How many broadleaved trees are enough in conifer plantations? The economy of land sharing, land sparing and quantitative targets

Summary

1. For biodiversity conservation to be an effective and significant social investment, non‐marketed values of biodiversity conservation and its associated opportunity costs should be evaluated in monetary terms.
2. In this study, we measured the willingness to pay (WTP) for bird abundance using a choice experiment (CE) based on the random utility model. We performed a cost–benefit analysis to identify the optimal proportion of broadleaved trees in conifer plantations on a volume basis to maximize the social benefits of bird conservation and wood production.
3. The results suggested that respondents to the CE were not satisfied with their current situation and preferred an increase in bird abundance. However, the estimated WTP indicated diminishing returns of bird conservation. More specifically, WTP first greatly increased before gradually experiencing decreasing marginal values, reaching its peak and finally decreasing slightly with increasing bird abundance.
4. Optimization analyses indicated that when the relationship between bird abundance and broadleaved tree proportion was convex, semi‐natural plantations with nonzero broadleaved tree proportion (0·02–0·22) were always optimal options. When the relationship was linear, optimal broadleaved tree proportion ranged from 0 to 0·78 and was greatly affected by wood values. When the relationship was concave, there were only two optimal broadleaved tree proportions: a very high proportion (approximately 0·90) and the lowest possible proportion (0). When the convex and concave relationships approached the linear form, comparable benefits could be attained across broad ranges of broadleaved tree proportion both within and across the relationships. In such cases, it would be useful to increase the likelihood of a feasible land‐use strategy of either land sparing or land sharing in order to be successful.
5. Synthesis and applications. It can be difficult to set quantitative targets in biodiversity conservation solely on an ecological basis, and social benefits of biodiversity conservation can create diminishing returns in many situations. The framework we propose shows how to reconcile resource production and biodiversity conservation in the real world.

Introduction

Biodiversity cannot be conserved for free. In general, conflicts emerge between resource production and conservation (Risser 1999), which is a predominant reason why biodiversity has been lost and its scarcity as well as values have been acknowledged. Effective biodiversity conservation as a social investment requires the evaluation of non‐marketed values of biodiversity in monetary terms and the accommodation of trade‐offs between biodiversity conservation and resource production (Arrow et al. 1996; Balmford et al. 2002; Bateman et al. 2015). Confronting the importance of the economic perspective in biodiversity conservation, Hunter (1990) suggested that money is the bottom line and permeates our culture.

The integration of economics and ecology allows biodiversity conservation to embrace reality and can change the strategy of biodiversity conservation. For example, although primary forests have the highest ecological value (Gibson et al. 2011; Edwards et al. 2014a), the management of logged forests, which have a lower ecological value, can have priority over the protection of primary forests given certain budgets (Wilson et al. 2010; Edwards et al. 2014b). This is because it is too expensive to conserve biodiversity in primary forests, and logged forests can be cost‐effective habitats for biodiversity conservation compared with forest conversion into agricultural fields (Fisher et al. 2011; Wilcove et al. 2013).

Setting quantitative targets is a prerequisite process in biodiversity conservation (Tear et al. 2005). If there are obvious thresholds across which an indicator of biodiversity precipitously declines, we could set these thresholds as conservation targets, for example minimum habitat structures or areas (Guénette & Villard 2005; Betts, Forbes & Diamond 2007). However, it can be difficult to empirically find clear thresholds (Lindenmayer, Fischer & Cunningham 2005; Swift & Hannon 2010), and the theoretical underpinnings of particular thresholds may not always exist. In such cases, how can quantitative targets be set? Crome (1997) suggested the difficulty of this problem in fragmented landscapes because alternative forest parcel can harbour unique biota. Therefore, solely on an ecological basis, we may end up concluding that every parcel is useful and important (Crome 1997).

