Introduction

There is a pressing need to develop environmentally and ecologically sustainable economic systems based on renewable energy and materials, as well as for restoration of degraded systems to halt species extinctions. Forests are a primary source of renewable carbon (C)-based materials needed to support the emerging bio-based economy that is decoupled from fossil fuels (WWF 2012b; European Commission. Directorate-General for Research Innovation 2018). Therefore, annual global wood removals from forests will likely need to increase three-fold by 2050 compared to 2010, to meet the future demand for solid wood, paper, and bioenergy products (WWF 2012a). Global wood production is currently dominated by countries in boreal and temperate zones (FAO 2018), but much of the increase in future demand will need to be met by tropical forests (FSC 2012) because of their vast land area (FAO 2020) and high production potential (Running et al. 2004). Southeast Asian forests will be subject to particularly high demand because of their proximity to the world’s largest consumer of wood products, China, and other growing markets like India (Ajani 2011; Malhi et al. 2014).

Historic ineffectiveness of natural regeneration of tropical forests by regrowth after logging (Chua et al. 2013; Yamada et al. 2016; Butarbutar et al. 2019) has switched focus from natural forest management towards tree plantations. Management practices in these plantations are intended to maximize production of renewable bio-based materials by nutrient optimization (achieved by fertilization) and by selectively planting often non-native fast-growing species (Putz et al. 2000; Sist et al. 2003; Curran et al. 2004; Hansen et al. 2008). Tree plantations in Southeast Asia are dominated by exotic Acacia and Eucalyptus species that are intensively managed and optimized for fast growth (Fah et al. 2008; Hardiyanto and Nambiar 2014; Harwood and Nambiar 2014a, b). Compared to natural forests, exotic tree plantations can have negative effects on local biodiversity (Barlow et al. 2007) and soil C stocks (Guillaume et al. 2018). Some could turn into invasive species, thereby leading to negative effects on biodiversity within, as well as outside plantations (Osunkoya et al. 2005). Furthermore, many second and third generation Acacia mangium plantations in Malaysia and Indonesia are suffering from severe pathogen attacks, resulting in heavily reduced outputs (Harwood and Nambiar 2014a; Nasution et al. 2019) and highligting a great need to diversify the portfolio of plantation species.

An alternative to exotic plantations is to use native tree species. Their adoption can reduce trade-offs and achieve synergies between raw material production and supply of ecosystem services within a bio-based economy (Piotto et al. 2004; Lamb et al. 2005; Subiakto et al. 2016; Crous et al. 2017; Bieng et al. 2021). Although it is often assumed that native tree species tend to be slower growing than many exotic tree species (Piotto et al. 2004; Lamb et al. 2005), native tree species may provide raw materials that are more valuable. As native species also share an evolutionary history with native ecosystems, they may be better adapted to pests and pathogens, and have a higher value for promoting native flora and fauna (Lamb 1998; Lamb et al. 2005; Crous et al. 2017), all while maintaining relatively high growth rates (Davis et al. 2012; Subiakto et al. 2016). Hence, it has been proposed that wider use of native trees for timber production could sustainably meet the growing demand for raw materials while ameliorating negative effects on other ecosystem services resulting from planting exotic species (Lamb et al. 2005; Davis et al. 2012). However, more research is needed to identify suitable native species for this purpose and to further determine whether they are financially viable alternatives (Lamb et al. 2005; Bieng et al. 2021).

Developing native tree species as economically viable options could also enable forest landscape restoration approaches that provide conservation benefits while simultaneously supporting social and economic objectives in accordance with the twin objectives of the Bonn challenge. Native species could be used over a gradient of land-use types ranging from intensively managed commercial plantations to enrichment plantings for restoration. However, there is little documented knowledge on growth potential and form, wood characteristics, correct management (Haggar et al. 1998; Piotto et al. 2004; Lamb et al. 2005), and genetic potential of native species (Thomas et al. 2014; Axelsson et al. 2020). Therefore, if native tree species are to be widely used for timber production, it is necessary to first identify which species are most suitable (Lamb et al. 2005). This is a challenging task given that the Indo-Pacific region, including Southeast Asia, has been estimated to contain between 19,000 and 25,000 tree species (Slik et al. 2015), each with unique adaptations that are not well known. Understanding of tree species’ adaptations to different environments can be derived from their traits, which could allow predicting function of a broad set of species from their composition of “functional traits”.

Plant functional traits are morphological, physiological, and phenological qualities that affect the organism’s growth, reproduction, and survival, thereby determining performance across variations in environmental conditions (Violle et al. 2007). For trees, commonly measured traits include specific leaf area, wood density, and leaf nitrogen content (Perez-Harguindeguy et al. 2016), all of which correspond to plant functions important for fitness.

Plant functional traits can be used to describe a tree species’ characteristics, ecosystem niche, and location in the “plant economics spectrum”, the latter which spans from traits associated with acquisitive to conservative growth strategies and incorporates a trade-off between the two (Violle et al. 2007; Reich 2014; Funk et al. 2017). Acquisitive traits include a high specific leaf area, pH, and high contents of nitrogen, phosphorus, and cations (Cornelissen et al. 2006; Patiño et al. 2012; Reich 2014; Perez-Harguindeguy et al. 2016). For example, specific leaf area is a proxy for resource investment per leaf area, whereby a high specific leaf area is associated with a high rate of resource capture and growth at the expense of leaf defense (Kurokawa and Nakashizuka 2008; Reich 2014).

A tradeoff between plant investments in defense versus growth can occur across different ecosystems, driven by gradients of resource availability (Endara and Coley 2011). Species adapted to low resource environments might thus invest relatively more resources in chemical and structural defense, such as high phenolic content, than in growth, compared to species in more resource rich environments (Jones and Hartley 1999; Endara and Coley 2011). Additionally, many species who invest in high growth rates early also have higher early life mortality (Sterck et al. 2006; Philipson et al. 2014; Gustafsson et al. 2016).

