Modelling Climate Using Leaves of Nothofagus cunninghamii—Overcoming Confounding Factors

Abstract

Fossil leaf anatomy has previously been used as a proxy for paleoclimate. However, the exposure of leaves to sun or shade during their growth can lead to morphotype differences that confound the interpretation of fossil leaf anatomy in relation to climate and prevent reliable paleoclimate reconstruction. This work aims to model the differences in leaf anatomy that are due to various climatic drivers and differences attributable to sun or shade positions, using Nothofagus cunninghamii as the model species. Leaves from the sun and shade parts of three trees have been sampled from each of 11 sites in Victoria and Tasmania, Australia. The gross morphological and cuticular features have been scored and modelled with climate data from the sites. Random forest models can accurately predict Nothofagus cunninghamii contemporary climatic conditions of the spring temperature and summer rainfall based on leaf anatomical measurements. Leaf area, stomatal density and epidermal cell density are the most accurate predictors of whether a leaf grew in the sun or shade. Leaf area is also the strongest predictor of the maximum and minimum spring temperatures and rainfall. The models have implications for the use of fossilised leaves in paleoclimate reconstruction. The models we have built can be used to effectively predict whether a fossil leaf was from a sun or shade position on the tree and thus enable more reliable inference of paleoclimate by removing the confounding issues of variable leaf anatomy due to sun exposure during growth. Finally, these models could conceivably be used to make predictions of past paleoclimatic conditions provided suitable training and validation data on climatic conditions are available.

1. Introduction

Leaves are valuable indicators of climate. Many parts of a leaf grow and change in response to climatic conditions and can therefore be used to reconstruct climatic conditions during their growth and development. Many studies have shown that we can quantify the response of leaves of extant plant species to climatic conditions and use this knowledge to infer how closely related species responded to climate change in the past, based on the anatomy of fossil leaves. Leaf measures that have been analysed in the past include leaf area and shape as a response to the mean annual temperature [1]; carbon isotope fractionation as a response to the mean annual precipitation [2]; stomatal size and density as responses to light availability and the temperature [3]; leaf size as an indicator of the growth temperature (although this has been criticised for over-use [4]); many novel physiognomic characters such as the tooth area and number of teeth on leaf margins as responses to the mean annual temperature [5]; and a combination of leaf characters to reconstruct how open the vegetation was [4].

A potential confounding factor affecting paleoclimate reconstructions based on paleobotanical measures is the morphological differences that exist on leaves from the same plant due to their position in the canopy. Leaf position often results in sun and shade leaf phenotypes, a well-known phenomenon in both the extant [6] and paleobotanical records [7]. For example, Keenan and Niinemets [6] identified confounding biases in global leaf trait data, with databases being based on sun leaves for the purpose of standardisation. Maslova et al. [7] noted problems of the misidentification of fossils due to large intra-specific differences in sun and shade leaves. Sun leaves tend to be thicker, have smaller areas with dense and small stomata and have dense venation compared with shade leaves [8,9,10]. Thus, the differing morphologies of sun and shade leaves need to be considered when leaves are used to reconstruct past climates.

The stomatal and epidermal cell density and leaf area have previously been shown to differ between sun and shade morphotypes in other species. For example, the stomatal density was found to be lower in shade leaves than sun leaves based on a herbarium study of multiple tree species with a 190-year time span [11]. Additionally, both the stomatal and epidermal cell density on the leaves of Alnus glutinosa show significant differences between sun and shade leaf morphotypes, with more epidermal and stomatal cells in sun leaves than shade leaves [12]. Likewise, for Toona ciliata, the leaves are smaller with a larger stomatal density when grown in high- versus low-light conditions [13]. This lends confidence to the generality of our modelling approach, which uses these three measurements together to determine whether a leaf was grown in a sun-exposed or shaded part of the tree canopy.

Here, we develop an approach, using a machine learning method, which directly accounts for leaf traits when modelling the relationship between leaf morphology and temperature. In doing so, we provide a potential new approach for overcoming an important source of uncertainty in paleoclimate reconstructions using fossil leaves. We applied and tested this approach using Nothofagus cunninghamii (Hook.) Oerst. (Nothofagaceae) leaves and contemporary climate parameters.