Economics, and more specifically cost–benefit analyses, can enable us to set quantitative targets to reduce the costs and maximize the benefits of biodiversity conservation (Lippke & Bishop 1999; Buongiorno & Gilless 2003). A cost–benefit analysis seeks optimal combinations of two products (typically, biodiversity and crops) to maximize the obtained benefits based on the relative values of these products. The solution can greatly depend on the functional forms of trade‐offs between biodiversity conservation and resource production (i.e. production possibility frontiers). In addition, functional forms without clear thresholds are usually treated (Perfecto et al. 2005; Zhang & Pearse 2011). Therefore, the cost–benefit analysis is promising for setting quantitative targets in many situations, as well as for offering insight into an important land‐use problem, namely whether resource production and biodiversity conservation should be separated or integrated – the so‐called land sparing vs. land sharing debate (Perfecto et al. 2005; Yamaura et al. 2012; Butsic & Kuemmerle 2015).

Few studies have attempted a cost–benefit analysis of biodiversity conservation and assumed that economic values per unit of conserved biodiversity are constant with increasing biodiversity conservation (Naidoo & Ricketts 2006; Edwards et al. 2014b; Teuscher et al. 2015). However, economic returns of products, which can be measured by willingness to pay (WTP), usually diminish (Field 2008; Zhang & Pearse 2011). Fisher et al. (2008) suggested that this diminishing return (marginality) should be incorporated in the cost–benefit analysis because political decisions frequently face the problem of what to do with the next unit. However, few studies have employed a formal economic analysis to consider diminishing returns (Naidoo & Adamowicz 2005) and its ecological substitute (i.e. species–area relationships; Wilson et al. 2007).

In this study, we examined the amount of broadleaved trees required in conifer plantations to improve bird abundance from an economic perspective. Conifer plantations are expanding in areas with increasing wood demands, and their replacement of native forests usually devastates the biota; the reconciliation of biodiversity conservation and wood production is actively discussed (Brockerhoff et al. 2008; Paquette & Messier 2010). Our previous stand‐level study revealed that bird abundance in conifer plantations can linearly increase with the amounts of mixed native broadleaved trees. We did not find clear thresholds indicating required amounts of broadleaved trees to maintain bird abundance on an ecological basis (Yoshii et al. 2015). Because broadleaved trees have lower wood values compared with planted coniferous trees and the increase in broadleaved trees requires the space of coniferous trees, the improvement in bird abundance in conifer plantations by increasing broadleaved trees decreases the revenue of foresters. This is a typical trade‐off between conservation and resource production. For this study, we conducted a choice experiment (CE) to measure the WTP to improve bird abundance and sought quantitative targets of bird conservation in monetary terms through a cost–benefit analysis. In addition, we performed a sensitivity analysis to examine how quantitative targets can be changed depending on the situation.

Materials and methods

A choice experiment to measure WTP

CEs allow us to measure individual preferences by asking respondents to choose among various multi‐attribute scenarios. This methodology was initially developed by Louviere & Hensher (1982) and Louviere & Woodworth (1983) and is classified as a family of stated preference approaches (Louviere, Hensher & Swait 2000). CEs are currently used in various fields, including marketing, transportation and environmental economics (Hensher 1994; Louviere 1994; Adamowicz et al. 1998). CEs are also an essential method to evaluate non‐marketed values of environmental goods and services in monetary terms (Adamowicz et al. 1998).

In the CE, we established hypothetical alternative plans of forest management to improve the bird abundance in conifer plantations by increasing broadleaved trees in Hokkaido prefecture, which comprises 14 districts. We asked respondents to select preferred scenarios from a range of plans. Hypothetical plans were applied to 1000 ha of plantations that were adjacent to human dwellings in each district. In other words, 14 000 ha of plantations were covered in total, which represents approximately 1% of the total plantation area in Hokkaido. Alternative plans differed according to three attributes: (i) the number of bird individuals in conifer plantations per ha, (ii) the number of bird‐watching stations in a district and (iii) the additional amount of tax payments needed to introduce new forest management plans (Table 1).