The allometry of the tree, here defined in terms of the relationships between stem diameter, tree height, and crown size, are life-history traits that also affect plant performance. Allometry can be strongly associated with a species’ resource capture strategy and early growth rates. Young trees with wide, flat crowns and thick stems that can support this crown shape may effectively capture light entering vertically through the canopy, and often have higher growth rates than trees with thinner crowns and more leaf layers (Sterck et al. 2001).

Functional traits are widely used in ecology to predict diverse ecological functions, including growth and survival (Wright 2010; Wills et al. 2018), drought tolerance (Harrison and LaForgia 2019), and interactions with associated communities (Kraft et al. 2008; Wright 2010; Wills et al. 2018; Harrison and LaForgia 2019; Axelsson et al. 2022). However, despite showing a strong correlation to productivity parameters such as growth and survival, functional traits have rarely been combined with estimates of financial value, especially for tropically indigenous tree species. Yet, there are many possible relationships and tradeoffs between financial value and traits, for example, variables capturing growth are consistently associated with financial performance. For instance, Lopez et al. (2010) found that mean annual increment explained most of the variation in financial performance of pine plantations in Colombia. Similarly, survival rates directly affect financial value by determining the quantity of resources needed for planting, re-planting, and other managerial adaptions to avoid high mortality of the planted material. Additionally, defensive compounds, such as phenolics, along with structural traits, such as wood density, are related to timber properties like resistance to decay, mechanical strength, and general durability, which in turn could contribute to define the market price of timber per unit volume (Chave et al. 2009; Kulbat 2016).

While reports on financial performance exist (for example Cubbage et al. 2007; Cuong et al. 2020), few comparisons between plant traits and financial relationships across species are available. Such an approach could potentially be used to identify suites of native tree traits that provide favorable balances between timber production and other ecosystem functions and services in species-rich tropical forests. In this study, we evaluated functional traits, estimated financial values, and calculated their associations for different species under similar ecological conditions.

We used a unique field trial with a range of native tree species in the Malaysian state of Sabah, northern Borneo (Gustafsson et al. 2016)—a global biodiversity hotspot with around 3000 native tree species (MacKinnon 1996; Myers et al. 2000). In this common garden, we measured 18 functional traits of 22 native tree species, examining the functional diversity within the group. We assessed the financial viability of each tree species in enrichment planting in a secondary forest by estimating their land expectation value, which was used to derive an internal rate of return under restoration treatment in the field trial site. Additionally, we conducted a sensitivity analysis to assess the robustness of these estimates, and to determine whether native tree species can be financially viable under different scenarios of restoration management.

We hypothesized that the financial value of trees could be predicted from a suite of functional traits. Specifically, we hypothesized that trees expressing high-value traits on the acquisitive side of the plant economics spectrum would have high financial value due to their higher growth rates. According to this hypothesis, high financial value was expected to be associated with a high specific leaf area and pH, high contents of nitrogen, phosphorus, and cations, and low wood density, leaf dry matter content, phenolic content, and leaf thickness (Cornelissen et al. 2006; Patiño et al. 2012; Reich 2014; Perez-Harguindeguy et al. 2016). High growth rates of species with these traits were expected to outweigh the potentially greater market value of timber from slow-growing conservative species with denser and more durable wood. Furthermore, with respect to tree allometry, we hypothesized that tree species that maximize capture of sunlight penetrating the canopy vertically would have the highest growth rates (Sterck et al. 2001). Such trees have crowns that are broad relative to the crown height (meaning a low relative crown length), but are also thin and short relative to the tree’s total height, meaning that they have a low relative crown width and relative crown depth (Sterck et al. 2001). In addition, since the crown is small and concentrated at the top of the tree, they have stems that are thin relative to the tree height (i.e., high stem slenderness). We tested these hypotheses and used the results obtained to analyze relationships between plant traits associated with financial value in the studied native tree species and ecosystem processes and functions.

Methods

Study site

This study was conducted within the 26,880 ha Sungai Tiagau Protection Forest Reserve in Sabah, Borneo, Malaysia (4°35′20″N 117°16′19″E). According to the Köppen-Geiger climate classification, the region’s climate is humid tropical equatorial (Peel et al. 2007), with a mean annual temperature of 27.6 °C (Walsh and Newbery 1999). The annual average precipitation (2004–2016) is 2400–2500 mm, based on data from three weather stations within a 15 km radius of the study location. The Sungai Tiagau Protection Forest Reserve consists of lowland mixed dipterocarp forest, dominated by Acrisols (Panagos et al. 2011). It was logged in the 1970s and burned during a forest fire in 1982–83. The forest reserve is managed by the state government foundation Yayasan Sabah. Since 1998, 18,500 ha of the site has been managed by the Innoprise-IKEA (INIKEA) Forest Rehabilitation Project, a collaboration between Yayasan Sabah, the Swedish University of Agricultural Sciences, and the Swedish company IKEA. The goals of the INIKEA Forest Rehabilitation Project are to rehabilitate the forest and to increase biodiversity, and efforts to achieve these goals have resulted in the planting of more than 90 native tree species and 4 million seedlings since its inception.

In November 2008, we established a tree species field trial with 34 different trees species within the INIKEA project area by raising 680 seedlings from seeds in the project plant nursery. These seedlings were then acclimatized to forest conditions in a second nursery close to the forest field site (Gustafsson et al. 2016) before being planted in the 3 ha forest site, located at ~ 300 m a.s.l. Within species, seedlings were of approximately the same size at the time of plantation (Gustafsson et al. 2016). Between species, the average height varied between 30 cm (Hopea ferruginea) and 105 cm (Pentace laxiflora) at the time of planting. The tree species were planted in a random order with one tree every 3 m along a total of 40 linear transects. Planting was conducted to avoid rocks, streams, and slopes exceeding 15% inclination, extending the transects when needed. Each transect was 2 m wide, 60–100 m long, and the distance between the center of transects was 10 m (Gustafsson et al. 2016), hence, the planting density was ca. 333 trees/ha. All transects are positioned perpendicularly to a stream and most of the study site was located on a south-facing slope, rising approximately 100 m above stream elevation (Gustafsson et al. 2016). Before planting, transects were cleared of bushes, small pioneer trees and climbers. No prior soil preparation was conducted; seedlings were planted directly into the forest soil. Four years after planting, climbers were removed and a few pioneer trees growing at the edges of the transects were girdled to promote growth of the planted seedlings (Gustafsson et al. 2016). The planting regime used for the study site is identical to enrichment planting done in large swathes of the INIKEA-project, and similar to methods in other regional restoration projects (Philipson et al. 2020; Ruslandi et al. 2017). Ideally, natural mortality rates among seedlings and clearing of nearby climbers and pioneer trees limit competition and enable growth of the surviving planted material.