2. Materials and Methods

2.1. Study Species

Nothofagus cunninghamii (Nothofagaceae) was chosen as the study species for several reasons:

• The species and its close relatives are found in Quaternary deposits through southeastern Australia, where paleoclimates are currently unresolved. Thus, determining the responses of N. cunninghamii to current climates can assist in hindcasting paleoclimatic conditions and understanding the species response to past climate change.

• Living Nothofagus cunninghamii occurs over a relatively large latitudinal and elevation gradient (Figure 1; elevation range = 112–1240 m), and the species has been found to be able to grow within a mean annual temperature range of 5 °C. This capacity to persist in a wide range of temperature environments makes N. cunninghamii ideal for modelling climate responses [14,15,16].

• The leaves of N. cunninghamii are evenly covered with stomata (including all veins except the mid-vein), and the stomata and epidermal cells are easily identified and counted, allowing their density to be measured relatively quickly and with accuracy. This is an important consideration, particularly when using fossil leaves for paleoclimatic reconstructions. In contrast, other species in the genus have stomata distributed in areoles that impact measures such as stomatal and epidermal cell density, especially in fragmentary fossil leaf material.

• The Cenozoic macrofossil record in eastern Australia contains many examples of leaves that appear to be intermediate in morphology between the living species N. cunninghamii and N. moorei, which is distributed further north in Australia, around the New South Wales and Queensland border [17,18]. This offers the potential to extend the range over which leaf anatomical data can be extended back into the Cenozoic.

2.2. Measurements and Field Collection

Leaf measurements were carefully chosen to reflect methods that would need to be applied to fossil Nothofagus leaves. For example, fossil Nothofagus leaves are often preserved as mummifications in southeastern Australia, but despite their high level of preservation they are often very delicate and leaf thickness cannot be measured, though it is usually possible to measure cuticular features and sometimes, especially within Australia, the leaf area [19]. The preservation of Nothofagus leaves as fossils outside of southeastern Australia is usually far lower in quality and it is more difficult to measure even the most basic traits. The measured and calculated variables chosen are listed in Table 1 and include the stomatal density (number of stomata mm⁻²), epidermal cell density (number of epidermal cells mm⁻²), epidermal cell area (µm²), epidermal cell perimeter (µm), undulation index (Equation (1), dimensionless), stomatal index (Equation (2), %) and leaf area (mm²).

(1)$\mathrm{UI}=\frac{{\mathrm{C}}_{\mathrm{e}}}{{\mathrm{C}}_{\mathrm{o}}}=\frac{{\mathrm{C}}_{\mathrm{e}}}{2\times \pi \times \sqrt{\frac{{\mathrm{A}}_{\mathrm{e}}}{\pi }}}$

where UI is the undulation index, C_e is the cell circumference (µm), C_o is the circumference of a circle with the same area as the cell, A_e is the cell area. The undulation index is defined by Kürschner, 1997 [19], as the ratio of the measured epidermal cell circumference to the circumference of a circle with the same area as the cell. Essentially, the higher the undulation index, the more sinuous the cell walls, the lower the undulation index, and the straighter the cell walls. The undulation index varies with the degree of cell wall undulation but not the cell area [20].

(2)$\mathrm{SI}=\frac{\mathrm{SD}}{(\mathrm{ED}+\mathrm{SD})}\times 100$

where SI is the stomatal index, SD is the stomatal density (stomata mm⁻²) and ED is the epidermal cell density (epidermal cells mm⁻²). The stomatal index is the proportion of epidermal cells that have transformed into stomata. It is independent of the epidermal cell size [21].

The leaf collection encompassed the entire natural latitudinal range of Nothofagus cunninghamii (collection sites are shown in Figure 1). The field collection design is shown in Figure 2. The Victorian specimens were collected using permit number 10007653 from Parks Victoria and the Tasmanian specimens were collected under permit number FL17019 from the Department of Primary Industries, Parks, Water and Environment.

The leaves were collected from the sun-exposed and shaded parts of trees. Often, the sun-exposed parts of the tree were the outward limbs of a group of trees (for example, at the roadside) and the shade-exposed parts were within the canopy. Once collected, the leaves were placed in herbarium presses. Five leaves of each of the sun and shade morphotypes were placed on a flatbed scanner (HP Scanjet 200 Flatbed Scanner; CanoScan Toolbox version 4.9.3.2 toolpak (X for Mac OSX) flatbed scanner) to create images from which to measure the leaf area using FIJI Is Just ImageJ [FIJI; [22,23]].

