Global Ecological Pattern of Local Leaf Size Diversity

Abstract

Local leaf size diversity (LLSD) is an essential functional indicator of plant biodiversity; however, massive challenges are encountered when quantifying it and decoding its global ecological patterns. To address this limitation, the present study defined a quantitative indicator of LLSD, termed coefficient of variation index (CVI), for the leaf sizes, regardless of plant species, collected in each sampling site. Then, we innovatively derived a set of global CVI values from a published dataset, which was obtained through a meta-analysis of global leaf area samples and their related climate factors. Our macroecological analyses indicate that the CVI values vary across continents and fluctuate with latitude. The global CVI values are predominantly influenced by the mean temperature of the coldest month during the growing season in the negative correlation mode. When two leading climate drivers are considered, the global CVI values are primarily influenced by the mean temperature during growing season and the mean annual sum precipitation. Overall, all of these contributions are pioneering in their implications for characterizing the global distribution and ecological patterns of LLSD and advancing the cutting-edge research domain of leaf functional biodiversity to a new quantitative stage.

1. Introduction

In the burgeoning field of biodiversity research, local leaf size diversity (LLSD) is a subject of extensive interest [1,2,3]. Leaves are plants’ primary functional organs that support fundamental physiological processes, such as photosynthetic carboxylation and catabolic respiration [1]. LLSD, undoubtedly, adds complexity to understanding the mechanisms that underlie these processes. As one of the most prominent functional traits in leaf ecology is leaf size [2], comprehensively investigating the determinants of leaf sizes may provide revolutionary ecological insights. Moreover, a diverse array of leaf sizes tends to exist within any given climate or latitude [3], whereas the ecological mechanisms regulating their individual characteristics remain unclear. Thus, exploring LLSD and its ecological patterns can enhance our understanding of its characteristics and determinants.

However, quantifying LLSD and, subsequently, decoding its ecological patterns has proven to still be a significant challenge. To date, there are efficient solution plans that can measure all the leaves on a plant without causing damage [4], yet few methods exist for quantifying LLSD in local areas. This challenge becomes more pronounced when considering its distribution patterns and ecological roles at the global scale. Notably, even state-of-the-art satellite-based remote sensing technologies, which have been widely used for global mapping, lack the resolution required for accurate measurement of individual leaf sizes [5]. Collectively, measurement across scales—from the individual level to ecosystem level—presents methodological challenges.

Specifically, traditional approaches to investigating leaf size typically involve direct physical measurements of collected samples using basic tools such as rulers, calipers, or graph paper [6,7,8,9,10,11]. These methods, while sometimes being labor-intensive, can provide fundamental metrics including leaf length, width, area, and perimeter—core parameters for assessing LLSD. However, these approaches [6,7,8,9,10,11] present challenges for large-scale studies due to time constraints and potential leaf damage during collection. Particularly for delicate leaves, the process of collection and transport can alter morphological traits, leading to measurement artifacts [12]. Additionally, traditional methods often involve destructive sampling, which may not be desirable for endangered species or long-term monitoring plots where repeated measurements are necessary [13].

Recent technological advancements have revolutionized the measurement of leaf shapes through non-destructive, high-throughput methods that enhance accuracy while reducing processing time. These approaches include portable scanning devices, digital image analysis, and automated classification algorithms that can capture not only size metrics but also shape characteristics [14,15,16,17]. Such approaches typically apply computer vision algorithms to distinguish leaf tissue from background, automatically extracting leaf morphological parameters with minimal human intervention [18,19]. Unfortunately, these techniques also present challenges for large-scale studies.

Additionally, understanding LLSD requires careful consideration of environmental influences and measurement scale, because leaf size varies significantly along climatic gradients [20,21,22] and across organizational levels [23]. Research has demonstrated that environmental factors such as temperature, precipitation, and solar radiation strongly influence leaf size patterns [24]. A comprehensive study across three distinctive plateaus in China found that leaf size was positively correlated with growing-season temperature and precipitation but negatively correlated with ultraviolet radiation, suggesting that environmental filtering plays a more important role than phylogeny in determining leaf size distribution [25]. Similarly, global analyses have revealed that diurnal temperature range significantly influences leaf size patterns, with larger leaves predominating in areas with smaller day-night temperature variations [26].