Sampling and trait measurement

Due to resource restraints, not all 34 species present in the common garden were sampled and measured. The 22 species sampled were selected randomly out of the pool of 34 total species. We characterized each selected tree species (n = 3–8 individuals per species) using a combination of 18 physical, chemical, and tree allometry traits (Table 1). Physical traits included specific leaf area (SLA), leaf dry matter content (LDMC), leaf area (LA), leaf thickness (LT), and wood density (WD). Chemical traits constituted concentrations of leaf nitrogen (N), phosphorus (P), potassium (K), calcium (Ca), magnesium (Mg), sodium (Na), and total and soluable phenolics, as well as pH. Tree allometry traits were stem slenderness, and relative crown length, depth, and width (Table 1). Traits are expressed as average values for each species. For physical and allometric traits, this value constitutes the arithmetic mean of measurements from all sampled individuals of the species. For chemical traits, leaves from all sampled individuals of a species were pooled and homogenized into one sample prior to analysis. Hence, the final sample size for each individual trait was 22.

Table 1 List of trait categories, associated sampled traits, their units of measurement, and the biological relevance of each trait
Full size table
Table 2 Data for 22 tree species that either contributed to or constitute our estimates of financial value
Full size table

During February–April 2019, 10 years and 2–4 four months after the planting of the trees, leaves were collected from 3 to 8 individuals of each of the 22 tree species, and allometric traits were measured (studied species are listed in Table 2; short descriptions of the silvicultural and ecological relevance of major taxonomic groups are presented in Box 1). Of these 22 species, 17 belonged to the Dipterocarpaceae family, which is the region’s dominant tree family (Symington et al. 2004; Ghazoul 2016). All 22 tree species were classified into local timber groups; trade names used to group closely related timbers with similar properties and appearance since the high number of tree species in the region would make it impractical to trade each on a per-species basis (Wong 1982). For each individual tree, we measured diameter at breast height (dbh, 1.3 m), tree height, and crown height, together with two perpendicular measurements of crown width, using measurement tape and a laser rangefinder (Nikon Forestry Pro). From these measurements we calculated the following tree allometry traits: tree height: diameter at breast height ratio (stem slenderness), crown height: crown width ratio (relative crown length), crown height: tree height ratio (relative crown depth), and crown width: tree height ratio (relative crown width) (Sterck et al. 2001). Leaf sampling was conducted by using ladders (approximately 10 m) and cutting or twisting branches or leaves off using a telescopic branch cutter or an approximately 5 m long stick with a forked ending. Sampling was performed between 8:00 AM and 1:00 PM. For each tree, leaves were collected (n = 5) from three different branches located in the outer canopy whenever possible. For very large trees where the outer canopy could not be reached, leaves from the inner canopy were collected instead. In cases where trees had compound leaves, we instead collected one leaflet per branch. Where possible, we sampled leaves that were undamaged by herbivores and not covered by lichen. If it was impossible to obtain enough material satisfying these criteria, the least damaged leaves were sampled. Immediately after sampling, leaves were wrapped in wet napkins and placed in plastic bags. Before sealing the bags, we exhaled into them to increase their CO2 concentration and moisture (Perez-Harguindeguy et al. 2016). Bags were then stored in an ice-cooled plastic box until trait measurements were taken later on the same day (Perez-Harguindeguy et al. 2016). Where necessary, lichen and dirt were removed from leaves before measuring traits (Perez-Harguindeguy et al. 2016). Physical trait measurements were performed as soon as the samples were retrieved from the field. First, LT was measured between major veins in the bottom, middle, and top thirds of each leaf (Perez-Harguindeguy et al. 2016) using a digital micrometer (RS Components Ltd., Corby, UK). Leaves were then scanned (Canon E410 series, Canon Inc., Tokyo, Japan) and LA determined by image analysis using the ImageJ software package (Schneider et al. 2012). Fresh and dry weight (after oven drying for 3 days at 70 °C) was determined for all samples (Perez-Harguindeguy et al. 2016). For each sample, fresh and dry weight measurements were used to determine LDMC, and SLA was calculated from the LA and dry weight (Perez-Harguindeguy et al. 2016).