The cuticles were removed from the sun and shade leaves by placing the leaves into 80% ethanol overnight then transferring them to a solution of two parts 35% hydrogen peroxide to one part 80% ethanol. The leaves were then slowly heated until translucent and then removed from the heat and transferred to water. Fine brushes were used to remove any remaining debris from the cuticle, and then cuticles were stained using crystal violet 0.05% w/v and permanently mounted onto slides using a drop of warmed glycerine jelly [24]. Entire leaves were used for the preparation of the cuticles because N. cunninghamii leaves are quite small (often <1 cm long) and it was easiest to avoid trying to cut the leaves prior to cuticle preparation.

For each leaf cuticle, three images were obtained at 20x magnification using an AX70 microscope (Olympus, Australia) mounted with a UC50 camera (Olympus, Australia). Examples of these images are shown in Figure 3.

2.3. Modelling

To generate predictions of the climate from the leaves of Nothofagus cunninghamii, we used a readily available machine learning approach, random forests [RF; [25]], which we implemented in the ‘randomForest’ package for R [26]. We opted to use random forests [RF; [25]] because the models are easily trained, highly accurate and have an in-built variable selection method. Random forests can deal with correlated variables [25]. Therefore, we modelled a full suite of predictors without screening for collinearity or performing variable selection or reduction. We used eight predictor variables related to leaf morphology (Table 1) and a binary variable (describing whether a leaf was found in the sun or shade) to generate regression-based estimates of contemporary climatic conditions for leaf development. These climate predictors were the maximum and minimum spring (mean values from September–November) temperature (°C), and total summer (summed over December–February) rainfall (mm). The climate data were extracted for each of the sample sites (Figure 1) from the Terrestrial Ecosystem Research Network (TERN) e-mast facility (http://portal.tern.org.au accessed on 27 May 2019) and represent a 30-year average (1976–2005). This temporal period coincides with the time period during which leaves were sampled. A classification model was also built to test whether a leaf could be correctly allocated as either a sun or shade leaf based on three measures of leaf morphology—stomatal density, epidermal cell density and epidermal cell area (see Results and Table 2).

The calculation of the variable importance was performed on both climate prediction and morphotype classification models to determine the parameters with the greatest influence on model predictions. For the random forest regression-based (RF) models of climate as a function of leaf variables, this was measured using unscaled permutation importance [27,28]. For the sun and shade leaf classification (RF) models, this was performed using the Gini index [25].

To train and assess the model performance of the regression models, we conducted 10 repeats of 10-fold cross validation, whereby the data were randomly split (without replacement) into 90/10 training/test sets. For our classification model, we conducted 10 repeats of 2-fold cross validation, whereby the data were randomly split (without replacement) into 50/50 training/test set. The models were built on the training dataset and tested on the holdout test dataset. We used 2-fold cross validation for our classification model to maintain the proportions of our binary target in the training and testing splits. All the regression models were stratified using the sun/shade variable to maintain class proportions when building and testing the models. Furthermore, we optimised the number of variables selected at each split within the trees of the RF models and selected the model that showed the best performance using the one standard error (SE) rule [29]. Accumulated local effects [ALE; [30]] plots were then built for each of the models to assess the effect of the predictor variables on average model predictions. ALE plots are preferable to traditional partial-dependence plots as they are unbiased when predictor variables are (potentially) correlated [30].

3. Results

Nothofagus cunninghamii sun leaves have more stomata and epidermal cells per unit area with a smaller perimeter and area than their shade-grown counterparts (Table 1). The undulation index and leaf area of N. cunninghamii sun leaves are smaller than for shade leaves, whereas the stomatal index is higher.