Table 1. Attributes and their levels in choice experiments

Attributes	Levels
Number of bird individuals in conifer plantationsa^a Levels were based on results of a prior empirical study (Yoshii et al. 2015).	6·2 (0% broadleaved trees: a control scenario)
	7·8 (including 20% of broadleaved trees)
	9·4 (including 40% of broadleaved trees)
	11·0 (including 60% of broadleaved trees)
	12·6 (including 80% of broadleaved trees)
	14·3 (forest of 100% broadleaved trees)
Number of bird‐watching station in a district	1, 2, 3 and 4
Additional amount of tax payments to introduce new forest management plans	1000, 2000, 3000, 5000 and 10 000 JPY

• ^a Levels were based on results of a prior empirical study (Yoshii et al. 2015).

The number of bird individuals in plantations was assumed to increase depending on the amounts of mixed broadleaved trees. Based on the previous empirical study that surveyed birds in plantations with different amounts of broadleaved trees (Yoshii et al. 2015), the minimum and maximum values were estimated as bird abundance in ‘pure’ plantations (with no broadleaved trees) and in natural broadleaved forests (without planted coniferous trees), respectively. Bird abundance (N) was assumed to increase linearly with the proportion of broadleaved trees in the basal area (p_bl) according to N = 6·21 + 8·04 × p_bl, following the results of Yoshii et al. (2015). However, we did not find predominant support of the linear response over the nonlinear response, possibly due to the small sample size. Furthermore, population densities can inherently take various forms against environmental gradients, depending on the situations (Austin 2002). Therefore, we changed the functional forms of bird abundance to p_bl in the sensitivity analysis (see cost–benefit analysis). Bird abundance was highly correlated with bird‐species richness (r = 0·98, P < 0·001), although we used abundance as an attribute since there would not be large differences in species richness among the hypothetical plans at the project level (i.e. in 14 000 ha). In the CE, we showed bird abundance per ha (6·2–14·3 individuals) as well as project‐level abundance (87 000–200 000 individuals) of individual alternative plans to respondents via proportional calculation.

We considered local habitat structure as a single determinant of bird abundance in order to simplify the cost–benefit analysis, especially since habitat structure can have larger effects on bird abundance than landscape structure in forested landscapes where the majority of forestry practices occur (Yamaura, Katoh & Takahashi 2008). We also noted that benefits of bird conservation were only measured by N, which was the total abundance of bird communities rather than the abundance of specific endangered species or functional groups sensitive to plantation forestry (e.g. cavity nesters, flycatchers). We excluded three species whose abundance increased with plantation intensity (coal tit Periparus ater, goldcrest Regulus regulus and Sakhalin leaf warbler Phylloscopus borealoides) from N (Yoshii et al. 2015). In this regard, these species are known for feeding in coniferous trees and are therefore positively affected by forestry plantations (Yamaura et al. 2009). In the CE, we told the respondents that bird communities are impoverished in conifer plantations compared with natural forests, and the focal birds are 31 common species inhabiting natural forests in Hokkaido. We expected that this crude attribute allowed the respondents to easily comprehend the benefits of bird conservation.

The second attribute (the number of bird‐watching stations) was used to separate recreational use values in conifer plantations with increased bird abundance from values of bird abundance itself. Essentially, new forest management plans were intended to increase the passive use values of forests (higher bird abundance), which is sometimes called non‐use values; however, some respondents may highly value forests with higher bird abundance since they may enjoy bird‐watching there. This effect was accounted for in the second attribute. We used the number of bird‐watching stations rather than other facilities (e.g. trail lengths) to confine the intended use to bird‐watching. The third attribute was additional tax payments required to achieve new forest management. The WTP for bird abundance was estimated on the basis of the trade‐off between the first and third attributes.

Each respondent evaluated three profiles (alternative management plans) with different levels of the three attributes. This evaluation process was repeated eight times (i.e. eight choice sets with different combinations of the levels were used). Profiles were designed following an orthogonal main effect design to avoid confounding the effects of individual attributes (Louviere, Hensher & Swait 2000). In February 2015, a research company sent invitation emails regarding our Internet questionnaire to 11 800 registered respondents in Hokkaido Prefecture. Of these respondents, 1194 (10·1%) completed the questionnaire. After removing 238 respondents who disagreed with the introduction of our proposed forest management plans regardless of the various attributes and levels, we used 956 responses (80%) for the analysis.

