspired) represent the water use efficiency in biomass and volume, respectively, iTB and iV are the increments in total biomass and stem wood volume between the first and last measurement of each PSP, and ET_a is the accumulated actual evapotranspiration between the date of the first and last measurement in each PSP.

2.6. Data Analysis

All analyses were performed using the statistical software R version 4.4.1 [48] and R-studio version 2024.04.2+764 [49]. A t-Student test was performed to examine differences between stand and site variables and WUE (in biomass and volume) for stands located in the lower basin of the Caribbean plains and in the inter-Andean valleys (upper basin) of the Magdalena River. To ensure the assumptions of parametric tests were met, we evaluated normality using Q–Q plots and assessed homogeneity of variances with Levene’s test. Any issues with non-normality and heteroscedasticity were addressed with log transformations when necessary. Furthermore, a bivariate Pearson correlation analysis was performed to investigate the relationship between WUE and stand and site variables.

Additionally, a canonical correlation analysis (CCA) was performed to identify stand and site (climate and soil) variables that drive WUE. The CCA statistical approach was used because it facilitates the study of relationships between sets of multiple dependent and independent variables [50]. CCA allows for the generation of two main outcomes: the ‘‘canonical variables’’, which represent the optimal linear combinations of dependent and independent variables, and the ‘‘canonical correlation’’, which represents the strength of the relationship between them [51,52].

For the CCA analysis, the random vector consisting of all the variables in the study was divided into two sets of variables, the vector x’ = [x₁, x₂, …, x_q₁], consisting of the 33 stand and site variables analyzed, and the vector y’ = [y₁, y₂, …, y_q₂], consisting of WUE_TB and WUE_V.

The approach used in CCA considers that the association between x and y is the highest correlation between two single variables u_k and v_k derived from x and y, where u_k = a’_k x is a linear combination of the variables x, and v_k = b’_k y is a linear combination of the variables y. The number of pairs of canonical variables is limited to the smaller dimension between x’ and y’ (k = 1, …, s) [53], in our case, s = 2, as this was the smallest number of variables between the vector x’ and y’; therefore, the maximum number of pairs of canonical variables was two.

The vectors a’_k and b’_k (k = 1, …, s) of the canonical variables (u_k, v_k), which define the required linear combinations of the variables x and y, are found as the eigenvectors of the matrices E₁(q₁ × q₁) and E₂(q₂ × q₂) defined as

$${\mathbf{E}}_1={\mathbf{R}}_{11}^{-1}{\mathbf{R}}_{12}{\mathbf{R}}_{22}^{-1}{\mathbf{R}}_{21},\hspace{1em}\hspace{1em}{\mathbf{E}}_2={\mathbf{R}}_{22}^{-1}{\mathbf{R}}_{21}{\mathbf{R}}_{11}^{-1}{\mathbf{R}}_{12},$$

(8)

where R₁₁ is the correlation matrix of the variables in set x, R₂₂ is the correlation matrix of the variables in set y, and R₁₂ (= R₂₁) is the q₁ × q₂ matrix of the correlations between the two sets of variables. The canonical correlations R₁, R₂, …, R_s are obtained as the square roots of the non-null eigenvalues of E₁ or E₂. These s canonical correlations express the association between the variables x and y once the correlation within the set is removed [53].

To test the significance of each canonical correlation coefficient, we used the Wilks’ Lambda test, which tests the sequential hypotheses that the k-th and all subsequent canonical correlations are null. Wilks’ lambda values are calculated from the eigenvalues and converted to F-statistics using Rao’s approximation [48].

Because our goal was to determine whether a relationship existed and how strong it was, and then to identify the original variables that exerted greater influence on WUE, in the study, only the first pair of canonical variables (u₁, v₁) (i.e., the strongest canonical correlation) was included in the analysis and discussion of results in the study. Additionally, to aid in the understanding and interpretation of the canonical variables (u₁, v₁), correlation coefficients were calculated between the canonical variables and each of the original variables in their respective set (canonical loadings) and the other set of variables (canonical cross-loadings).

3. Results

3.1. Regional Variations in Stand and Site Variables

Figure 2 compares various stand and site variables between the inter-Andean valleys and the Caribbean plains. Significant regional differences are observed in site quality, stocking, and productivity of G. arborea stands between these two regions. The Caribbean plains exhibit more productive stands, as reflected in higher annual volume production (Figure 2a–c).

