Allometric Equations for Predicting Culm Surface Area of Three Bamboo Species (Phyllostachys spp.)

Abstract

It is strongly debated whether bamboo forest ecosystems are a carbon sink or a carbon source because of insufficient knowledge regarding carbon loss via CO₂ emissions from these forests. The surface area (S) of bamboo culms (stems) is important for estimating culm respiration, a major component of carbon cycling in bamboo forests. However, few studies have attempted to formulate predictive equations for S. In this study, we developed allometric equations for predicting S in three bamboo species grown in Kyushu Island, western Japan: Phyllostachys pubescens Mazel ex Houz., P. bambusoides Sieb. et Zucc. and P. nigra var. henonis. We used a power equation between S and diameter at breast height (D) and a linear equation between S and D × total culm length (H). The results indicated that P. bambusoides and P. nigra shared common site-independent equations. In contrast, P. pubescens required species-specific equations due to interspecific variation in culm slenderness and tapering. We also found that D was a better predictive variable than DH when quantifying S because of its satisfactory predictive performance and simplicity. These findings will be beneficial for evaluating the contribution of bamboo forest ecosystems to carbon cycling.

1. Introduction

Bamboo (family: Poaceae, subfamily: Bambusoideae) is a diverse plant group with nearly 1500 species across approximately 119 genera and is native to all continents except Antarctica and Europe [1,2]. Bamboo forests cover roughly 31.5 million hectares worldwide [2,3] and occupy approximately 1% of the total global forested area [4]. International interest in bamboo has rapidly increased because of its highly versatility, fast growth and environmentally friendly characteristics [4,5,6].

Many studies in the past decade have shown the potential of bamboo forests as a carbon sink (a process that removes carbon dioxide from the atmosphere) [6,7,8,9,10,11]. Nevertheless, there remains considerable debate whether bamboo forest ecosystems are a carbon sink or a carbon source (a process of releasing carbon dioxide to atmosphere) [12,13,14]. One possible reason for this uncertainty is insufficient knowledge regarding carbon loss via CO₂ emissions from bamboo forests [14].

Bamboo forests release CO₂ into the atmosphere through respiration of the culm (stem), leaves, branches and soil, including roots [10,14,15,16,17]. In particular, culm respiration (R_C) is a major component of carbon cycling in bamboo forests, accounting for nearly 30% of aboveground respiration [18,19]. R_C can be quantified by scaling the point-measured respiration rate per unit area (R_AREA) up to the individual- or stand-level with the culm surface area (S), i.e., R_C = S × R_AREA. This convenient approach is widely used to quantify the stem respiration of a tree as well as R_C [20,21,22,23].

During this scaling, errors in S and R_AREA can increase errors in the R_C estimate [24]. For bamboo, a reliable R_AREA must consider vertical variation, culm size and age, and response to temperature change [14,19]. By contrast, previous studies have not focused on the quantification of S. A previous study assumed cone geometry for culms when estimating R_C in a bamboo forest [19]. However, this simplistic assumption is violated when quantifying S [25]. Because small errors in S could lead to large errors in scaled-up R_C [24], S must be characterized accurately and precisely.

In general, measuring the surface area of a stem or culm requires time and labor-intensive destructive sampling [26]. Therefore, many allometric equations have been developed to estimate stem surface area in various tree species based on commonly measured tree attributes such as diameter at breast height and total tree height [27,28,29,30,31]. Numerous equations have also been derived for estimating culm biomass or volume of bamboo species [11,32,33,34,35,36,37,38], but only one study has investigated how S is related to other culm dimensions [39]. In Phyllostachys pubescens Mazel ex Houz., a common bamboo species in Asia with high potential for carbon sequestration [6,9,10,40], a power equation described the relationship between S and external culm diameter at breast height (D) [39]. In addition, a linear equation was suitable for representing the relationship between S and the product of D times total culm length (H) [39]. However, to our knowledge, no other allometric equations for predicting S are found in the literature, raising the question whether such predictive equations are site- and species-specific. If they are specific, destructive sampling is required for every site and species for which an equation is developed. In contrast, generic equations can be applied to multiple sites and species, which results in efficient R_C prediction.