Random utility model: Measuring the benefits of bird conservation

Results of the CE were analysed using a random utility model with Stata 12 (StataCorp, College Station, TX, USA). Utility for a profile i (U_i) is described as a linear combination of deterministic (V_i) and random (ε_i) terms: U_i = V_i + ε_i. The probability that profile i is chosen is equated to the probability that U_i is larger than U_j:

(eqn 1)

where C is a choice set. Because we assume that the error term follows a Gumbel distribution (McFadden 1974), the above probability, Pr(i|C), can be expressed as a conditional logit model (McFadden 1974):

(eqn 2)

Without losing generality, deterministic terms can be described as a linear combination of parameters. We modelled the observable part as μβx_i, where x_i denotes a vector of attributes and β denotes its coefficient. The scale parameter, μ, is conventionally assumed to be 1. The marginal WTP for an increase in bird abundance (per ha) is obtained by dividing the ‘bird abundance’ parameter by the ‘additional tax payment’ parameter, which indicates the marginal utility of income. Considering the possibility that the marginal WTP for bird abundance takes a nonlinear functional form, we fitted linear and quadratic models for bird abundance and compared the models with Akaike Information Criterion (AIC).

Cost–benefit analysis: Searching for optimal plantation intensity

We assumed that forests were mature plantations with 300 m³ ha⁻¹ of wood stock (Forestry Agency 2014b) and that its composition was represented by its proportion of broadleaved trees in the basal area: (p_bl): p_bl + p_cnf = 1, where p_cnf was the proportion of planted coniferous trees. The lowest value of p_bl (0) indicates pure conifer plantations without any broadleaved trees, whereas the highest value of p_bl (1) indicates mature natural forests without any planted coniferous trees. Wood values were based on stumpage prices of broadleaved trees (500 yen m⁻³) and coniferous trees (1000 yen m⁻³), according to the current prices in Hokkaido (Forestry Agency 2014a). It means that wood values decreased with increasing p_bl due to the replacement of coniferous trees by broadleaved trees.

We also considered other silvicultural (opportunity) costs, which was represented by further decreases in the amounts of planted coniferous trees. Assuming the retention of broadleaved trees during the harvest as an alternative silvicultural approach, we expected that there would be few additional harvest and weeding costs with the increase in broadleaved trees. The major additional costs would be shading effects of the retained trees on the planted coniferous trees (Rose & Muir 1997). Yoshida et al. (2005) examined the stand structure of a 60‐year‐old larch Larix kaempferi plantation with retained broadleaved trees (mostly Quercus crispula) in central Japan. Retained trees (larger than 40 cm d.b.h.) occupied 18% of the stand space, and the basal area (BA) of the larches within 10 m of the retained trees could be halved. No silvicultural practices had been conducted in this stand after the initial weeding. We considered this result as the worst‐case scenario (with the most severe impacts on the planted trees), and modelled the stock of the planted conifers by 300 m³ × p_cnf², which can roughly reproduce this shading effect (Fig. 1a). Conversely, we did not consider the benefits of increasing p_bl other than bird conservation, such as the conservation of other taxa (Ohsawa 2007) and the maintenance of long‐term site productivity (Franklin 1989).

Figure 1

Modelled response forms of opportunity cost and bird abundance to increasing broadleaved tree proportion. (a) Opportunity cost (shown in grey) was expressed as the differences between the wood value of pure plantation (p_bl = 0) and that of plantations with broadleaved trees (p_bl > 0). Wood value was summed values of coniferous trees and broadleaved trees, and the single model was used to express the value of broadleaved trees (masked by the model for opportunity cost with power = 1·0). The values of coniferous trees (and thus wood value) were modelled in different ways using three different levels of powers in stock calculation to account for the additional silvicultural costs: 1·0, 1·5 and 2·0 (i.e. 300 m³ × p_cnf^1·0, 300 m³ × p_cnf^1·5 and 300 m³ × p_cnf^2·0, respectively). (b) With regard to bird abundance in relation to broadleaved tree proportion, deep convex, linear and deep concave forms are represented by 6·21 + 8·04 × (1 − ((1 − p_bl)²)²), 6·21 + 8·04 × p_bl and 6·21 + 8·04 × (1 − (1 − ((1 − (1 − p_bl²))²))), respectively. Shallow convex and concave forms are represented by intermediate formulations.