The regression models accurately predicted the maximum spring temperature (R² = 0.97 [±0.01]; RMSE = 0.32 [±0.02]), minimum spring temperature (R² = 0.95 [±0.01]; RMSE = 0.28 [±0.03]) and total summer rainfall (R² = 0.98 [±0.01]; RMSE = 12.23 [±1.10]) based on the leaf trait variables. The variable importance for the RF regression-based model showed that the leaf area (mm²) was the strongest predictor of the spring temperature (both minimum and maximum) followed by the epidermal cell density (epidermal cells mm⁻²; Table 2). The epidermal cell area (µm²) and stomatal index (%) were also important predictors (variable importance rank ≤ 4th; Table 2). The most important variable for predicting rainfall was the leaf area (mm²), followed by whether the leaf was from the sun or shade, the stomatal index (%) and the epidermal cell density (epidermal cells mm⁻²; Table 2).

The accumulated local effects plots (Figure 4) show that the leaf area increased at sites where the minimum and maximum spring temperatures were higher, and at sites where the summer rainfall was lower than that observed across most of the collection sites. A reduction in the stomatal index occurred at sites with cooler minimum spring temperatures whilst, conversely, sites that had higher maximum spring temperatures and higher summer rainfall saw an increase in the stomatal index.

The random forest classification model was able to accurately predict whether a leaf grew in a sun or shade environment based on the leaf morphological measurements, with a cross-validated AUC score of 0.99 (±0.001; sensitivity = 0.98, specificity = 0.99). An assessment of the variable importance, measured as the mean decrease in the Gini index, suggested the leaf area was the most important variable (mean decrease in Gini index = 833.8), followed by the stomatal density (mean decrease in Gini index = 660.5) and then the epidermal cell density (mean decrease in Gini index = 620.8). The probability of a leaf being classified as a sun morphotype decreased with an increase in the leaf area up to ~100 mm² when a leaf had a 100% chance of being classified as shade (Figure 5A). The stomatal and epidermal cell density displayed similar responses with an increase in the probability of being classified as a sun leaf as their respective values increased (Figure 5B,C).

4. Discussion

Our new approach, based on N. cunninghamii leaf traits, accurately predicts whether a leaf grew in the sun or shade. Furthermore, it does well at predicting the relationship between some very simple leaf morphological measurements and the spring temperature and, to a lesser extent, the summer rainfall, across a latitude and elevation gradient. This provides us with high confidence that the fossilised leaves of N. cunninghamii that preserve the morphological traits used as predictors in our models will be useful for reconstructing past climates across the present and past range of the species.

Our results suggest that it is possible to reconstruct past temperature and, at least to some extent, rainfall using predictive models based on leaf traits from fossilised leaves, whilst removing the confounding factor of sun or shade leaf morphotypes. Specifically, we found that the traits of the leaf area and the stomatal and epidermal cell density were consistently good predictors in our models using extant species that are also present in the fossil record. Such models can theoretically be used on fossilised leaves, provided that the leaves can be accurately identified to species level. This is the case for N. cunninghamii at several Quaternary localities (see, for example, Figure 9 in [31]—other Quaternary records of N. cunninghamii leaves are unpublished). Predictive models based on the leaf area and the stomatal and epidermal cell density, that account for whether individual leaves are a sun or shade morphotype, are likely to be able to reconstruct past temperatures and, at least to some extent, rainfall using predictive models.

Predictions of the climatic tolerance of plants through leaf analysis are likely to be influenced by whether the morphological traits of sun or shade leaves are used as proxies. For example, any climate modelling based on fossil leaves should ensure that fossil sun leaves are compared with extant sun leaves rather than extant shade leaves. Light is a strong driver of leaf morphology [32], and its effects need to be accounted for prior to modelling climate conditions, otherwise the confounding of leaf sun exposure with climate conditions may occur. A modern example of where light exposure confounds leaf trait interpretation is the CANTRIP database [database of CANopy TRaIt Plasticity, 6]. CANTRIP data include the sun and shade leaf morphology and physiological, chemical and ecological measurements for 830 species. While this database is highly valuable for extant species, the CANTRIP database cannot be applied to fossil leaves, because its measurements cannot be made on fossil leaves, except for leaves from deciduous trees [33]. The purpose of CANTRIP is to highlight the importance of the role of light availability in interpreting the leaf economics spectrum [6]. This purpose is vital for understanding plant adaptation to certain environments and making recommendations for conservation and revegetation.