The objective of this study was to develop allometric equations for predicting S in three bamboo species grown in Kyushu Islands, southern Japan: P. pubescens, P. bambusoides Sieb. et Zucc. and P. nigra var. henonis, which are called the three major useful bamboos in Japan [34,41]. In Japan, bamboo forests are estimated to cover more than 150,000 ha [42,43]; these three species account for more than 99% [44]. Hence, developing allometric equations is helpful to evaluating the potential of the bamboo forests as a carbon sink. The findings obtained in this study will also be beneficial when determining equations for S of other bamboo species in other countries. Specifically, we address the following questions regarding our newly developed allometric equations for S:

(1)

Are the allometric equations site-specific?

(2)

Are the allometric equations species-specific?

(3)

What predictive variable is optimal for S prediction?

2. Materials and Methods

2.1. Study Sites

Data used here were gathered in Kyushu Islands, western Japan, which is situated between latitudes 30°59′ N to 33°58′ N and longtitudes 129°33′ E to 132°5′ E. Phyllostachys pubescens and P. bambusoides data were collected in a suburban forest of Toshima Mountain, located in eastern Kumamoto City, Kumamoto Prefecture, western Japan (32°49′ N, 130°48′ E, 80–90 m a.s.l.; hereafter TO). Average annual temperature and rainfall between 1981 and 2010 at the nearest observatory were 16.9 °C and 1986 mm, respectively [45]. The soil type was classified as ando soil [46]. Stands of both species were located adjacent to each other across an approach to Toshima Shrine at the foot of Mt. Toshima. These sites were originally used for upland farming until a few decades ago and became bamboo stands after the planting of bamboo and subsequent abandonment [33,35,36].

Phyllostachys nigra data were collected from two sites: the campus of Fukuoka University of Education in eastern Munakata City, Fukuoka Prefecture (33°48′ N, 130°36′ E, 50 m a.s.l.; MU) and the Ito campus of Kyushu University in western Fukuoka City, Fukuoka Prefecture, western Japan (33°36′ N, 130°13′ E, 60 m a.s.l.; IT). Average annual temperature, rainfall and soil type were, respectively, 15.6 °C, 1640 mm and a brown forest soil for MU and 16.1 °C, 1677 mm and a gley soil or brown forest soil for IT [45,46]. Both sites were very dense, and no other vegetation was found on their forest floors [34,35].

2.2. Field Measurements

For each site, healthy, living culm samples were felled at ground level after the growing (elongation) period from October to January [47]. In total, 200 culm samples were gathered from each of the P. pubescens and P. bambusoides stands in TO, while 60 and 110 culms were collected from P. nigra stands in MU and IT, respectively. Total culm length (H) and external culm diameter at breast height (D) were directly measured with a tape measure to the nearest 0.1 m and a digital caliper to the nearest 0.1 cm, respectively. To quantify S accurately and efficiently, external culm diameters at 10 successive points were measured with a caliper to the nearest 0.1 cm [25]. The measurement points were located along the culm at equal intervals of one tenth of the direct measured H. It should be noted that the ages of sample culms were not recorded in our sampling. However, this did not distort our results because of the following reasons: (1) the bamboo culms attain maximum total culm length by completing culm elongation [6,48,49], (2) the lack of vascular cambium results in no secondary growth of the culm after the elongation period [50,51], and thus (3) the relationships of S to other culm dimensions do not vary with culm age after completing elongation. The dendrometric parameters of sample culms are shown in

Table 1. Dendrometric parameters of sample culms (n = 890).