We considered five possible response forms of bird abundance to broadleaved tree proportion (p_bl): convex, shallow convex, linear, shallow concave and concave (Fig. 1b). As described above, we increased bird abundance per ha from 6·2 at 0 p_bl to 14·3 at 1 p_bl. We converted bird abundance into an amount of money per ha paid by a person using an equation obtained from the CE and random utility model (i.e. eqn 3; see below). We then multiplied this value by 80% (the proportion of respondents who agreed with the introduction of management plans) of the total labour force in Hokkaido (2·12 million) and considered this as a monetary value of bird conservation per ha, which we call bird values.

A question with the cost–benefit analysis was whether forgone wood values as an opportunity cost can be compensated by increasing bird values. The analysis identified the specific p_bl required to maximize economic forest values represented by the sums of wood and bird values. We searched for the optimal level of p_bl with an algorithm of simulated annealing (SANN) using the ‘optim’ R function. Because net forest values could have more than one peak against p_bl (local optima), we conducted optimization analyses with 99 different initial values from 0·01 to 0·99 with 0·01 increments and took the proportion with the highest forest value as the globally optimal proportion. Since p_bl only took values ranging from 0 to 1, we used an inverse logit transformation of the parameter in the optimization process and treated the transformed optimized values as optimal p_bl.

To examine the sensitivity of the optimal p_bl, we iterated this search for each combination of the five bird response forms and three levels of WTP (using factors of 1, 2/3 and 1/3), wood values (using factors of 2, 1 and 0·5), broadleaved tree values (using factors of 1·5, 1 and 0·5) and silvicultural costs (using powers of p_cnf to calculate stock of coniferous trees: 2·0, 1·5 and 1·0; Fig. 1a). The variabilities of wood values were motivated by their historical ranges in Japan (Yamaura et al. 2012), and the lower limit of WTP was set at 1/3 since WTP can be overstated up to three times (List & Gallet 2001) and also given the relatively low response rate of the questionnaire (10·1%). The total number of combinations was 5 × 3⁴ = 405. In the calculation of the wood values, we first summed the values of broadleaved and coniferous trees before factoring the wood values. Therefore, broadleaved tree values are related to the economic costs of increasing p_bl, and the factors of wood values are the general economic wood values. Finally, an ANOVA was conducted for optimal p_bl with five control variables as main factors to examine their relative effects on the optimal p_bl. We conducted all of the optimization analyses using R version 3.0.3 (R Core Team 2014).

Results

Choice experiment for WTP

In the analysis of conditional logit models, all of the estimated parameters except for bird‐watching stations were significantly different from zero at the 5% level (Table 2). The parameter of additional tax payments was negative, which indicates that respondents preferred a cheaper alternative. The parameter of alternative‐specific constant for status quo profile was also negative, indicating that respondents avoided choosing the current situation, i.e. respondents found values in introducing new forest management plans regardless of their attributes and levels. Comparing AIC of the models suggested that the quadratic model was better supported than the linear model (∆AIC ~ 36), and both models showed the medium fits (Table 2). In the linear model, the parameter of bird abundance was positive, but in the quadratic model, the quadratic parameter was negative, suggesting that the WTP for bird abundance had a nonlinear form (Fig. 2a). WTP first greatly increased before gradually experiencing decreasing marginal values, reaching its peak and finally decreasing slightly with increasing bird abundance (see Appendix S1 in Supporting Information for the estimation uncertainty of this form). Specifically, WTP (yen per person) was modelled as a quadratic function of bird abundance N to take 0 values when N = 6·21 as follows:

(eqn 3)

where U_N is a utility level for a scenario with N and U_C is a utility level for the control scenario (N = 6·21). Parameters β₁ (= 0·5823), β₂ (= −2·5493 × 10⁻²) and β_tax (= −0·0353 × 10⁻²) represent linear terms of bird abundance, quadratic terms of bird abundance and additional tax payments, respectively. The effect of the alternative‐specific constant is not included in this formulation of WTP since this effect is not attributable to the WTP for bird abundance. It may rather be attributable to the context for payment (e.g. respondents may always choose the status quo profile because of disapproval of tax increase). Since bird abundance was expressed as the linear combination of p_bl (Fig. 1), WTP was accordingly described as a function of p_bl (Fig. 2).