In this paper, we asked whether leaves from sun or shade positions can be used to predict climatic conditions, after accounting for light exposure. The answer to this question is yes. Nothofagus cunninghamii leaves decreased in leaf area with an increase in summer rainfall, and this relationship is strong (Table 2). The leaf area was found to increase with increasing mean annual precipitation in a meta-analysis of over 1900 species [1]. We chose to focus on summer precipitation as this variable is likely to be a greater limiting factor to leaf maturation than the mean annual precipitation for N. cunninghamii, which grows in a region where summer is the driest season [34]. Thus, N. cunninghamii leaves may still follow the trend concluded by Peppe et al. [1] related to the mean annual rainfall. Future research could explicitly test this, but this hypothesis is not within the scope of this paper.

The two-step morphologically based modelling approach performed in this study is a cutting-edge solution to a problem that will enable leaf anatomy to be used as a paleoclimate proxy in a highly robust manner. The first step allows climatic conditions to be reconstructed and the second determines whether a leaf has been exposed to high levels of light. In the future, it is feasible to develop a method for using this model to predict whether fossil leaves were likely to be from sun or shade locations on the tree, and thus reduce the confounding effect of climate and the effect of sun exposure on leaf morphology. This paper demonstrates proof of concept for our approach, by showing that we can successfully estimate climatic variables, whilst accounting for the differences between sun and shade leaves from an extant Nothofagus species based solely on leaf morphology. However, we acknowledge that N. cunninghamii has several features that make it ideal for this kind of research and the challenge will be to extend this research to other species to determine whether similar trends are observed there. For example, future challenges for this approach include extending it to other species where stomata are less evenly spread on the leaf surface and applying it to older fossil species, where extant affinities may be well-understood, but the fossil species no longer exists in the extant vegetation. This represents a major research challenge. Even within the set of fossil leaves that are related to living N. cunninghamii, there are some obvious challenges to applying the results presented here. For example, while Quaternary N. cunninghamii leaves have robust cuticles that relate them directly to the living species, older fossils (Late Eocene–Miocene) often have quite fragile and fragmentary cuticles which may be the result of the sedimentary conditions where they occur, more equable climates that required less robust cuticles, or perhaps this indicates that these older fossils were from winter deciduous plants, which is certainly possible given that Australia was much further south when these fossils were growing and in light environments where increased winter darkness may have made winter deciduousness a more competitive strategy [18].

Author Contributions

K.E.H.—designed project, carried out field work, prepared cuticles, scored cuticular traits and leaf area, obtained climate data, wrote manuscript. S.C.B.—co-designed statistical analysis and modelling, carried these out and wrote the methods section for this part, edited manuscript. A.J.—co-designed statistical analysis and modelling, edited manuscript. D.F.—co-designed statistical analysis and modelling, edited manuscript. R.S.H.—co-designed project, co-wrote manuscript. All authors have read and agreed to the published version of the manuscript.

Data Availability Statement

Data are not publicly available, although the data may be made available upon request from the corresponding author.

Acknowledgments

We thank Fiona McQueen for lab assistance and preliminary data exploration, Josh Edwards and Kimberly Edwards for lab assistance, Matthew DeBoo for field assistance and Greg Jordan and Thomas Baker for field collections.

Conflicts of Interest

The authors declare no conflict of interest.

Footnotes

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Figures and Tables

Figure 1. Map of field collection sites of Nothofagus cunninghamii sun and shade leaves in Victoria and Tasmania, Australia. Each site is marked with a white circle.

Figure 2. Nested design for leaf sampling.

Figure 3. Cuticle images of (A) shade and (B) sun leaf morphotypes of Nothofagus cunninghamii. Scale bar is 100 µm. Note the larger cell size and highly undulating cell walls in the shade morphotype when compared with the sun counterpart. Stomata are obviously fewer in shade leaves. Three large trichomes are present on the sun leaf. These varied in density across the collection, but they did not interfere with measurements.

Figure 4. Accumulated local effects plots depicting the effects of (A) minimum spring temperature (°C), (B) maximum spring temperature (°C) and (C) total summer rainfall (mm) on (A–C) leaf area (mm2); (D) minimum spring temperature (°C), (E) maximum spring temperature (°C) and (F) total summer rainfall (mm) on (D–F) stomatal index (%); and (G) minimum spring temperature (°C), (H) maximum spring temperature (°C) and (I) total summer rainfall (mm) on (G–I) undulation index. The x-axis is the leaf measurement. The y-axis describes how likely a leaf is to be sun or shade; if the plot is above zero on the y-axis, then the leaf is from a shade position, and if the line is below zero, then the leaf is from a sun position. For example, in A, leaves with a leaf area less than 100 mm2 are sun leaves and those with a leaf area more than 100 mm2 are shade leaves.