However, almost no research has been conducted to facilitate the explicit analyses of the ecological determinants and patterns of LLSD in different scenarios, let alone its global scale. The objective of this study was to fill this gap by specifically addressing the issue of quantitatively characterizing LLSD and revealing its global ecological pattern. The validated indicator can serve as a common index for use by the LLSD research community, and the derived ecological models can generate new insights into the macro-scale LLSD-indicated distribution of phytosphere worldwide. The findings have wide implications for advancing research in leaf ecophysiology, plant biogeography, plant community ecology, ecosystem process modeling, biosphere productivity, and even Earth system science.

2. Materials and Methods

2.1. General Scheme

We proposed a quantitative indicator of LLSD—coefficient of variation index (CVI)—for the leaf sizes, regardless of plant species, collected in each sampling site. The decision not to consider plant species, as in [20], is rooted in the difficulty of collecting leaf size samples at the global scale, in contrast to those in intraspecific [22] and interspecific [24] studies. Utilizing this indicator, we systematically examined the macroecological pattern of LLSD based on a published dataset that was generated through a meta-analysis of global leaf size samples and their associated climate factors [20]. The 16 climate factors include mean annual temperature (MAT, °C), mean temperature during growing season (Tgs, °C), mean temperature of the coldest month (T_CM, °C), mean temperature of the coldest month during growing season (T_CMgs, °C), mean temperature of the warmest month (T_WM, °C), mean annual sum precipitation (MAP, mm), mean growing season precipitation (PPTgs, mm), coefficient of variation of monthly precipitation (cvPPT, mm), annual equilibrium moisture index (MIann, mm mm⁻¹), growing season equilibrium moisture index (MIgs, mm mm⁻¹), sum annual equilibrium evapotranspiration (ET_q, mm), sum growing season equilibrium evapotranspiration (ET_qgs, mm), mean daily irradiance, annual (RADann, W m⁻²), mean daily irradiance during growing season (RADgs, W m⁻²), mean annual daytime relative humidity (RHann, %), and mean daytime relative humidity during growth season (RHgs, %).

2.2. LLSD Quantification

The rationale of introducing CV into the characterization of LLSD is that the ranges of leaf sizes may be different across different sampling sites, which, in terms of locations (latitude and longitude), were determined by following the rules in [20]. In this context, CV can serve as a viable solution. Specifically, CV is a statistical measure that quantifies the relative dispersion of data values in a dataset around the mean. It has been utilized in the development of indicators for retrieving plant attributes [27]. For the selected sites, each exhibiting more than nine sampling specimens in terms of leaf sizes, CVI is defined as follows:

$$\mathrm{CVI}={\displaystyle \frac{\sigma }{\mu }},$$

(1)

$$\mu ={\displaystyle \frac{1}{n}}{\displaystyle \sum _{i=1}^n{x}_i},$$

(2)

$$\sigma =\sqrt{{\displaystyle \frac{1}{n-1}}{\displaystyle \sum _{i=1}^n{\left(x_i-\mu \right)}^2}},$$

(3)

where x_i denotes the leaf size value of the ith sample within the considered site that comprises n samples, μ is the mean of those leaf sizes, and σ is the standard deviation of those leaf sizes. As a standardized measure of the dispersion within a probability distribution or a frequency distribution, the CV-based definition of LLSD can quantify the situations of leaf diversities in different environments of plant growths.

For the 16 ancillary climate factors, their statistical distribution modes were derived by examining their histograms. Subsequently, the identified strongly right-skewed climate factors (MAP, PPTgs, MIann, and MIgs) were log-transformed. In contrast, the strongly left-skewed climate factors (MAT and T_CM, pre-scaled by dividing by 10 and 20, respectively) were exponentially transformed. All transformations yielded approximately Gaussian distributions.

2.3. Global Ecological Analysis

To explore potential relationships between CVI and the climatic factors, Pearson’s correlation analysis [28] was employed to assess the strength of associations between these variables. A strong correlation, indicated by a high correlation coefficient (R) and a low p-value, suggests a close relationship, whereas a weak correlation (low R and high p-value) indicates minimal association [28]. Specifically, relationships between CVI and each of the 16 climatic factors were quantified using the Correlation Analysis module in MATLAB (www.matlab.cn, accessed on 3 September 2024). Results from these analyses help elucidate the influence of climatic factors on CVI variation.