Species	Site 1	Number of Samples 2	Soil Type 3	Total Culm Height (m) 4	Diameter at Breast Height (cm) 4	Culm Surface Area (m2) 4
P. pubescens	TO	200 (100)	Ando soil	14.1 ± 2.5	9.2 ± 2.7	2.46 ± 1.01
P. bambusoides	TO	200 (100)	Ando soil	11.1 ± 3.0	5.0 ± 1.7	1.26 ± 0.70
	KI	102 (51)	Brown forest soil or ando soil	12.8 ± 3.4	6.2 ± 2.1	1.77 ± 1.00
	FU	122 (61)	Brown forest soil or ando soil	11.8 ± 2.5	5.7 ± 1.7	1.50 ± 0.81
	HI	96 (48)	Brown forest soil or ando soil	9.0 ± 2.1	4.1 ± 1.2	0.87 ± 0.46
P. nigra	MU	60 (30)	Brown forest soil	7.3 ± 2.0	2.9 ± 1.1	0.50 ± 0.30
	IT	110 (55)	Gley soil or brown forest soil	6.2 ± 1.5	2.7 ± 0.8	0.37 ± 0.19

¹ TO: Toshima, KI: Kitakyushu, FU: Fukuoka, HI: Hisayama, MU: Munakata, IT: Ito; ² Values in parentheses are sample sizes of modeling and validation data; ³ The soil type classification is based on website information [46]; ⁴ Average ± S.D.

.

2.3. Analyses Methods

For all culm samples, S was calculated from H and successive diameters measured at one-tenth intervals of H using the sectional measurement method [25,39]. Similarly, apparent culm volume (V) [52,53] was computed with the same method [33,34,35,36]. To improve the reliability of the allometric equations, we also used published data on D, H, S and V of P. bambusoides from three sites [54]: Fukuoka (FU; n = 122), Kitakyushu (KI; n = 102) and Hisayama (HI; n = 96), Fukuoka Prefecture. Therefore, as shown in Table 1, the resulting total number of samples was 890 (200 for P. pubescens from one site, 520 for P. bambusoides from four sites and 170 for P. nigra from two sites). Despite the reliable data of site location, the soil type was estimated to be brown forest soil or ando soil [46]. These data were used previously to develop height-diameter curves, two-way volume equations and one-way volume tables [33,34,35,36,55]. Furthermore, a part of the P. pubescens data (150 of 200 culms) was used to determine predictive equations for S [39].

For each species and site, samples were randomly split into modeling and validation data. The former was used to develop allometric equations for S, while the latter was used independently to test the predictive performance. Scatterplots of H and D for both datasets were very similar (data not shown), suggesting that robust conclusions could be reached from the validation.

We used the following two allometric equations:

S = αD^β,

(1)

and

S = γDH,

(2)

where α, β and γ are the parameters [27,39]. Phyllostachys bambusoides and P. nigra data were gathered from multiple sites, which allowed us to test whether the allometric equations were site-specific. In addition, P. bambusoides and P. nigra had similar culm taper [34], suggesting that they should share a common allometric equation for S. Furthermore, if the equations for the three species are generic, it is not necessary to develop the equations for each species by tedious destructive sampling. For these reasons, the allometric equations were determined for (1) each species and site (hereafter site-specific equation), (2) each species (species-specific equation), (3) P. bambusoides and P. nigra (common equation) and (4) all species (generic equation) (see

Table 2. Explanation of the allometric equations developed in this study.

Equation	Species	Site 1	Number of Samples 2
Site-specific	P. bambusoides	TO	200
		KI	102
		FU	122
		HI	96
	P. nigra	MU	60
		IT	110
Species-specific	P. pubescens	TO	200
	P. bambusoides	TO, KI, FU and HI	520
	P. nigra	MU and IT	170
Common	P. bambusoides and P. nigra	TO, KI, FU, HI, MU and IT	690
Generic	P. pubescens, P. bambusoides and P. nigra	TO, KI, FU, HI, MU and IT	890

¹ TO: Toshima, KI: Kitakyushu, FU: Fukuoka, HI: Hisayama, MU: Munakata, IT: Ito; ² The number of sample culms used for both modeling and validation data.

). Goodness of fit was evaluated with the adjusted coefficient of determination (r²), root mean square error (RMSE) and Akaike’s Information Criteria (AIC).