Table 2. Results of conditional logit models

Attribute	Linear model		Quadratic model
	Coef. (S.E.)	P‐value	Coef. (S.E.)	P‐value
Bird abundance in conifer plantations (per ha)
Linear term	0·0267 (0·0084)	0·002	0·5823 (0·0909)	0·000
Quadratic term (×10^−2)			−2·5493 (0·4153)	0·000
Number of bird‐watching stations in a district	0·0162 (0·0129)	0·280	−0·0002 (0·0131)	0·990
Additional amount of tax payments to introduce new forest management plans (×10^−2)	−0·0355 (0·0009)	0·000	−0·0353 (0·0009)	0·000
Alternative‐specific constant for status quo profile	−0·9488 (0·0550)	0·000	−0·5422 (0·0861)	0·000
Log‐likelihood		−7217·3		−7198·3
Adjusted log‐likelihood ratio indexa^a This index is called a McFadden's pseudo‐R^2, and values of 0·2–0·4 represent an excellent fit (McFadden 1979).		0·1410		0·1433
AIC		14442·6		14406·7

• Parameter estimates with positive and negative signs mean that the attributes influence respondents’ utility positively and negatively, respectively.
• ^a This index is called a McFadden's pseudo‐R², and values of 0·2–0·4 represent an excellent fit (McFadden 1979).

Figure 2

Willingness to pay (WTP) for birds and dependency of forest values on WTP, wood values and bird response forms. (a) We calculated bird values (per ha) by multiplying the WTP (per ha per person) by the total labour force in Hokkaido. (b–d) Five bird response forms led to varied forms of bird and forest values as a function of p_bl. Wood values decreased with increasing p_bl due to low prices of broadleaved trees. Net values (forest values) were obtained by summing bird and wood values. The deep convex form (the top line in Fig. 1) corresponds to the leftmost lines of bird and net values, while the deep concave form (the bottom line in Fig. 1) corresponds to the rightmost lines. Silvicultural costs were held constant at the most optimistic scenario (power = 1·0).

Optimization analysis for quantitative targets

Due to the nonlinear form of WTP, bird values as a function of p_bl were all unimodal for all five bird response forms, though the p_bl with maximized bird values and the deepness of the curvature varied among the response forms (Fig. 2b). In the standard scenario (Fig. 2b) with the factors of WTP, wood value, broadleaved tree value and silvicultural cost equal to 2/3, 1, 1 and 1, respectively, bird values at the highest p_bl (i.e. 1) were approximately equal to the wood value. Unimodal forms of bird values created a problem on how to maximize forest values from the identification of optimal p_bl to be less than 1. The convex response form had the maximized bird values at low p_bl and the corresponding net values (forest value) was the highest among the five forms since its wood value was also high. The concave form had the highest optimal p_bl and the lowest forest value at optimal p_bl. The linear form had the shallowest curves of bird and forest values. Shallow convex and concave forms also had shallow curves, indicating that they can attain comparable forest values at the broad ranges of p_bl within and across these forms although they have different specific optimal p_bl.

In the high‐WTP and low‐cost scenario (Fig. 2c) with the factors of WTP, wood value, broadleaved tree value and silvicultural cost equal to 1, 0·5, 1·5 and 1, respectively, bird values were larger than wood values for many levels of p_bl, and the decreased economic disadvantages of increasing p_bl made the maximized forest values comparable among the five response forms. Although the optimal p_bl of forest values was always smaller than that of bird values, the differences were small in this scenario. In the low‐WTP and high‐cost scenario (Fig. 2d) with the factors of WTP, wood value, broadleaved tree value and silvicultural cost equal to 1/3, 2, 0·5 and 1, respectively, optimal p_bl was more than zero only for the two convex forms. The linear form and the two concave forms had zero optimal p_bl since increasing bird values could not compensate for any forgone opportunity costs of wood values.