Additionally, the studied regions show significant differences in climate and soil conditions. The Caribbean plains experience a significantly drier climate (Figure 2d) and higher maximum air temperatures (Figure 2e). However, the maximum VPD shows no significant differences compared to the inter-Andean valleys (Figure 2f). The soils in the Caribbean plains are characterized by fine textures (Figure 2g), higher available water capacity (Figure 2h), and higher calcium content (Figure 2i).

3.2. WUE Variation

G. arborea showed a mean WUE_TB of 0.93 kg m⁻³ H₂O, with a range of 0.07–2.34 kg m⁻³ H₂O, and a mean WUE_V of 0.0017 m³ m⁻³ H₂O, with a range of 0.0002–0.0040 m³ m⁻³ H₂O. Significant differences in WUE_TB (p < 0.05) and WUE_V (p < 0.05) were found between G. arborea plantations located in two regions of the Magdalena River basin, in the Caribbean plains and the inter-Andean valleys (Figure 3). The Caribbean plains had 30.3% and 37.4% more WUE_TB and WUE_V, respectively, than the inter-Andean valleys (Figure 3). The mean WUE_TB was 1.00 kg m⁻³ H₂O, with a range from 0.14 to 2.34 kg m⁻³ H₂O, for those stands located in the Caribbean plains, and 0.77 kg m⁻³ H₂O, with a range of 0.07 to 1.96 kg m⁻³ H₂O, for the stands located in the inter-Andean valleys (Figure 3a). Considering WUE_V, the mean was 0.0018 m³ m⁻³ H₂O, with a range of 0.0003 to 0.0040 m³ m⁻³ H₂O, for the Caribbean plains, and 0.0013 m³ m⁻³ H₂O, with a range of 0.0002 to 0.0035 m³ m⁻³ H₂O, in the inter-Andean valleys (Figure 3b).

3.3. Relationship between WUE and Stand and Site Variables

Figure 4 presents a matrix of Pearson bivariate correlations between WUE_TB and WUE_V and stand and site variables (climatic and soil variables) for G. arborea plantations. WUE_TB and WUE_V showed moderate and strong significant correlations with stand variables, especially positive correlations with MAI_TB (r = 0.85 for WUE_TB and r = 0.89 for WUE_V) and MAI_V (r = 0.75 for WUE_TB and r = 0.86 for WUE_V), SI (r = 0.65 for WUE_TB and r = 0.74 for WUE_V), and LAI (r = 0.47 for WUE_TB and r = 0.58 for WUE_V), indicating that more productive stands have higher WUE (Figure 4). Additionally, there was a significant moderate positive linear correlation between stand density measures, especially with N (r = 0.43 for WUE_TB and r = 0.51 for WUE_V), and a negative significant correlation with RS (r = −0.39 for WUE_TB and r = −0.52 for WUE_V), indicating that the species requires high stand densities to reach its productive potential. A weak negative significant correlation was observed between age and WUE_TB (r = −0.20), but the correlation was not significant for WUE_V (r = −0.15, p > 0.05), although stands of different ages were included in the study.

A significant weak negative correlation was found between altitude and both WUE_TB (r = −0.21) and WUE_V (r = −0.24). Similarly, a significant moderate inverse relationship was found between WUE_TB and WUE_V with precipitation (r = −0.28 for WUE_TB and r = −0.30 for WUE_V), and a positive relationship was found with the aridity index (r = −0.27 for WUE_TB and r = −0.30 for WUE_V), indicating that sites with higher WUE correspond to lower and generally drier sites (Figure 4). In addition, weak significant positive correlations were found with net solar radiation and air temperatures, especially with mean and maximum air temperatures (Figure 4).

Among the soil variables evaluated, moderate positive relationships were found with AWC (r = 0.43 for WUE_TB and r = 0.46 for WUE_V) and soil calcium content (r = 0.33 for WUE_TB and r = 0.37 for WUE_V). In addition, weak positive correlations were also observed between WUE_TB and WUE_V and CEC and soil potassium content (Figure 4), and a weak negative relationship was also found with soil texture, showing that stands located on sandy soils had the lowest values of WUE_TB and WUE_V.

3.4. Canonical Correlations

Table 4. Canonical correlations between stand and site variables (u_k) and WUE (v_k), with eigenvalues (R²) and their statistical significance (p-value).