Next, dominance analysis [29] was conducted to evaluate the relative contributions of the 16 climatic factors to the spatial distribution of CVI. This method of statistical analysis can determine the relative importance of predictors within a linear regression model [29]. The procedure is designed as follows: CVI is designated as the dependent variable, and the 16 climatic factors as predictors. These variables are individually input into the Dominance Analysis module of DPS software [29] for analysis, consequently without the need to consider the autocorrelation between the predictors. The resulting model coefficients reflect the relative contributions of the corresponding climatic factors, and the comparison of the coefficients identifies the leading climatic factor exerting a primary influence on the global CVI pattern.

To examine nonlinear relationships between CVI and the 16 climatic factors, two univariate regression methods were employed, namely quadratic regression and Gaussian regression [30], and the analyses were performed by using MATLAB. Additionally, bivariate regression was implemented using the poly11 and poly22 tools in MATLAB to investigate interactive effects of paired climatic factors on CVI variations. Note that this explicit scenario relates to the unnecessity of considering the autocorrelation between the 16 climatic factors. At last, all the results provide insights into the macroecological mechanisms through which climate shapes global CVI patterns.

3. Results

3.1. Global Pattern of LLSD

Based on the dataset of leaf size samples extracted from the published dataset [20], the global distribution of LLSD was quantitatively derived, as shown by the global map of 504 CVI sample sites in Figure 1. Across the range of calculated CVI values, most of the sampling sites exhibit relatively lower CVI magnitudes.

Specifically, the CVI values fluctuate with latitude, with decreasing values toward the equator and increasing values toward the poles (Figure 2a). In terms of continental variations, Oceania is characterized by the highest mean CVI values, while Africa demonstrates the lowest mean CVI values (Figure 2b). Oceania displays the largest range of CVI variation, whereas South America shows the smallest range of CVI variability (Figure 2b). Overall, Earth exhibits pronounced spatial heterogeneity in local leaf size distribution.

The histogram of the CVI values for all the considered samples in the present study demonstrates a pronounced right-skewed distribution (Figure 2c). Given this statistical distribution, the original CVI values were log-transformed for both logical and statistical reasons. Log transformation enables a more meaningful interpretation as it operates on a multiplicative rather than arithmetic scale, which is biologically more relevant for size variables. Additionally, it reduces the positive skewness, thereby helping to stabilize the variance in relation to the mean [27]. This approach was implemented to support the subsequent macroecological analyses aimed at uncovering those yet-unknown global determinants of LLSD.

3.2. Global Ecological Pattern of LLSD

For the task of exploring the macroecological relationships between the global CVI values and 16 climate factors, linear regression analyses revealed their correlations (Figure 3a), and dominance analyses quantified the relative contributions of the latter in influencing the former (Figure 3b). Most climate factors show explicit relationships with CVI values. The dominant climate factor is T_CMgs, which has the highest absolute correlation coefficient value (−0.41) and the highest contribution (9.32%).

This suggests that, at the global scale, the primary driving force on CVI is the temperature-related climate factor. However, the total variation in CVI is inadequately explained by the climate factors when the linear regression fitting techniques are used, as indicated by the derived R² values, which show that climate factors account for only 0.66% to 16.82% of the variations in CVI. These results underline the need for applying nonlinear (e.g., polynomial) regression models to more effectively elucidate the potential mechanisms through which climate influences LLSD.

The quadratic and Gaussian regression models applied to each of the 16 climate factors account for relatively higher ratios of variation in the global CVI values than linear models. In these two scenarios, the highest ratios of variation are 16.82% (Figure 4a) and 16.83% (Figure 4b), respectively, both of which are best explained by T_CMgs, as indicated by the 5th and 95th quantile regression fits. In comparison, the Gaussian regression model marginally outperformed the quadratic one. Although CVI decreases with T_CMgs in both models, the minimal improvements in R² values (0.1682 and 0.1683) compared to the best result (0.1681) of linear regression analyses suggest that the global variation in CVI in response to climate drivers is more likely to be nonlinear.

Furthermore, first-order and second-order bivariate regression analyses were made for all possible combinations of the 16 climate factors considered in the present study. Collectively, the fitting performance shows a gradual improvement. In these analyses, the combination of Tgs and logMAP is identified as the most effective predictor pair. The fitted plane and curved surface corresponding to their respective highest R² values are shown in Figure 5a,b (0.1794 and 0.1982, respectively), representing an increase in explanatory power from 17.94% to 19.82%. In other words, the ecological predicting model corresponding to the fitted curved surface provides a slightly better explanation for the variation in CVI values. Consequently, it can be inferred that the varying climatic conditions exert a more complex influence on global LLSD patterns.