Our optimization analyses under varying conditions adequately showed the dependency of optimal p_bl on response forms and wood values (Figs 2-4). The results of the ANOVA suggested that wood values had the greatest effect on optimal p_bl (Table 3). When two intermediate (shallow convex and concave) response forms were excluded, bird response forms were the second most important (evaluated by mean SS), followed by WTP. Although silvicultural costs had minor effects, their increases actually decreased the optimal p_bl (Figs 2, 3). The mean values of optimal p_bl at low, medium and high silvicultural costs across the 405 combinations were 0·34, 0·30 and 0·28, respectively. The deep convex form was the most robust to the uncertainties in that optimal p_bl lay in the relatively narrow range of 0·02 to 0·22 (Fig. 4). Optimal p_bl of the linear form ranged from 0 to 0·78 and was greatly affected by the situations. In the deep concave response, there were just two available options: 0 and approximately 0·90. In the scenarios with the lowest wood values, 0·90 p_bl was always optimal except for three cases with the lowest WTP and broadleaved tree values. When wood values increased, however, the optimal p_bl suddenly jumped from 0·90 to 0. There were no intermediate options. Optimal p_bl of shallow convex and concave forms was higher and lower, respectively, than their respective deep forms.

Figure 3

Dependency of forest values on willingness to pay (WTP), wood values and bird response forms under the severe silvicultural costs. The most severe silvicultural costs were presumed (power = 2·0). (a–c) The same scenarios from Fig. 2 were used. The optimal proportions with the highest net forest values were decreased from Fig. 2. (d) The highest WTP and wood value as well as lowest broadleaved tree value were used.

Figure 4

Boxplots of optimal broadleaved tree proportion to maximize total benefits from the forests in relation to bird response forms and wood values. (a) Optimal broadleaved tree proportion was separated by five response forms. (b–c) Results of deep convex, linear and deep concave responses were also individually shown in relation to factors of wood values. Individual optimal values for the combination of four factors with three levels were depicted by grey circles.

Table 3. ANOVA table for optimal broadleaved tree proportion by five controlling variables as main effects

Variables	Data with five response typesa^a Overall data (405 combinations of five variables) were used.					Data with three response typesb^b Data for shallow convex and concave responses were excluded (with 243 combinations).
	d.f.	SS	Mean SS	F	P	d.f.	SS	Mean SS	F	P
Response type	4	3·2	0·80	18·5	0·000	2	2·8	1·39	30·0	0·000
Willingness to pay	2	3·9	1·95	44·8	0·000	2	2·5	1·24	26·6	0·000
Wood value	2	10·7	5·37	123·2	0·000	2	5·9	2·94	63·3	0·000
Broadleaved tree value	2	2·6	1·28	29·4	0·000	2	1·9	0·93	20·1	0·000
Silvicultural cost	2	0·3	0·13	3·0	0·052	2	0·1	0·04	0·9	0·417
Residuals	392	17·1	0·04	NA	NA	232	10·8	0·05	NA	NA

• ^a Overall data (405 combinations of five variables) were used.
• ^b Data for shallow convex and concave responses were excluded (with 243 combinations).

Discussion

To the best of our knowledge, this is the first study to identify quantitative conservation targets (optimal plantation intensity) to reconcile biodiversity (bird abundance) conservation and resource production by measuring WTP for biodiversity conservation. Although it has been difficult to set quantitative targets without clear thresholds on an ecological basis (Ficetola & Denoël 2009), we successfully identified optimal targets on an economic basis. Our analysis showed that the optimal plantation intensity depends on a range of factors and found a value of semi‐natural plantations (plantations with broadleaved trees) as social capital in many cases. Semi‐natural plantations are optimal forest types that maximize the total benefits of bird conservation and wood production unless bird–broadleaved tree proportion relationships are concave or unless relationships are linear and wood values greatly exceed the WTP for bird abundance.