As a preliminary analysis for examining the validity of Equation (2), we also tested the following three equations:

S = γDH + δ,

(3)

S = γ(DH)^δ,

(4)

and

S = γD^δH^ε,

(5)

where γ, δ and ε are the parameters. Values of δ in Equations (3) and (4) were not significantly different from zero and unity, respectively. In addition, neither δ nor ε in Equation (5) differed significantly from unity. These facts imply that these three equations are essentially identical to Equation (2), which also exhibited smaller AIC values. Hence, we selected Equation (2) as the predictive equation with two variables (D and H).

Using validation data, S was predicted by substituting measured D and H into the determined equations. Accuracy and precision were then evaluated with average relative error (%AE) and relative root mean square error (%RMSE), respectively. These were defined as:

$$\%\mathrm{AE}=100\times {{\displaystyle \sum }}_n\left\{\left(S_{\mathrm{predicted}}-S_{\mathrm{observed}}\right)/S_{\mathrm{observed}}\right\}/n,$$

(6)

and

$$\%\mathrm{RMSE}=100\times \sqrt{{{\displaystyle \sum }}_n{\left(S_{\mathrm{predicted}}-S_{\mathrm{observed}}\right)}^2/n}/S_{\mathrm{average}},$$

(7)

where S_predicted and S_observed are predicted and observed S, respectively; n is the number of culm samples; and S_average is the average S_observed values. Differences between S_predicted and S_observed were examined with a Wilcoxon signed rank test.

To compare culm characteristics, height to diameter ratio (HDR) [56] and breast height form-factor for apparent culm volume (λ_b) [57] were calculated from modeling data. The former is known as a slenderness coefficient, whereas the latter is a measure of culm tapering. λ_b was calculated as:

λ_b = 4V/πD²H.

(8)

Culms with higher HDR and λ_b are more slender and non-tapering, respectively. These values were compared using Tukey’s HSD test. All statistical analyses were performed using R software version 3.4.3 [58], and p < 0.05 was considered significant.

3. Results and Discussion

3.1. Are the Allometric Equations Site-Specific?

Figure 1 shows the relationship between S and D for each species and site. The parameter estimates and fit statistics of the developed equations are summarized in

Table 3. Parameter estimates of the developed allometric equations and their fit statistics.

Equation	Species	Site 1	Power Equation					Linear Equation
			α	β	r2	RMSE (m2)	AIC	γ	r2	RMSE (m2)	AIC
Site-specific	P. bambusoides	TO	159.03	1.65	0.97	0.155	−83.25	2.08	0.99	0.051	−306.68
		KI	178.44	1.68	0.99	0.147	−46.26	2.04	0.99	0.073	−120.05
		FU	184.19	1.70	0.98	0.089	−115.35	2.07	0.99	0.068	−150.95
		HI	242.78	1.78	0.98	0.073	−106.07	2.23	0.99	0.055	−136.63
	P. nigra	MU	143.52	1.63	0.97	0.060	−79.96	2.14	0.99	0.025	−134.44
		IT	263.16	1.83	0.94	0.039	−191.73	2.12	0.99	0.011	−322.93
Species-specific	P. pubescens		83.21	1.49	0.98	0.174	−58.68	1.80	0.99	0.101	−168.49
	P. bambusoides		179.50	1.69	0.98	0.131	−315.58	2.07	0.99	0.069	−649.89
	P. nigra		239.91	1.79	0.95	0.051	−260.28	2.13	0.99	0.017	−441.90
Common	P. bambusoides and P. nigra		187.66	1.70	0.98	0.118	−495.65	2.08	0.99	0.061	−950.25
Generic	All species		116.01	1.57	0.97	0.294	−3.27	1.92	0.98	0.142	−471.77

¹ TO: Toshima, KI: Kitakyushu, FU: Fukuoka, HI: Hisayama, MU: Munakata, IT: Ito.