Overall, compared to the scenarios based on univariate regression analyses, the combinations of two climate factors demonstrated a greater capacity to explain the variations in CVI values. In addition to verifying the initial hypothesis that LLSD should be influenced by its growth environment, this stepwise improvement facilitated the discovery of new ecological knowledge about LLSD.

4. Discussion and Conclusions

The present study has accomplished the groundbreaking task of developing a primary indicator of LLSD and characterizing its global macroecological patterns. This pioneering research provides new quantitative insights into the global variation and spatial distribution of LLSD. However, the quantitative characterization of CVI and its macroecological patterns may involve various aspects of plant biodiversity [31] and their complex growth environments. Nevertheless, many facets remain unexplored. Relying solely on the coefficient of variation in leafvariation of leaf sizes as an LLSD indicator is insufficient; consequently, subsequent research must refine LLSD indicators, e.g., by introducing new statistical models.

To this end, subsequent work will incorporate as many factors influencing LLSD as possible. It is noted that leaf size represents only one functional trait for characterizing leaf biodiversity. Apart from the ecological drivers examined in the present study, traits characterizing plant leaf diversity span structural, genetic, functional, biogeographic, biochemical, physiological, and phylogenetic dimensions [32,33,34,35,36,37,38,39]. For example, species and phylogenetic lineages have evolved to differ in the way that they acquire and deploy resources, with consequences for their physiological, chemical, and structural attributes [36], and more such information can be disclosed by exploring LLSD. This theoretical prospect suggests that additional factors may influence CVI, underscoring the need to determine the relative weights of LLSD functional traits and their systematic relationships to establish robust leaf biodiversity indicators. Such an approach will enable the development of more effective indicators and potentially a comprehensive framework for characterizing leaf biodiversity.

Although our macroecological analysis of CVI was conducted globally, achieving a comprehensive ecological understanding of LLSD remains quite distant. Enhancing the ecological assessment of leaf biodiversity is thus imperative. Potential improvements include expanding sample sizes, adopting novel measurement techniques, acquiring additional leaf feature parameters, enhancing functional synergy among heterogeneous parameters [40], implementing spatiotemporal big data indexing methods [41], and integrating improved LLSD indices. These advancements will yield a more systematic LLSD distribution map applicable across scales. The ecological mechanisms identified can subsequently inform biomimicry applications, support plant conservation, and deepen insights into natural evolution.

When selecting LLSD measurement methods, researchers must weigh trade-offs among accuracy, efficiency, destructiveness, and technical requirements. Traditional methods offer low technical barriers but are often time-consuming and potentially destructive, while technological approaches provide greater efficiency and digital archiving capabilities but require specialized equipment and expertise. The choice of methods should align with research objectives—intraspecific variation studies tend to require detailed measurements of sample leaves, whereas community-level assessments may prioritize rapid sampling across numerous species [42]. This point also explains the theoretical foundation of this study without regarding plant species.

Future methodological developments should focus on standardizing protocols to facilitate cross-study comparisons. Current variability in measurement techniques may complicate meta-analyses and large-scale synthesis. Developing universal calibration models that incorporate leaf shape parameters, akin to those proposed in global studies of the Montgomery equation [43], offers a promising path toward standardizing leaf area estimation across diverse floras. Similarly, integrating machine learning algorithms for automated leaf classification and measurement [44,45,46,47] could enhance throughput and accuracy while minimizing human bias.

As technology advances, maintaining connections to traditional validation methods and establishing robust quality control protocols will be essential. Ultimately, methodological integration—using high-throughput techniques for large-scale screening and traditional methods for calibration and validation—will provide the most comprehensive understanding of LLSD across ecosystems. This integrated approach will be crucial for addressing fundamental questions in plant ecology and predicting vegetation responses to ongoing environmental change.

In summary, this study’s contributions from a functional synergy [40] and big data [41] perspective have pioneering significance for deciphering global ecological patterns of LLSD and, in a broader sense, advancing the emerging field of leaf functional biodiversity into a quantitative macroecological stage.