Figure 5. Probability of a leaf being classified as a ‘sun’ leaf (y-axis) as a function of the variables on the x-axis. (A) Leaf area (mm2). Note the ticks along the x-axis are individual measurements. (B) Stomatal density of the leaf (stomata mm−2). (C) Epidermal cell density (epidermal cells mm−2). Probabilities more than zero define sun leaves and those below zero define shade leaves.

Table 1

Mean ± standard deviation for all leaf variables.

Variable	Sun	Shade
Stomatal density (stomata mm−2)	396 ± 1	281 ± 9
Epidermal cell density (epidermal cells mm−2)	5788.6 ± 1255.4	4537.6 ± 1036.8
Epidermal cell area (µm2)	197.5 ± 94.8	286.1 ± 142.0
Epidermal cell perimeter (µm)	63.3 ± 22.7	86.8 ± 32.7
Undulation index	1.3 ± 0.2	1.5 ± 0.2
Leaf area (mm2)	108.6 ± 43.3	114.5 ± 42.2
Stomatal index (%)	6.4 ± 1.4	5.8 ± 1.4

Table 2

Variable importance scores (rescaled to between 0 and 1) of each measured or calculated leaf anatomy variable in predicting maximum and minimum spring temperatures (defined as mean of maximum or minimum temperatures in September, October and November) and total summer rainfall (defined as total rainfall for the months of December, January and February).

Variable	Maximum Spring Temperature (°C)	Minimum Spring Temperature (°C)	Total Summer Rainfall (mm)
Leaf area (mm2)	1	1	1
Epidermal cell density (epidermal cells mm−2)	0.29	0.31	0.16
Epidermal cell area (µm2)	0.29	0.26	0.13
Stomatal index (%)	0.20	0.31	0.21
Stomatal density (stomata mm−2)	0.16	0.15	0.11
Sun or shade leaf	0.14	0.08	0.32
Undulation index	0.06	0.09	0.07
Epidermal cell perimeter (µm)	0	0	0

References

1. Peppe, D.J.; Royer, D.L.; Cariglino, B.; Oliver, S.Y.; Newman, S.; Leight, E.; Enikolopov, G.; Fernandez-Burgos, M.; Herrera, F.; Adams, J.M. et al. Sensitivity of leaf size and shape to climate: Global patterns and paleoclimatic applications. New Phytol.; 2011; 190, pp. 724-739. [DOI: https://dx.doi.org/10.1111/j.1469-8137.2010.03615.x]

2. Diefendorf, A.F.; Mueller, K.E.; Wing, S.L.; Koch, P.L.; Freeman, K.H. Global patterns in leaf ₁₃C discrimination and implications for studies of past and future climate. Proc. Natl. Acad. Sci. USA; 2010; 107, pp. 5738-5743. [DOI: https://dx.doi.org/10.1073/pnas.0910513107] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/20231481]

3. Hill, K.E.; Hill, R.S.; Watling, J.R. Pinnule and stomatal size and stomatal density of living and fossil Bowenia and Eobowenia specimens give insight into physiology during Cretaceous and Eocene paleoclimates. Int. J. Plant Sci.; 2019; 180, pp. 323-336. [DOI: https://dx.doi.org/10.1086/702643]

4. Jordan, G. Uncertainty in palaeoclimatic reconstructions based on leaf physiognomy. Aust. J. Bot.; 1997; 45, pp. 527-548. [DOI: https://dx.doi.org/10.1071/BT96035]

5. Royer, D.L.; Wilf, P.; Janesko, D.A.; Kowalski, E.A.; Dilcher, D.L. Correlations of climate and plant ecology to leaf size and shape: Potential proxies for the fossil record. Am. J. Bot.; 2005; 92, pp. 1141-1151. [DOI: https://dx.doi.org/10.3732/ajb.92.7.1141] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/21646136]

6. Keenan, T.F.; Niinemets, Ü. Global leaf trait estimates biased due to plasticity in the shade. Nat. Plants; 2016; 3, 16201. [DOI: https://dx.doi.org/10.1038/nplants.2016.201]