Canonical Variable Pairs	Canonical Correlation (R)	Canonical R2	Mult. F	df1	df2	p-Value
(u1, v1)	0.949	0.901	11.862	66	158	<0.001
(u2, v2)	0.845	0.714	6.253	32	80	<0.001

shows the canonical correlations obtained from the canonical correlation analysis, the canonical values of R², and the results of the significance test. These results show the existence of two highly significant canonical correlations, R_{(u₁, v₁)} = 0.949 and R_{(u₂, v₂)} = 0.845, between the vectors of variables consisting of water use efficiency (WUE_TB and WUE_V) and stand and site variables. The results show that 90.1% and 71.4% of the variance in the data is explained by canonical variable pairs 1 and 2, respectively (Table 4).

The most significant canonical correlation from Table 2 (u₁, v₁) indicates that the canonical variables are directly proportional and highly correlated (Figure 5a). The values for each canonical variable were obtained from the standardized data of the original variables in the determined linear combinations for each canonical variable.

Figure 5b shows the canonical loadings between the canonical variable v₁ (cyan) and each of the observed variables composing it, WUE_TB (canonical loading of 0.89) and WUE_V (canonical loading of 0.99). As can be seen in Figure 5b, there is a high correspondence between the values of the canonical variables, as the canonical cross-loadings between u₁ (red) and WUE_TB and WUE_V also show high values of 0.84 and 0.94, respectively. Figure 5c shows the canonical loadings (red) and cross-loadings (cyan) of each of the 33 observed stand and site variables as well as the canonical variables formed by the stand and site variables (u₁ = red) and water use efficiency (v₁ = cyan).

Considering the canonical cross-loadings, the results in Figure 5c indicate that stands with higher rates of WUE (WUE_TB and WUE_V) were positively associated with all stand variables, except age and RS, which showed negative associations. Of these, the stand variables that showed the highest association with the canonical WUE variable (v₁) were (in descending order) MAI_V (canonical cross-loading of 0.88), MAI_TB (0.87), SI (0.76), LAI (0.61), RS (−0.56), N (0.53), BA (0.52), V (0.44), and TB (0.44).

Among the observed site variables, those with the highest positive association with the canonical variable WUE (v₁) were AWC (canonical cross-loading of 0.46), Ca content (0.38), T_mx (0.32), and CEC (0.30). Conversely, the original site variables that showed the strongest negative association, in descending order, were AI (−0.31), Tex (−0.31), and P_m (−0.31).

3.5. Main Driver of WUE

Figure 6 illustrates the relationships between stand and site variables and their mean effect on WUE_TB. Age was categorized into three age classes < 5, 5–10, >10 years. Site quality was categorized into three ranges, as defined by [39]: low site quality stands (SI ≤ 18.6 m), medium site quality stands (SI between 18.6 and 24.1 m), and high site quality stands (SI > 24.1 m). The stocking levels were categorized as overstocked stands (RS ≤ 0.20), fully stocked stands (RS > 0.20 and RS ≤ 0.30), and understocked stands (RS > 0.30). Climate type was derived from the aridity index (AI) and categorized as drier sites (AI ≤ 1) and humid sites (AI > 1).

As shown in Figure 6a, a slight decrease in WUE_TB is observed with increasing stand age, with younger stands showing up to 37% higher WUE_TB compared to older stands. Among stand variables, site quality exhibited the strongest effect on WUE_TB, with increases of up to 53% and 69% when moving from low to medium and high site quality, respectively (Figure 6b). Stand density also significantly impacted WUE_TB, with variations of up to 60% between stand densities of <400 trees ha⁻¹ and >800 trees ha⁻¹ (Figure 6c). Additionally, a decrease of 52% was observed between overstocked and understocked stands (Figure 6d).

Soil texture also had a strong effect on the WUE_TB of G. arborea. Stands on sandy soils had WUE_TB values between 86% and 90% lower than those on loamy and clayey soils, respectively (Figure 6e). Finally, the analysis indicated that drier sites had, on average, 28% higher WUE_TB than humid sites (Figure 6f).

Figure 7 illustrates the relationship between the MAI_TB and WUE_V and WUE_TB. Circles represent PSPs in the inter-Andean valleys, and triangles represent PSPs in the Caribbean plains. Figure 7a shows a positive correlation between MAI_TB and WUE_TB (kg m⁻³ H₂O), with a coefficient of determination of 0.72. Similarly, Figure 7b shows a positive correlation between MAI_TB and WUE_V (m³ m⁻³ H₂O), with a coefficient of determination of 0.79, indicating that higher MAI_TB is associated with greater WUE_V. These results suggest that in both the inter-Andean valleys and the Caribbean plains, higher MAI_TB is positively correlated with greater WUE, both in terms of total biomass and volume.

