1. Introduction
The leaf economics spectrum (LES) is a set of interconnected and synergistic functional traits that quantitatively represent a range of steadily shifting plant resource trade-off strategies [
1,
2]. At the heart of this complex and multifaceted trait network lies leaf mass per area (LMA) [
3]. LMA is the ratio of dry leaf mass to the corresponding leaf area and is a combination of various leaf anatomical characteristics [
4]. It is widely used to estimate leaf area indices [
5,
6] and simulate canopy photosynthesis [
7,
8]. LMA values vary among different tree species, different environmental conditions, and different leaf developmental stages. Thus, accurate and swift measurements of LMA and its dynamic changes are highly important for understanding the growth processes of trees, simulating canopy photosynthesis, and estimating forest productivity.
The crown is the primary organ responsible for photosynthesis in trees. Its intricate three-dimensional structure of branches and leaves affects the local microenvironment of the crown, resulting in spatial differences in leaf functional traits in different areas of the crown [
9,
10,
11]. Studies have shown that the LMA of crowns in various forest types tends to increase from the bottom to the top of the crown [
12,
13,
14]. This increasing pattern of LMA is usually associated with the light gradient through the crown and the water potential gradient from the root to the crown [
15,
16,
17,
18]. In addition to the vertical variation in the crown, LMA also varies significantly across the different developmental stages of trees. Nouvellon et al. [
19] reported that LMA changes drastically over different months, which was also confirmed by Rossatto’s research [
20] on savanna grassland tree species and forest tree species in central Brazil. This variation in LMA can be attributed to differences in temperature, precipitation, and solar radiation among the different periods.
The alteration of LMA typically depends on the leaf dry matter content (LDMC). Plants can acclimatize to diverse circumstances through varied dry matter investments. Consequently, the correlation between traits is strongly associated with the resources and environment in which plants are located. Previous studies have demonstrated that there is a significant correlation between LMA and LDMC, and the association is significantly disparate under diverse environmental conditions [
21,
22]. For trees with a conspicuous canopy structure, the gradient discrepancy of the canopy microenvironment in the vertical direction [
23] will cause a shift in the correlation between different canopy depths. Zhang [
24] studied the vertical changes in the LMA and LDMC of
Pinus yunnanensis with canopy height. The results showed that different LMA and LDMC values exhibited distinct changes with canopy height. Tian et al. [
25] reached the same conclusion. Additionally, alterations in the temperature, precipitation, and solar radiation of plants during different growing seasons will make the environment in which leaves are situated highly heterogeneous. This will lead to different changes in LMA and LDMC with canopy height at different growth stages.
Due to the limitations of leaf area measurement technology [
26], it is very difficult to measure LMA. Currently, LMA is measured by retaining some leaves of the analysed tree and establishing a single tree leaf biomass model based on the leaf biomass of the analysed tree and its diameter at breast height [
27], or by directly calculating LMA from the measured total leaf area and leaf dry mass. However, for coniferous plants, these methods require considerable manpower and material resources [
28,
29] because of their three-dimensional structure and large number of leaves. To address this issue, an increasing number of researchers are estimating LMA by establishing regression models between LMA and plant traits, leaf morphology, or environmental conditions, such as leaf length, leaf width [
30,
31], branch height [
32,
33], and LDMC [
34,
35,
36]. As the correlation between LMA and the vertical direction of a tree crown is significant, variables related to vertical height, such as branch height, depth into the crown, and relative depth into the crown, are often used as the main fitting factors. LDMC is also a common fitting factor, and many studies have discussed the relationship between these two parameters. Typical linear models or nonlinear models are used to fit LMA based on LDMC [
37]. Peng’s research [
38] showed that the LMA of Chinese fir can be estimated by the LDMC and that the model meets the estimation requirements. Therefore, it is important to establish a simple and accurate LMA prediction model for the purpose of simplifying canopy models. Determining LDMC and RDINC is simpler than determining LMA, and both methods meet the estimation requirements of LMA. However, previous studies have taken only leaf samples at one particular point in time or at a specific canopy position. It is yet to be determined whether different leaf development stages and depths of the canopy have an effect on LMA prediction models. A few studies have examined which vertical factor or LDMC can most accurately predict LMA. Furthermore, whether leaf development time can be used as a single factor to predict LMA has not been tested.
Larix principis-rupprechtii, one of the most widely planted trees in North China, is characterized by strong light tolerance, rapid growth, and longevity and is a valuable native species. This study aimed to clarify the variation patterns of the LMA of needles in different vertical layers and at various leaf development stages and to reveal the main factors influencing the LMA. Finally, the best prediction model of LMA for L. principis-rupprechtii plantations was established, which can be used to simulate crown photosynthesis and estimate regional primary productivity.