. The relationship between S and D can be described by a power equation for all species and sites (r² > 0.94). The relationship between S and DH can also be expressed by a linear equation (r² > 0.99; Figure 2). For all species and sites, %AE of the site-specific equations was small (less than ±3%;

Table 4. Accuracy (%AE) and precision (%RMSE) of the culm surface area estimated by the developed allometric equations.

Equation	Species	Site 1	Power Equation		Linear Equation
			%AE (%)	%RMSE (%)	%AE (%)	%RMSE (%)
Site-specific	P. bambusoides	TO	−1.45	12.04	−0.55	4.04
		KI	2.06	6.17	−1.95	2.73
		FU	−0.43	7.91	−1.17	5.54
		HI	0.95	9.44	1.12	7.86
	P. nigra	MU	−1.94	12.76	−1.73	3.80
		IT	−3.09	11.06	−1.31	3.33
Species-specific	P. pubescens	TO	0.30	8.70	−0.51	4.74
	P. bambusoides	TO	−1.42	11.30	−0.88	4.10
		KI	0.05	5.78	−0.28	3.16
		FU	0.91	8.03	−1.21	5.53
		HI	0.73	9.19	−5.77	9.94
	P. nigra	MU	−10.45	14.47	−2.15	3.99
		IT	0.51	11.13	−0.83	3.41
Common	P. bambusoides	TO	−2.32	11.41	−0.80	4.08
		KI	−0.46	5.91	−0.20	3.21
		FU	0.31	8.13	−1.13	5.56
		HI	−0.48	9.25	−5.70	9.89
	P. nigra	MU	−3.28	11.17	−4.76	6.01
		IT	9.08	14.06	−3.47	4.01
Generic	P. pubescens	TO	16.54	21.75	5.81	8.41
	P. bambusoides	TO	−8.39	19.40	−8.48	10.30
		KI	−9.59	15.81	−7.92	7.77
		FU	−8.32	14.42	−8.78	9.06
		HI	−4.62	14.10	−12.99	16.34
	P. nigra	MU	−2.28	15.18	−12.12	14.02
		IT	−10.98	13.13	−10.94	11.37

¹ TO: Toshima, KI: Kitakyushu, FU: Fukuoka, HI: Hisayama, MU: Munakata, IT: Ito.

). Hence, Equations (1) and (2) are acceptable as the predictive equations for every species and sites examined in this study.

In P. bambusoides and P. nigra, a single, site-independent power equation represented the relationship between S and D (Figure 3). The values of r² and RMSE of the power equation were, respectively, 0.98 and 0.131 m² for P. bambusoides and 0.95 and 0.051 m² for P. nigra. The relationship between S and DH was fitted by a single linear equation for both species, with r² = 0.99 for both species (Figure 4). RMSE was 0.069 m² for P. bambusoides and 0.017 m² for P. nigra. Site- and species-specific equations for both species had comparable fit statistics and predictive performance, suggesting that separate equations are not necessary for different P. bambusoides and P. nigra sites. Therefore, the answer to the first question is that allometric equations for P. bambusoides and P. nigra are not site-specific. However, further studies are necessary to verify whether equations for P. pubescens are site-specific, as our data were insufficient to clarify this point.

3.2. Are the Allometric Equations Species-Specific?

Figure 5 depicts the relationships of S to D and DH for P. bambusoides and P. nigra. Similar to the two-way volume equations for these species [33,34], the relationships were well expressed by the common power and linear equations with r² = 0.98 for the former and r² = 0.99 for the latter. Common equations had intermediate RMSEs compared to site- and species-specific equations. The relationships of S to D and DH for all three species are shown in Figure 6. Although the generic power equation for all three species was significant (r² = 0.97), S and D scatterplots indicate a clear separation of P. pubescens and the other two species; for the same D, P. pubescens had a smaller S than P. bambusoides and P. nigra (Figure 6a). Similarly, the relationship between S and DH for the three species could be fitted by the single linear equation (r² = 0.98). However, for the same DH, S was slightly smaller in P. pubescens than in the other two species (Figure 6b). In both generic equations, %AE was positive for P. pubescens, but negative for P. bambusoides and P. nigra. Therefore, when applying the generic equations, R_C will be overestimated for P. pubescens and underestimated for the other two species. Moreover, the generic equations had larger %RMSEs than the other equations. These results indicate that generic equations are inappropriate when estimating S, and the predictive equations for P. pubescens should be species-specific. For these reasons, the optimal equations for predicting S are as follows:

S = 83.21D^1.49,

(9)

S = 1.80DH,

(10)

for P. pubescens (species-specific equations) and

S = 187.66D^1.70,

(11)

S = 2.08DH,

(12)

for P. bambusoides and P. nigra (common equations).

For P. pubescens, both power and linear equations had %AE less than ±0.5%, with no significant differences between S_predicted and S_observed. Generally, S_predicted of P. bambusoides from power and linear common equations did not differ significantly from S_observed, with %AE less than ±1.2% for all sites and equations except for S_predicted in the TO from the power equation (%AE = −2.3%) and in the HI from the linear equation (−5.7%). For P. nigra, a significant difference was found between S_predicted and S_observed at both sites; %AE from the power equation was relatively large for IT (9.1%), but less than ±5% for other sites and equations. It should be noted that the common power equation in MU had smaller %AE than the species-specific power equation. Therefore, the predictive performance of common equations is not inferior to that of species-specific equations. However, common equations have broader application than species-specific equations, even if they are not relevant for P. pubescens.

Why are the predictive equations different between P. pubescens and the other two species? Figure 7a compares HDR computed from modeling data across species and sites. P. pubescens had significantly lower HDR, and therefore shorter culm length, than P. bambusoides and P. nigra. Hence, P. pubescens has a relatively smaller S for the same D than the other species. Moreover, P. pubescens had lower λ_b (Figure 7b), indicating a more tapered culm. A tapered culm results in smaller S for the same D and H. These facts indicate that interspecific variation in culm slenderness and tapering explains differences in allometric power equations, while only variation in culm tapering explains differences in the linear equation. Therefore, the answer to the second question is that species-specific predictive equations are required for P. pubescens, but not for P. bambusoides and P. nigra, because of interspecific variation in culm characteristics.

3.3. What Predictive Variable Is Optimal for S Prediction?

We compared two allometric equations for predicting S with the different variables, i.e., D and DH. Which predictive variable is better? We found that the DH-based equation, Equation (2), showed slightly higher r² and smaller RMSE and AIC than the D-based equation, Equation (1) (Table 3). However, %AE and %RMSE magnitudes exhibited inter-site and inter-specific variation (Table 4), suggesting that the inclusion of the additive variable H does not improve the predictive performance. Diameter at breast height and tree height are easy measureable dendrometric variables of trees [28]. D of bamboo is also the most easily measured culm attribute. In contrast, it is difficult to measure H of standing bamboo culm, since the culm tip is generally bent askew [35,52,59]. Equations (10) and (12) clearly demonstrated that inaccurate measurement of H causes errors in S prediction, which results in uncertain R_C prediction. Similar arguments are found in estimating V or biomass of tree species [60,61,62]. For these reasons, our answer to the third question is that D is the optimal variable for estimating S of the three examined bamboo species.

4. Conclusions

In this study, we developed allometric equations for predicting S in three Phyllostachys species grown in Kyushu Island, western Japan. We demonstrate that P. bambusoides and P. nigra share the common site-independent equations, whereas P. pubescens requires species-specific equations because of interspecific variation in culm slenderness and tapering. We also conclude that D is the optimal variable when predicting S of all three species. These equations enable us to scale point-measured R_AREA up to R_C at individual- or stand-levels efficiently through simple measurement of D. Therefore, the equations developed here will be beneficial for evaluating R_C, and bring a consensus regarding the contribution of bamboo forest ecosystems to the carbon cycle. In future, it is necessary to examine whether the equations vary with culm sizes, climatic conditions and soil properties.