Box 1 Relevance of studied tree species’ taxonomic groups
Full size table

Dried leaf samples were transported to the Forest Research Center of the Sabah Forestry Department in Sepilok, Malaysia, for all chemical trait analyses, where they were bulked by species, air dried to a constant weight, and ground using a Wiley Mill ED-5 (Thomas Scientific, US). A subset of each sample was dried at 105 °C, weighed, and used to convert all data to oven-dry values. Bulk samples were then analyzed for pH, soluble phenolics, total phenolics, as well as total contents of N, P, Mg, K, Ca, and Na. Specifically, for each ground sample, leaf tissue pH was analyzed by mixing a subsample in distilled water (volume ratio 1:8) and measuring the resulting pH using a Mettler Toledo inLab pH electrode (Greifensee, Switzerland) connected to a Corning pH meter 240 (Corning Life Sciences, Texas, US) (Cornelissen et al. 2006). Another subsample was used to measure total and soluble phenolics by extraction in methanol and deionized-water, respectively, both at a ratio of 1:50 (Gundale et al. 2010), and extracts were analysed by the Folin-Ciocalteu assay at 765 nm on a Hitachi U-2900 UV/Vis Spectrophotometer (Tokyo, Japan) (Singleton and Rossi 1965). Total N contents were determined by dry combustion at 900 °C on Vario Max CN Elemental Analyzer (Elementar Analysensysteme, Germany). To determine P, Mg, K, Ca, and Na contents of each sample, subsamples were digested using the sulfuric acid-hydrogen peroxide method (Allen 1989). Additionally, Mg, K, Ca, and Na content was measured on a Spectro Arcos ICP-OES (Spectro Analytical Instruments, Kleve, Germany), and P content was determined using the molybdenum-blue method (Anderson and Ingram 1993) at 880 nm with a Hitachi U-2900 UV/Vis Spectrophotometer (Tokyo, Japan). For Hopea ferruginea, Pentace laxiflora, Shorea parvifolia and Sindora irpicina, bulked samples were too small to perform all analyses. In these cases, additional leaves were collected, weighed, and dried on 29 March 2019 and analyzed as described above.

We used the Global Wood Density Database to acquire wood density values for 16 of the 22 species (Zanne et al. 2009). The remaining six species, Canarium sp., Diospyros sp., Pentace adenophora, Pentace laxiflora, and Sindora irpicina, were not in the database. In these cases we instead used average wood densities for species of the same genera growing in tropical Southeast Asia.

Growth rate estimation

To estimate growth rates for the 22 species, we used dbh measurements collected on all planted trees growing at the study site. These measurements were taken on seven occasions, starting 3 years after planting, and then repeated roughly every second year, up to 14 years and 5 months after planting. For each species, we generated linear mixed effect models with dbh as the response variable and number of years since planting as the explanatory variable, with random intercepts and slopes based on tree identity (i.e., replicate). Using the coefficients from the generated models, we extrapolated the average time it would take each species to reach a dbh of 30 cm and defined this as one rotation period. This dbh value is below the lower dbh limit of 60 cm stipulated by the current regulations concerning logging in natural forests in Sabah (Berry et al. 2010; Reynolds et al. 2011; Bryan et al. 2013) but was chosen because the volumes extracted under this cutting limit would better match logging intensities seen in the region. Harvesting up to 333 trees at 60 cm dbh, after adjusting for survival rates presented in Table 2, would result in average wood extractions of 783 m3/ha. This greatly exceed the average volume of 152 m3/ha extracted by conventional logging in the region (Fisher et al. 2011), and those extracted from timber plantations, which rarely exceed 400 m3/ha (Cuong et al. 2020). Using 30 cm as the cutting limit gave average wood extractions of 154 m3/ha, thus enabling a more realistic comparison to conventional logging and plantation practices.

Internal rates of return

Internal rate of return (IRR) corresponds to the estimated discount rate at which net present value of all cash flows equal zero, and helps define the total rate of return from investment in a project and compare it against the opportunity cost of capital. It is commonly used to rank different investment alternatives where the higher the IRR, the more profitable the investment. Here, IRR is given per hectare for each tree species and was estimated based on calculations of land expectation value (LEV, USD/ha), a standard discounted cash flow technique in timberland appraisal, used to determine financial value of a patch of land, assuming the land will be under this forest management indefinitely. The LEV, and subsequently IRR, was calculated following Straka and Bullard (1996) in (1):

$$LEV = \frac{{\sum_0^T {R_t \left( {1 + i} \right)^{T - t} } - \sum_0^T {C_t \left( {1 + i} \right)^{T - t} } }}{(1 + i)^T - 1}.$$
(1)

where ‘Rt’ is the revenue received in year t, ‘Ct’ is the cost incurred in year t, ‘T’ is the rotation length in years, ‘t’ is the year of a particular revenue or cost, and ‘i’ is the annualized discount rate, expressed as a decimal. The internal rate of return is defined as the value of ‘i’ where LEV = 0.

Assuming that the planting pattern does not change between rotations, 333 trees could be grown on a hectare of land. Additionally, in each calculation we assumed that all 333 trees would be of the same species. Lastly, we assumed that all trees in a hectare could be harvested simultaneously when they reach an average of 30 cm dbh. Tree survival rates were calculated for each species based on their observed survival between planting and the most recent plot survey (a 15-year period), and were used to adjust the number of trees remaining at harvest time, assuming no further mortality after 15 years.

Financial value of each species per unit volume of timber was based on the price per cubic meter of exported round logs, free on board, reported by the Sabah Forestry Department (2016, 2017). Price per cubic meter is reported for timber groups used in the local region based on similarities in timber properties and phylogenetic relatedness. Timber group identities for the 22 tree species were retrieved from Soepadmo and Wong (1995), Soepadmo et al. (1996), Soepadmo and Saw (2000), Soepadmo et al. (2002), Lee (2003), Soepadmo et al. (2004), and Soepadmo et al. (2007). All price data were taken from the Sabah Forestry Department 2017 report except in the case of Canarium sp., which were only mentioned in the 2016 report. Values from the reports were adjusted for inflation and converted to spring 2023 USD via consumer price index (CPI) and exchange rates (MYR CPI mid 2016/2017 − spring 2023 = 1:1.13/1.09 = 0.229 USD).

We used allometric volume equations obtained from the Indonesian STREK-project (Bertault and Kadir 1998) to calculate tree volume at the 30 cm dbh threshold for each tree species. Volume was calculated in (2):

$$V=A+B*dbh+C*{dbh}^{2}$$
(2)

where ‘V’ is the volume given in m3, ‘dbh’ is the diameter at breast height (1.3 m, cm), and ‘A’, ‘B’, and ‘C’ are constants estimated through field measurements from the STREK-project (Bertault and Kadir 1998). The STREK-project produced species- and genera-specific constants, together with an “others”-category based on measurements on species for which specific equations could not be made due to a lack of measurements. For the 17 dipterocarp species in this study, we used species- or genera-specific constants. However, no such constants were available for the five non-dipterocarp species, so the non-specific equation constants were used in these cases. The complete list of equation constants can be found in Lussetti et al. (2016).