4. Discussion

Stand-level WUE is a critical indicator of water use strategy in forests and reflects ecophysiological processes related to biomass production and transpiration. This study presents estimates of WUE in G. arborea plantations based on PSPs measurements, examining its relationship with various stand and site variables. Additionally, the silvicultural implications of these findings are discussed.

G. arborea showed lower WUE_TB values compared to other hardwood species such as Eucalyptus benthamii (2.86 kg biomass per m³ H₂O) [54] and Eucalyptus grandis × Eucalyptus urophylla (1.8 to 3.2 kg biomass per m³ H₂O) [23]. However, the WUE_V values of G. arborea are within the ranges reported by [55] for Eucalyptus grandis in South Africa, who reported WUE_V values ranging from 0.0011 to 0.0123 m³ of wood produced per m³ of water consumed. Differences in WUE compared to other species can be attributed to genetic variability and the physiological adaptations that each species develops, which determine their efficiency in water use and capacity to produce biomass. Previous research has reported significant variations in WUE and biomass growth among different Eucalyptus clones in Brazil, attributed to genetic differences in drought tolerance [7]. In that study, the observed WUE ranged from 1.5 g biomass L⁻¹ to 2.3 g biomass L⁻¹, highlighting the impact of genetic variability and physiological traits on WUE and biomass production across species [7]. Another study demonstrated that half-sib families of G. arborea exhibit significant variations in intrinsic WUE, with moderate heritability, emphasizing the potential to genetically select progenies with higher intrinsic WUE without compromising biomass production [42].

The study reports significant differences in WUE_TB and WUE_V between the lower and upper Magdalena River basin regions. The Caribbean plains exhibited 30.3% higher WUE_TB and 37.4% higher WUE_V compared to the inter-Andean valleys. The Caribbean plains demonstrated more productive plantations, indicated by higher site quality, greater annual volume production, and stands managed with higher stockings. These differences suggest that silvicultural practices have been more effectively applied in the Caribbean plains, leading to relatively greater success. Additionally, despite the drier climate, the soils in the Caribbean plains possess fine textures with superior water retention and fertility. Consequently, forest plantations in this region can support more intensive management practices due to their enhanced productivity [39].

Despite the wide range of stand ages included in this study, encompassing both young and rotation-aged stands, the analyses indicated that stand age exerts a weak negative effect on WUE. This result is similar to that reported for Pinus strobus in Canada [56]. The WUE of G. arborea was directly related to stand variables associated with productivity, such as MAI_V, MAI_TB, SI, and LAI. Highly productive stands exhibited higher WUE, supporting the hypothesis that more productive forests are more efficient in resource use [8,57]. The positive relationships with MAI suggest that the WUE of G. arborea is primarily controlled by carbon assimilation rates, which facilitate greater biomass and volume accumulation [58].

High WUE is generally associated with increased carbon assimilation, leading to a positive correlation with growth [59]. Conversely, if higher WUE results from stricter stomatal control, it may inversely correlate with growth. This assertion is further supported by the positive correlation found between WUE and LAI, indicating that an increase in leaf surface area available to intercept sunlight enhances carbon assimilation and, consequently, biomass production [60,61]. Additionally, a higher LAI can significantly alter the stand environment by reducing soil temperature and water evaporation. The shade provided by a higher LAI decreases leaf temperature and transpiration, thereby conserving water and enhancing WUE [62,63].

Basal area and stand density showed a positive relationship with WUE, whereas RS exhibited a negative relationship with WUE. This indicates that higher stand densities, which result in more intense competition for resources, positively affect WUE. Previous studies have demonstrated that G. arborea is highly dependent on stand density for its development [31,64]. Additionally, higher stand densities allow for greater growth per hectare through increased resource competition and earlier site occupancy [24], leading to higher stand growth rates [65,66]. This enhanced growth is likely due to optimized resource use, as denser stands maximize light interception and water utilization, illustrating a balance between competition and resource availability inherent in density-dependent growth.

Within the climate variables, WUE_TB and WUE_V were negatively associated with variables related to water availability, such as P_m and AI. Conversely, air temperatures, particularly T_mx, exhibited a positive association with WUE. These results suggest that G. arborea is well adapted to regions with dry climates [58,64].

Evaluations in temperate and subtropical coniferous and broadleaf forests have shown that ecosystem WUE generally decreases with increasing temperature and precipitation [67]. However, these relationships are not universal across all forest types. For instance, past research has found that WUE was positively correlated with temperature and solar radiation in broadleaf and coniferous forest ecosystems [68]. This indicates that the response of WUE to climatic variables can vary significantly depending on species-specific adaptations and the ecological characteristics of different forest types.