4. Materials and Methods
4.1. Site Description
The study site was located at the Saihanba Forest Farm, Hebei Province, northern China (42°02′~42°36′ N, 116°51′~117°39′ E), at an altitude of 1010~1939.9 m. The main soil type is sand. The climate type is a typical temperate continental monsoon climate, with an annual average temperature of −1.3 °C, an extreme minimum temperature of −43.3 °C, an extreme maximum temperature of 33.4 °C, an annual average snow cover of 7 months, an annual average precipitation of 460 mm, an average annual frost-free period of 64 days, and an annual average windy day of 53 days. The main tree species are L. principis-rupprechtii, Picea asperata, Betula platyphylla, P. sylvestris var. Mongolica sylvestris, etc. The forest coverage rate was 82%, and the total forest stock was 5.025 million m3.
4.2. Sample Selection
In this research, five sample plots (20 m × 30 m) were established within a 17-year-old, pure
L. principis-rupprechtii plantation in the same habitat as the Saihanba Forest Farm. All trees with a diameter at breast height (DBH) larger than 5 cm in the sample plot were measured, and factors such as DBH, tree height (H), crown width (CW), and relative coordinates (X,Y) were included. Subsequently, five sample trees with a DBH similar to the quadratic mean diameter (Dg), representing the average state of each sample plot, were chosen. The basic information about the sample plots and sample trees is displayed in
Table 4.
4.3. Measurement of LMA
For a single tree, the crown was divided into various whorls by the whorls of branches from top to bottom. In each group, 3–4 healthy clusters were chosen as samples. The relative depth into the crown (the ratio of depth into the crown to crown length, RDINC) of every sample cluster was recorded, and then the samples were immediately removed and taken back to the laboratory for scanning and weighed immediately to determine fresh weight (WF). The scanned needle samples were dried to a constant weight of 85 °C and weighed (WD). The images were analysed using image analysis software (Image-Pro Plus 6.0, Media Cybernetics, Inc., Bethesda, MD, USA), resulting in a projected leaf area (LA, m
2). The LMA and LDMC of each cluster of needle samples were then calculated. The data were collected every half month during the growing phases (approximately from 1 June to 15 September) in 2017. The basic statistics were listed in
Table 5.
The LMA and LDMC of each cluster of needle samples were calculated as follows:
where
i represents the sample whorls,
j represents the date of the measurement, WD represents the dry weight, and WF represents the fresh weight.
4.4. Model Descriptions
4.4.1. Basic Model Selection
Based on previous research and the scatter plot distribution and correlation between LMA and LDMC, as well as the spatial position and leaf growth phase (LGP) obtained in this study, a basic model was established with RDINC, date of year (DOY), and LDMC as independent variables (see
Table 6).
4.4.2. Discrete Analysis and Reparameterization
According to prior research, the relationships between LMA and LDMC, leaf spatial position, and leaf growth phase are evident. Thus, to improve the accuracy of the model, it is necessary to further discretize the LMA data for feature analysis. Models 1–4 simulated the LMA in 9 groups, with intervals of 0.1, based on RDINC. Model 5 simulated the LMA in 6 groups, with groupings of 150, 165, 180, 195, 210, and 225, based on the DOY. Models 6–7 simulated the LMA on RDINC and DOY. Then, reparameterization was conducted in Models 1–7, according to the correlations between the parameters and RDINC and DOY, to form 6 new models with multiple independent variables (RDINC, DOY, and LDMC). Finally, the optimal LMA model for L. principis-rupprechtii was selected based on its goodness-of-fit (Equations (3)–(5)) and validation performance (Equations (6)–(8)). The LMA prediction model was then established through parameterization.
4.4.3. Model Assessment and Validation
When fitting the model, 75% of the data were randomly chosen for model fitting, and 25% were used for model validation (
Table 7). The indicators chosen to assess the model’s goodness-of-fit are the adjusted determination coefficient (
Ra2), root mean square error (RMSE), and Akaike information criterion (AIC). The indicators for validation are the mean error (ME), absolute mean error (MAE), and fit index (FI). The formulas for computing each index are as follows:
where yi is the observed value,
i is the average of the observed values, ŷi is the predicted value, n is the number of samples, and
p is the number of parameters.
4.5. Data Analysis
A two-factor ANOVA was used to examine whether the LMA of L. principis-rupprechtii significantly differed between different trees and different ring whorls at various growth and developmental stages. Furthermore, Pearson correlation coefficients among LMA, LDMC, RDINC, and DOY were computed, and the correlations between LMA and other factors were analysed. The LMA and LDMC of different canopies and different growth periods were fitted by standardized principal axis analysis. Tests were conducted to determine whether there was a significant difference in slope among the different groups and to ascertain whether different canopy depths and growth periods had a significant influence on the correlation between LMA and LDMC.
Microsoft Excel 2010 was used to collate the data of the study; descriptive statistical analysis was performed using SPSS 24; model fitting was completed by the nls package in R 4.0.5; standardized principal axis analysis was completed by the smatr package in R 4.0.5; and diagrams were drawn with the ggplot2 package in R 4.0.5 and Origin 2019.