Management costs were taken from Philipson et al. (2020), which includes expenditures from site preparation, seed collection, seedling raising, planting, post site management, and replanting. These costs are based on a separate restoration project consistent with those incurred during the INIKEA project (personal communication, David Alloysius, manager of INIKEA). Logging costs were taken from FRIM and ITTO (2002) and Fisher et al. (2011), and represent average logging costs per m3 timber in conventional logging operations in Peninsular Malaysia. Logging costs include operations such as road construction, felling, skidding, as well as transport to the mill, taxation, administration, and other expenditures (FRIM and ITTO 2002; Fisher et al. 2011). Cost estimates for each activity is avaliable in Table 3. All costs were adjusted for inflation and converted to USD (MYR CPI mid 2001 − spring 2023 = 1:1.59 = 0.229 USD).

To capture variation in our data and evaluate robustness of our IRR values, we performed a sensitivity analysis by varying management costs and tree growth rates. For each of these variables we used a high, average, and low value, and produced three scenarios representing the upper, middle, and lower limit of our IRR estimates. Philipson et al. (2020) estimated management costs between 1500 and 2500 USD/ha, depending on the intensity of site preparation. For the sensitivity analysis, we used management costs of 1500, 2000, and 2500 USD/ha for the high, mid, and low-value scenarios, respectively. These values where then adjusted for inflation, which gave management costs of 1742, 2323, and 2903 USD/ha for each respective scenario (MYR CPI mid 2020 − spring 2023 = 1:1.16). We assumed that 80% of these costs would be used in the first year to cover field preparation and seedling planting, and that the remaining 20% would be spread out during the following nine years, covering site maintenance, for a total management period of 10 years. To account for variation in tree growth rates, we used the bounds of the 80% confidence interval in our mixed effects models. The lower bound of the confidence interval represents the shortest growth time needed to reach the cutting limit and was thus used in the high-value scenario, while the average growth rate and the upper bound of the interval were used in the mid and low-value scenarios, respectively.

A full list of species names, species acronyms, timber group identity, wood market value, growth rate, survival, volume of wood per tree, and time needed to reach 30 cm dbh, for each species and scenario from the sensitivity analysis is available in Table 2.

Statistical analyses

To evaluate variation in plant functional traits in relation to financial value within the studied species, we performed a multivariate partial least squares (PLS) regression in Simca (version 17.0.0.24543 (64-bit), Sartorius Stedim Data Analytics AB 2021). We included the mid-scenario estimate of IRR as the response variable and the 18 measured traits as explanatory variables for all 22 species. To determine how many components to include we used the Q2 value, the fraction of the total variation of IRR that could be predicted by a component, produced by cross-validation. We only considered components with Q2 values above 5%. Variables were considered significant if the 90% confidence intervals of their loadings obtained by jack-knifing (Quenouille 1949; Efron and Gong 1983) during cross-validation did not overlap with zero. The best-fitting model was determined by backward selection, removing the trait with the confidence interval furthest from zero in each model, one variable at the time. We continued this way until the model only contained significant variables and then compared Q2 values of the resulting models. The model with the highest Q2 value was deemed the best fitting and was kept while others were discarded.

In addition, to distinguish between direct and indirect effects traits had on IRR, we ran four additional PLS models, following the same procedure as in the trait-IRR model. Each model was run with the same traits as explanatory variables, but one of the four variables that determined IRR as the response variable (e.g., time needed to reach 30 cm dbh, survival rate, market value per cubic meter of wood, and volume of wood per tree at the 30 cm dbh threshold).

To test which variables in the IRR equation were the most influential, we used simple linear regression models with IRR as the response variable. Explanatory variables were survival rate, time needed to reach 30 cm dbh, i.e., the reverse of growth rate, market price for a cubic meter of wood, and volume of wood per tree at the given cutting limit of 30 cm dbh. Additionally, to improve the model assumption of normally distributed data, survival rate was logit transformed and time needed to reach 30 cm dbh was log transformed. In order to find the most appropriate model we performed a backward selection, starting with a model containing all four variables and their interactions, excluding three- or four-way interactions. We removed one variable at a time and evaluated model fit using AIC values adjusted for small sample sizes, AICc (Hurvich and Tsai 1989), with the model with the lowest value being retained. This continued until the AICc was no longer improved by removing additional variables. We interpreted models using a type III anova, using Kenward–Rogers “F” tests with Satterthwaite’s degrees of freedom and restricted maximum likelihood estimation, which are especially appropriate for small sample sizes (Fox and Weisberg 2019; Kenward and Rogers 1997; Satterthwaite 1941). We validated the final model via visual inspection of residuals distribution and linearity assumption for the variables and interactions. While the model for low levels of survival showed signs of non-linearity in the interaction term with the time needed to reach the set cutting limit of 30 cm dbh, this was likely due to a few species having very low survival rates compared to the rest. However, the R2-value for the overall model remained above 95% without including non-linear terms. Therefore, we did not include any quadratic terms in the model.

Results

Tree species’ functional traits associated with financial viability

Our first PLS model, including 22 tree species, explained 43.6% of the variation in plant functional traits, and 23.3% of the variation in IRR with plant functional traits by one significant component (Q2, predicted variation by the single component = 11.6%, Fig. 1a). According to the model, IRR was negatively correlated with leaf Ca, Mg, N content, pH, and SLA (Fig. 1b). Leaf area and relative crown length were also included in the model but did not significantly contribute to explained variance, i.e., the loading 90% confidence interval overlapped with zero.