The study highlights the crucial role of soil properties in determining the WUE of G. arborea plantations. A positive relationship was observed between soil texture, AWC, and WUE, indicating that higher soil moisture levels significantly enhance water uptake efficiency. Fine soil textures, which retain more water, improve water availability and tree absorption efficiency [11,12]. Consistent with our findings, G. arborea grown in forest plantations exhibited the best annual growth on loamy soils in Venezuela [69]. In contrast, unfavorable soil textures can restrict water infiltration and root penetration, leading to water stress and reduced WUE, as trees must allocate more resources to cope with water scarcity rather than biomass production [70].

Furthermore, the study found that sufficient soil calcium content is crucial for the ability of G. arborea to withstand water stress. Calcium plays a vital role in maintaining cell turgor and membrane stability, allowing trees to function normally under drought conditions [71,72]. This adaptation enables trees to conserve and use water more efficiently during periods of scarcity. Past research has demonstrated a significant relationship between forest water uptake and soil calcium content, establishing it as a reliable predictor of WUE in forests in northern China [72]. Soils with higher CEC also demonstrated a positive relationship with WUE, as they can retain essential nutrients and moisture more effectively, thereby enhancing plant health and productivity [70].

Results from this study have important implications for selecting sites for new G. arborea plantations. Higher-quality sites, characterized by higher site indices, clay and loamy soil textures with higher AWC, higher calcium content, and higher CEC, are ideal for this species [69,73]. These sites support greater forest productivity and WUE, especially in lower altitude regions [23,31]. Avoiding sites with excessive precipitation and selecting those with favorable air temperatures will contribute to more sustainable and productive plantations [67,74].

Past studies have shown that proper site preparation is essential to ensure optimal soil structure and drainage for G. arborea forest plantations [75]. Practices such as subsoiling can break compacted layers, improve water infiltration, and enhance root penetration [24,75].

Basophilic forest species, such as G. arborea, thrive in soils with higher pH levels, which are typically rich in calcium [73,75]. Research has indicated that G. arborea benefits from the addition of calcium and magnesium to the soil [76,77]. Therefore, applying lime to increase calcium content in calcium-poor soils is essential for this species. These conditions enhance nutrient availability and water retention, supporting better growth and productivity [68,71,72]. Additionally, mineral fertilization can further improve nutrient availability and water retention capacity [77].

Research on G. arborea shows that higher stand densities lead to taller trees and greater stem diameter growth [64]. Studies have similarly demonstrated a significant positive correlation between site productivity (e.g., site index) and stand density for this species [31]. Additionally, thinning studies have shown that less intensive thinning is suitable for ensuring full site occupancy, promoting a higher LAI, more uniform growth, and reduced intraspecific competition in the stands of this species [78]. Therefore, maintaining appropriate stand stockings throughout the rotation length is crucial for the effective management of G. arborea stands.

Implementing these practices will enhance the sustainability and profitability of G. arborea plantations. This management approach is particularly relevant in the context of climate change, where efficient water management is crucial for maintaining the productivity and resilience of forest plantations. Additionally, integrating the selection of appropriate genotypes into management strategies can further enhance the benefits obtained from optimal silvicultural practices, ensuring that plantations reach their maximum productivity potential [8,57].

5. Conclusions

G. arborea plantations exhibited significant variability in WUE, both in terms of biomass and volume, along the Magdalena River basin. The Caribbean plains showed notably higher WUE compared to the inter-Andean valleys, indicating the influence of regional stand and site characteristics on plantation productivity.

The study identified that stand variables, particularly those related to site productivity such as MAI_V, MAI_TB, SI, and LAI, were strongly associated with WUE. Higher productivity stands demonstrated greater WUE, supporting the concept that more productive forests are more efficient in resource use. Additionally, stand stocking played a significant role in maintaining higher productivity and WUE.

Soil variables significantly modulate the WUE of G. arborea forest plantations. Fine soil textures with higher AWC and higher calcium content are correlated with increased WUE. Conversely, climate variables exhibit a minor influence on WUE in G. arborea plantations. Nevertheless, this species exhibits superior adaptation to drier climates compared to humid environments.

Overall, these findings underscore the importance of comprehensive site selection and effective silvicultural practices in maximizing the WUE and overall productivity of G. arborea plantations, especially in the context of climate change, where efficient water management is increasingly critical.