Given the most likely scenario (i.e. a linear bird response, present wood values and 2/3–1 WTP), an approximately 50% coverage of broadleaved trees was suggested to be optimal (Fig. 3c). However, we only quantified opportunity costs required to maintain semi‐natural plantation forests by the presumed decreases in the amounts of planted coniferous trees. If we measure opportunity costs empirically and consider them, the optimal broadleaved tree proportion may be lower than that of this study. For example, Hunter (1990) and Newton (1994) suggested a retention of 5–10 snags per ha to maintain cavity users, and Gustafsson et al. (2012) suggested tree retention levels between 5 and 10% to reduce the ecological impacts of forest harvest. Similar retention levels may be feasible targets under varied opportunity costs.

Optimal broadleaved tree proportion never reached 1, suggesting that strict nature reserves excluding any human interventions are always a suboptimal option to maximize forest values. Optimal broadleaved tree proportion was approximately 90% rather than 100%, even under a deep concave response due to the diminishing returns of conserved bird abundance. It is suggested that the marginal benefits of further bird conservation do not greatly increase given that certain amounts of bird abundance are already conserved. To this extent, the exploitation of forests (and resulting slight loss in bird abundance) would be socially accepted. This idea is already acknowledged and manifested by the existence of multiple types of nature reserves (Cumming et al. 2015).

Our analyses also showed situations in which the advantages of semi‐natural plantations are limited. These occur when bird abundance responds to broadleaved tree proportion in a concave way or the value of birds is greatly restricted compared to that of resource production. In such cases, separation of biodiversity conservation and resource production (land sparing or land‐use specialization) is superior to their integration on the same land parcels (land sharing or multiple uses). The same observations and conclusions were also made by previous modelling studies (Green et al. 2005; Perfecto et al. 2005; Butsic & Kuemmerle 2015). Although optimal options can be suddenly shifted even due to small changes in resource values (Fig. 4), the existence and nature of such thresholds can be predicted a priori using the models.

What makes this study distinct from others is that we have formulated economic values of biodiversity conservation as a function of land‐use intensity. Without this formulation, we would have to seek land‐use strategies that reconcile biodiversity conservation and resource production by maximizing the resource production while specifying the quantitative target of biodiversity conservation. However, specifying the quantitative target of biodiversity conservation can also be difficult (e.g. Flather et al. 2011); we circumvent this problem by evaluating monetary values of two conflicting products. This approach enables us to compare values of biodiversity conservation and resource production, and to manage landscapes to maximize social benefits (see also below).

Conservation implications

We showed that when species response forms approach the linear form, we are able to obtain comparable benefits from forests at the broad ranges of intensity. When linear response forms are competing with other models, to increase the social benefits and the likelihood of an adopted strategy to be successful, it would be useful to adopt a feasible land‐use strategy in focal landscapes. Another consideration is the number of products involved. We only dealt with two conflicting products, and it can be difficult to evaluate more than seven products simultaneously with a choice experiment (Miller 1956). However, only if we evaluate additional products in monetary terms (e.g. carbon, water use), our framework can be extended to the landscape‐level problem with more than two products (see Appendix S2). Different products are likely to be functions of different land uses (Raudsepp‐Hearne, Peterson & Bennett 2010; Crouzat et al. 2015), and their benefits are also likely to show diminishing returns (Field 2008; Zhang & Pearse 2011); we suggest that a flexible hybrid approach may be a sustainable option for real landscapes in varied ecological and social contexts (Wilson et al. 2007; Yamaura et al. 2012; Butsic & Kuemmerle 2015).

Acknowledgements

Comments from anonymous four reviewers, José Alves and J. Barlow greatly improved the manuscript. We also thank M. Forsyth to improve our English expressions. This research was supported by JSPS KAKENHI Grant Number 24310029. Y. Yamaura was supported by the Mitsui & Co., Ltd. Environment Fund Grant Number R12‐G2‐225.

Data accessibility

R scripts for the optimization analysis: uploaded as online supporting information.

Data related to the CE: Dryad Digital Repository doi: 10.5061/dryad.5145k (Yamaura et al. 2016).