Fig. 1
figure 1

Results from a partial least square model (PLS) including data for 17 Dipterocarpaceae species, and 5 non-dipterocarp species grown in a common garden experiment with active restoration management in Sabah, Malaysia. Plot a shows observed and predicted internal rate of returns (IRR) based on the land expectation value of the study system. Observed values represent mid-scenario IRR values obtained from a sensitivity analysis of costs and revenue from managing trees planted as part of the restoration effort. Predicted IRR values are based on values of functional traits included in each model after backward selection, taken from a pool of 18 traits measured for each tree species (n = 22 per trait). The dotted line represents a 1:1 correspondence between measured and predicted values. Tree species are denoted using acronyms of their scientific names (Table 2). Plot b shows PLS loadings. Orange bars correspond to functional traits measured for each tree species, chosen via backwards selection from a pool of 18 traits. The blue bar represents the same IRR values as the observed IRR in plot a, i.e., the mid-scenario values obtained in a sensitivity analysis of financial viability of the restoration planting. Loadings indicate the importance and direction of influence of variables contributing to the component in question; large loadings, either positive or negative, indicate substantial influence on the model. Error bars represent 90% confidence intervals. A variable was considered to significantly improve the model’s prediction if its confidence interval did not overlap with zero. Significant variables are highlighted with an asterisk

Full size image

Our four additional models that each includes one of the four variables used to calculate IRR, could all explain their respective response variable with one significant component (Table 4; Figure 4a, c, e, and g). In the first model, high values in time needed to reach the cutting limit of 30 cm dbh was positively correlated with LT, LDMC, relative crown width, and stem slenderness, and negatively correlated with leaf N (Fig. 4b). In the second model, survival rate was positively correlated with LT, relative crown depth, relative crown length, and relative crown width (Fig. 4d). In the third model, market price per cubic meter of wood was positively correlated with stem slenderness, soluble phenolics, and WD, and negatively correlated with leaf Ca, Mg, K, and N content (Fig. 4f). Lastly, in the fourth model, volume of wood per tree at the 30 cm dbh threshold was negatively correlated with leaf K and Mg content, as well as LT (Fig. 4h). In all four models additional traits were included, but these did not significantly contribute to the explained variance (Fig. 4b, d, f, and h).

Distribution of financial value in sensitivity analysis

In our mid financial value scenario (for which moderate growth rates and management costs were assumed), the average IRR was 2.58%. Shorea macrophylla had the highest IRR (4.89%), and Sindora irpicina the lowest (0.96%) (Table 2; Fig. 2a).

Fig. 2
figure 2

Internal rate of return (IRR) for each tree species and each scenario included in the sensitivity analysis, in which the assumed growth rate and management costs were varied. Green bars indicate species in the Dipterocarpaceae family; yellow bars indicate non-dipterocarp species. Components contributing to IRR and their exact values are listed in Table 2. Tree species are denoted using acronyms of their scientific names (Table 2). Species in b, c are sorted after IRR values in a

Full size image

In the high financial value scenario (high growth rates and low management costs), the average IRR was 4.07%. As for the mid financial scenario, the species with the highest IRR was Shorea macrophylla (7.8%) while Sindora irpicina had the lowest IRR (1.51%) (Table 2; Fig. 2b).

Finally, in the low financial value scenario (low growth rates and high management costs), the average IRR was 1.6%. The highest IRR was obtained for Shorea ovalis (3.15%) and the lowest for Shorea parvifolia (0.03%), (Table 2; Fig. 2c).

A simple linear regression model with survival rate, time needed to reach the set cutting limit of 30 cm dbh, market price per cubic meter, and number of cubic meters of wood per tree as explaining variables could explain 97% of the variation in IRR (F (5, 16) = 158, p < 0.0001). Internal rate of return was positively related to the survival rate (F = 305, p < 0.0001), the market price per cubic meter of wood (F = 71, p < 0.0001), and the number of cubic meters of wood per tree at time of harvest (F = 20, p = 0.0004). Additionally, IRR was negatively related to the time needed to reach the set cutting limit of 30 cm dbh (F = 445, p < 0.0001), i.e., the shorter the rotation time, the higher the IRR. Furthermore, there was an interaction effect between survival and growth rate (F = 230, p < 0.0001). Each additional unit of survival rate increased the negative relationship between IRR and time needed to reach the cutting limit of 30 cm dbh, and vice versa, i.e., decreasing time needed to reach the cutting limit of 30 cm dbh increased the positive relationship between IRR and survival rate (Fig. 3). For example, an increase in survival rate of one standard deviation, while time needed to reach the cutting limit was fixed at 33 years (ln = 3.5), would increase IRR by approximately 1.3% (Fig. 3).

Fig. 3
figure 3

Linear regression analysis, showing the relationship between internal rate of return (IRR) and numbers of years needed to reach the size threshold for when harvest would take place in our model (30 cm dbh), for 22 tree species. Each dot represents a species, and lines represent the relationship between IRR and the time needed to reach the harvesting threshold, when the survival rate for each tree species 15 years after planting is varied around the mean ± one standard deviation. Both predictor variables are log transformed. Colored bands represent the 80% confidence interval

Full size image

Discussion

The aim of this study was to determine whether plant functional traits could predict financial values of native trees in a forest restoration setting by enrichment planting in tropical rainforests and to identify tree species as economically viable options for use in such settings. Our results show that for a subset of traits, a plant functional trait approach could be used to predict long-term financial values for tree species native to northern Borneo, but that the predictions where opposite to our expectations. We hypothesized that acquisitive traits indicative of high growth rates would be related to high financial value. However, we instead found an opposite pattern; high financial value was correlated to conservative trait values such as low SLA and low N content.

We found that a subset of five traits (Ca, Mg, N, pH, and SLA) was useful for predicting IRR of tree species. Hence, of the 18 traits we measured, thirteen were not significantly related to IRR in the model. In order to support our hypothesis that acquisitive trait values would be related to high IRR, the traits included in the PLS model would all have been positively related to IRR (i.e., high Ca, Mg, N, pH, and SLA would be accompanied by high IRR). This would indicate high nutrient content and low content of defensive compounds, which is typical for a plant species with an acquisitive strategy. However, the tree species exhibited the opposite relationship between their traits and IRR, meaning that species with trait values in ranges normally associated with more conservative growth strategies had higher IRR than species with more acquisitive values. The most likely reason is that the fastest-growing species with acquisitive traits also tended to have lower survival rates, which reduces overall income. This is in line with other studies showing that species with conservative traits are more robust and can endure environmental stress in a better way than species with more acquisitive traits (Wright 2010; Díaz et al. 2016; Harrison and LaForgia 2019).