7. Maslova, N.P.; Karasev, E.V.; Kodrul, T.M.; Spicer, R.A.; Volkova, L.D.; Spicer, T.E.V.; Jin, J.; Liu, X. Sun and shade leaf variability in Liquidambar chinensis and Liquidambar formosana (Altingiaceae): Implications for palaeobotany. Bot. J. Linn. Soc.; 2018; 188, pp. 296-315. [DOI: https://dx.doi.org/10.1093/botlinnean/boy047]

8. McMillen, G.G.; McClendon, J.H. Dependence of photosynthetic rates on leaf density thickness in deciduous woody plants grown in sun and shade. Plant Physiol.; 1983; 72, pp. 674-678. [DOI: https://dx.doi.org/10.1104/pp.72.3.674]

9. Ashton, P.; Berlyn, G. Leaf adaptations of some Shorea species to sun and shade. New Phytol.; 1992; 121, pp. 587-596. [DOI: https://dx.doi.org/10.1111/j.1469-8137.1992.tb01130.x]

10. Abrams, M.D.; Kubiske, M.E. Leaf structural characteristics of 31 hardwood and conifer tree species in central Wisconsin: Influence of light regime and shade-tolerance rank. For. Ecol. Manag.; 1990; 31, pp. 245-253. [DOI: https://dx.doi.org/10.1016/0378-1127(90)90072-J]

11. Kürschner, W.M.; van der Burgh, J.; Visscher, H.; Dilcher, D.L. Oak leaves as biosensors of late Neogene and early Pleistocene paleoatmospheric CO₂ concentrations. Mar. Micropaleontol.; 1996; 27, pp. 299-312. [DOI: https://dx.doi.org/10.1016/0377-8398(95)00067-4]

12. Poole, I.; Weyers, J.D.B.; Lawson, T.; Raven, J.A. Variations in stomatal density and index: Implications for palaeoclimatic reconstructions. Plant Cell Environ.; 1996; 19, pp. 705-712. [DOI: https://dx.doi.org/10.1111/j.1365-3040.1996.tb00405.x]

13. Carins Murphy, M.R.; Jordan, G.J.; Brodribb, T.J. Differential leaf expansion can enable hydraulic acclimation to sun and shade. Plant Cell Environ.; 2012; 35, pp. 1407-1418. [DOI: https://dx.doi.org/10.1111/j.1365-3040.2012.02498.x]

14. Hovenden, M.J.; Vander Schoor, J.K. The response of leaf morphology to irradiance depends on altitude of origin in Nothofagus cunninghamii. New Phytol.; 2006; 169, pp. 291-297. [DOI: https://dx.doi.org/10.1111/j.1469-8137.2005.01585.x]

15. Worth, J.R.P.; Jordan, G.J.; McKinnon, G.E.; Vaillancourt, R.E. The major Australian cool temperate rainforest tree Nothofagus cunninghamii withstood Pleistocene glacial aridity within multiple regions: Evidence from the chloroplast. New Phytol.; 2009; 182, pp. 519-532. [DOI: https://dx.doi.org/10.1111/j.1469-8137.2008.02761.x] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/19210718]

16. Hovenden, M.J. The influence of temperature and genotype on the growth and stomatal morphology of southern beech, Nothofagus cunninghamii (Nothofagaceae). Aust. J. Bot.; 2001; 49, pp. 427-434. [DOI: https://dx.doi.org/10.1071/BT01001]

17. Hill, R. Evolution of Nothofagus cunninghamii and its relationship to N. moorei as inferred from Tasmanian macrofossils. Aust. J. Bot.; 1983; 31, pp. 453-465. [DOI: https://dx.doi.org/10.1071/BT9830453]

18. Hill, R.S. Biogeography, evolution and palaeoecology of Nothofagus (Nothofagaceae): The contribution of the fossil record. Aust. J. Bot.; 2001; 49, pp. 321-332. [DOI: https://dx.doi.org/10.1071/BT00026]

19. Scriven, L.J.; Hill, R.S. Relationships among Tasmanian Tertiary Nothofagus (Nothofagaceae) populations. Bot. J. Linn. Soc.; 1996; 121, pp. 345-364. [DOI: https://dx.doi.org/10.1111/j.1095-8339.1996.tb00761.x]