We also found a potential tradeoff between growth rate (i.e., time needed to reach 30 cm dbh) and survival. Specifically, our results revealed that traits positively correlated with time needed to reach the cutting limit (i.e., traits related to low growth rates) were similar to traits that were positively correlated with survival rate (Fig. 4b, d). In the growth-rate and survival models, LT along with an array of allometric traits were positively correlated to the time needed to reach 30 cm dbh (i.e., low growth rate) and survival rate, respectively. We hypothesize that the contrasting pattern by which traits related to high growth versus survival rates is a plausible explanation for our finding that neither architectural traits nor LT were correlated to IRR. This tradeoff where the two main variables, growth and survival rate, associated with plant traits in opposite ways, might also explain why the traits-IRR model only accounted for a modest part of the observed variance.

Aside from its effects on financial value, species selection could be used to promote other services to increase utility of enrichment plantings. Species’ traits can affect ecosystem functionally through a multitude of pathways, and are not limited to one trophic level (de Bello et al. 2010). For example, the Diospyros, Canarium, and Shorea genera all include species with edible fruits. Identifying financially viable fruit tree species could support biodiversity and provide additional food security to local communities. Additionally, leaf pH, and by extension cation content, which influence pH levels, has been shown to slow down decomposition, affecting nutrient cycling and soil carbon sequestration (Cornelissen et al. 2006). Furthermore, traits that were not related to IRR in our study could promote other services independently of financial value. In the case of decomposition, tropical forests generally have very poor litter quality and decomposition rates that are slower than expected given the climate conditions that are excellent for fast decomposition (Makkonen et al. 2012). This poor litter quality is driven partly by traits we found related to IRR, but also by others, such as lignin content.

Our study design, which focused on financial value of different tree species, grown within a secondary forest setting, could be expanded upon by considering other management goals, such as forest ecosystems’ ability to supply a multitude of services and respond to disturbances of various kinds, i.e., functional resilience (Messier et al. 2019). Recovery of functional resilience is often one of the main goals of forest restoration efforts (Lamb et al. 2005). For example, the original purpose of the INIKEA project was to promote recovery of tree biodiversity after heavy disturbance in the forest reserve. Further, a different restoration project in Sabah showed that enrichment planting can accelerate recovery of carbon stocks after human disturbance (Philipson et al. 2020). Selecting suites of multiple species, based on plant functional traits, that increase the likelihood of survival and form a functional network that cover a wide array of management goals, may improve forest resilience and versatility in the enrichment planting approach (Aquilué et al. 2021).

Survival rates among the tree species ranged from 95% to, in some cases, as low as 15% and this variation was highly influential in predicting IRR of the species. There are several possible reasons for why we saw such a large variation among species. For example, low seedling survival may simply be common for some species, particularly on the acquisitive side of the trait spectrum, or because the site conditions poorly matched the species’ niche. Regardless, low survival rates reduced IRR, to the point where the two fastest growing species, Shorea leprosula and Sh. parvifolia, had some of the lowest IRR values in two out of three scenarios of our sensitivity analysis, due to their very low survival rate of 15% (Figs. 2a, 3; Table 2). If survival rates for these species could be raised to the mean of our 22 species, then their IRR in the mid scenario of our sensitivity analysis would be around 6% (Fig. 3), surpassing the 4.89% of Shorea macrophylla, currently the highest valued species in the scenario. Therefore, management efforts to improve survival, selection of species with higher survival rates, or increased planting to compensate for low survival could be used in conjunction with selection for fast early growth to maximize IRR. The last option might be necessary, given that there likely are tradeoffs between fast growth and survival for some species, as seen for Shorea leprosula and Sh. parvifolia. However, species like Shorea macrophylla show that this growth-survival trade-off can be tapered, achieving both a relatively high survival rate (70%) and growth rate (40 years to achieve cutting limit dbh in mid-scenario). Yields could probably also be improved by a range of management interventions outside the scope of this study. Potentially useful interventions could for example include optimizing the planting regime and harvesting dimensions, applying additional canopy, and/or climber treatments to improve light conditions (Gustafsson et al. 2016), limited selective logging of trees not included in the enrichment plantings, and acquiring deeper knowledge of species-specific site adaptations and genetic variability (Axelsson et al. 2020).

Shorea leprosula and Sh. parvifolia also stood out as the two species that varied most in our sensitivity analysis (Fig. 2). While a few species swapped rankings between different scenarios, e.g., Sh. macrophylla and Sh. ovalis, for most species, variation between scenarios were relatively minor. In comparison, for Sh. leprosula and Sh. parvifolia, the relative change between different scenarios was much larger. For example Sh. parvifolia was ranked 19/22 in the mid-IRR-scenario (Fig. 2a), 9/22 in the highest value scenario (Fig. 2b), and 22/22 in the low-value scenario (Fig. 2c). This fluctuation is likely explained by the fact that the two parameters that vary in the sensitivity analysis, management cost and growth rate, interact with survival and growth characteristics (i.e., very high growth rate and very low survival rate; Table 2) of the two species differently across scenarios compared the other 20 species.