20. Kürschner, W.M. The anatomical diversity of recent and fossil leaves of the durmast oak (Quercus petraea Lieblein/Q. pseudocastanea Goeppert)—Implications for their use as biosensors of palaeoatmospheric CO₂ levels. Rev. Palaeobot. Palynol.; 1997; 96, pp. 1-30. [DOI: https://dx.doi.org/10.1016/S0034-6667(96)00051-6]

21. Salisbury, E.J. On the causes and ecological significance of stomatal frequency, with special reference to the woodland flora. Philos. Trans. R. Soc. Lond. Ser. B Contain. Pap. A Biol. Character; 1927; 216, pp. 1-65.

22. Schindelin, J.; Arganda-Carreras, I.; Frise, E.; Kaynig, V.; Longair, M.; Pietzsch, T.; Preibisch, S.; Rueden, C.; Saalfeld, S.; Schmid, B. Fiji: An open-source platform for biological-image analysis. Nat. Methods; 2012; 9, pp. 676-682. [DOI: https://dx.doi.org/10.1038/nmeth.2019] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/22743772]

23. Rueden, C.T.; Schindelin, J.; Hiner, M.C.; DeZonia, B.E.; Walter, A.E.; Arena, E.T.; Eliceiri, K.W. ImageJ2: ImageJ for the next generation of scientific image data. BMC Bioinform.; 2017; 18, 529. [DOI: https://dx.doi.org/10.1186/s12859-017-1934-z]

24. Hill, K.E.; Guerin, G.R.; Hill, R.S.; Watling, J.R. Temperature influences stomatal density and maximum potential water loss through stomata of Dodonaea viscosa subsp. angustissima along a latitude gradient in southern Australia. Aust. J. Bot.; 2015; 62, pp. 657-665.

25. Breiman, L. Random forests. Mach. Learn.; 2001; 45, pp. 5-32. [DOI: https://dx.doi.org/10.1023/A:1010933404324]

26. Liaw, A.; Wiener, M. Classification and Regression by randomForest. R News; 2002; 2, pp. 18-22.

27. Meng, Y.A.; Yu, Y.; Cupples, L.A.; Farrer, L.A.; Lunetta, K.L. Performance of random forest when SNPs are in linkage disequilibrium. BMC Bioinform.; 2009; 10, 78. [DOI: https://dx.doi.org/10.1186/1471-2105-10-78] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/19265542]

28. Strobl, C.; Boulesteix, A.-L.; Zeileis, A.; Hothorn, T. Bias in random forest variable importance measures: Illustrations, sources and a solution. BMC Bioinform.; 2007; 8, 25. [DOI: https://dx.doi.org/10.1186/1471-2105-8-25]

29. Breiman, L.; Friedman, J.; Olshen, R.; Stone, C. Classification and Regression Trees; 1st ed. Routledge: New York, NY, USA, 1984.

30. Apley, D.W. Visualizing the effects of predictor variables in black box supervised learning models. arXiv; 2016; arXiv: 1612.08468[DOI: https://dx.doi.org/10.1111/rssb.12377]

31. Jordan, G.J. A new Early Pleistocene species of Nothofagus and the climatic implications of co-occurring Nothofagus fossils. Aust. Syst. Bot.; 1999; 12, pp. 757-765. [DOI: https://dx.doi.org/10.1071/SB98025]

32. Körner, C. When meta-analysis fails: A case about stomata. Glob. Chang. Biol.; 2017; 23, pp. 2533-2534. [DOI: https://dx.doi.org/10.1111/gcb.13700] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/28346734]

33. Scriven, L.J.; McLoughlin, S.; Hill, R.S. Nothofagus plicata (Nothofagaceae), a new deciduous Eocene macrofossil species, from southern continental Australia. Rev. Palaeobot. Palynol.; 1995; 86, pp. 199-209. [DOI: https://dx.doi.org/10.1016/0034-6667(94)00145-A]

34. Busby, J.R. A biogeoclimatic analysis of Nothofagus cunninghamii (Hook.) Oerst. in southeastern Australia. Aust. J. Ecol.; 1986; 11, pp. 1-7. [DOI: https://dx.doi.org/10.1111/j.1442-9993.1986.tb00912.x]