In our methodology, we used growth rate estimations based on the first 15 years of growth in our plots, and assumed linear growth throughout the tree’s lifespan. However, the growth rate of tropical trees normally accelerates as trees grow larger in diameter and height, and gain access to more sunlight (Appanah and Turnbull 1998; King et al. 2006). Thus, our growth rates should be viewed as conservative estimates. For example, in plantation trials by the Forest Research Institute of Malaysia (FRIM), Shorea macrophylla reached 48 cm dbh in 23 years, which is a much higher growth rate than in our study, where the species reached 30 cm dbh in 29 years in our most optimistic scenario (Appanah and Weinland 1993). The exception would be Shorea leprosula and Sh. parvifolia, which exhibited very rapid growth (Table 2), matching values in literature. For example, the same FRIM trials included Sh. leprosula that achieved 33.6 cm dbh in 30 years, and Sh. parvifolia that achieved 30.9 cm dbh in 23 years, which is similar to our results (Appanah and Weinland 1993). Generally, our species outpaced growth rates from trees under canopy shade in primary forests, where the mean annual increment can be as low as 0.1 mm/year (Manokaran and Kochummen 1994), at which rate it would take 300 years to reach 30 cm dbh for the average tree. However, in logged forests, mean growth rates are generally higher than most of our estimates. A traditionally used assumption for growth rate in logged forests is 0.8 cm/year, i.e., 30 cm would be reached in 37.5 years (Appanah and Turnbull 1998). While this assumed growth rate is likely higher than realized growth, it would need to be cut approximately in half in order to match the average species growth rate in our mid-scenario, and that is with our study plot receiving liberation and thinning treatments designed to improve growth, that a logged forest does not.

Our findings also allowed us to evaluate viability of native tree species in forest landscape restoration or reforestation efforts as an alternative to exotic tree plantations and selective logging. For example, the IRR for Shorea macrophylla was 4.89% in the average scenario and ranged from 2.82% in the low-value scenario to 7.8% in the high-value scenario. Even in the average-value scenario, its IRR is comparable to that for hardwood plantations in Australia, where the average IRR is 6% (Venn 2005). However, the IRR in the high-value scenario is lower than the 13–14% estimated for rubber plantations (Winarni et al. 2018) as well as the 9–14% estimated for Acacia and Eucalyptus (Mackensen and Fölster 2000). It is also considerably lower than another estimate for plantation forestry using exotic tree species in Vietnam, for which IRRs of 26–34% were obtained (Cuong et al. 2020). Finally, IRR in the high-value scenario was lower than low-end estimates of profits from oil palm plantations, which are around 15% (Svatoňová et al. 2015). However, it should be noted that other sources suggest that oil palm plantations have higher IRR values of 21–27% (Latif et al. 2003), 68% (Noormahayu et al. 2009), or 58–74% (Wildayana 2016). The financial potential of restoration plantings thus seems to be lower than that of most alternative and comparable land-use types. Nevertheless, it is still profitable and could be viable in areas where other land uses are prohibited or are marginal for practical reasons. Alternatively, it could find viability by filling market niches that other land-use types cannot.

One challenge in the global wood value chain is meeting the demand for high-value tropical hardwood used in furniture and décor (Sarshar 2012). These demands have traditionally been met by harvesting natural forests that have now largely been depleted or are being protected (Bieng et al. 2021). While hardwood plantations with Acacia and Eucalyptus are common, these species are mainly used for pulpwood and are poor options for the production of high-value hardwoods (Sarshar 2012). If enrichment planting with native hardwood species in successional lowland dipterocarp forests can be made financially viable, it could be an attractive alternative to both conventional logging and forest plantations, neither of which have historically been able to provide a sustainable supply of hardwoods. Combining selective logging with enrichment planting has been proposed to address this issue at various times in recent decades (Schulze et al. 1994; Rimbawanto 2006) and is actively being implemented in Indonesia (Rimbawanto 2006; Izuno et al. 2013; Ruslandi et al. 2017). There, the four main species used in enrichment plantings are Shorea leprosula, Sh. johorensis, Sh. parvifolia and Sh. macrophylla (Ruslandi et al. 2017). Our study included all of these species except Sh. johorensis and our results suggest that Sh. macrophylla has the highest financial value in this context.

The trait approach applied in this study could potentially be used to select for specific purposes in the wood industry, although this would require future studies to be verified. For example, species with large investment in defense compounds such as phenolics could be expected to be more resistant to fungi and insect attacks (Kulbat 2016). Similarly, high wood density is sometimes used as a proxy for mechanical stability, and is further correlated to other important attributes like elasticity to bending and resistance to breakage (Chave et al. 2009). Furthermore, trait composition is also likely to influence species’ ability to adapt to climate change. While long-term forest resilience to increased temperatures may be higher than previously thought, higher temperatures are projected to reduce forest productivity, especially in ecosystems with high mean annual temperatures (Sullivan et al. 2020). Species with a conservative growth strategy are generally more resilient to environmental stress and might thus to some extent be better adapted to rapidly changing abiotic conditions (Wright 2010; Díaz et al. 2016; Harrison and LaForgia 2019). For example, the abundance of moisture-demanding species on Barro Colorado Island have been shown to decline due to climate change, even in a relatively short time frame (Condit et al. 1996). Additionally, within-species genetic variation might also be a source of considerable trait variation, which can be used to improve seed source selection by identifying beneficial local adaptations (Axelsson et al. 2023). As global wood demand is only expected to increase, in order to secure the supply of wood in the tropics, selecting species for climate resilience, along with other properties, will be an important part of all enrichment planting programs moving forward (WWF 2012a).

Conclusions

Our results showed that a subset of functional traits could be used to predict financial values of native tree species. Financial value was negatively correlated with Ca, Mg, and N content, along with pH and specific leaf area. Thus, financial value was highest when values for these traits corresponded to those associated with the more conservative side of the plant economics spectrum. This is likely due to tradeoffs between survival and growth rates, which vary heavily among the studied species. On the financial side, enrichment plantings generally seem to be profitable, although not to the same degree as the most common competing land-use types. Overall, we conclude that enrichment planting using selected native tree species could be a financially sound and viable alternative to traditional land management strategies that possibly improves biodiversity and ecosystem services on degraded lands. We further conclude that functional traits together with growth and survival rates have potential to be used to screen unexplored native species for financial viability in enrichment planting.