Introduction
The global carbon cycle that underpins all life on earth is dependent upon the forested ecosystems of the tropics (Lewis et al. Reference Lewis, Edwards and Galbraith2015). Carbon maintained in woody biomass restricts the amount of atmospheric carbon, limiting the rise in global temperatures (Guillaume et al. Reference Guillaume, Kotowska, Hertel, Knohl, Krashevska, Murtilaksono, Scheu and Kuzyakov2018; Pan et al. Reference Pan, Birdsey, Fang, Houghton, Kauppi, Kurz, Phillips, Shvidenko, Lewis, Canadell, Ciais, Jackson, Pacala, McGuire, Piao, Rautiainen, Sitch and Hayes2011). Carbon stocks are dependent upon community diversity via multiple indirect effects (Grace et al. Reference Grace, Anderson, Olff and Scheiner2010; Liang et al. Reference Liang, Crowther, Picard, Wiser, Zhou, Alberti, Schulze, McGuire, Bozzato, Pretzsch, de-Miguel, Paquette, Hérault, Scherer-Lorenzen, Barrett, Glick, Hengeveld, Nabuurs, Pfautsch, Viana, Vibrans, Ammer, Schall, Verbyla, Tchebakova, Fischer, Watson, Chen, Lei, Schelhaas, Lu, Gianelle, Parfenova, Salas, Lee, Lee, Kim, Bruelheide, Coomes, Piotto, Sunderland, Schmid, Gourlet-Fleury, Sonké, Tavani, Zhu, Brandl, Vayreda, Kitahara, Searle, Neldner, Ngugi, Baraloto, Frizzera, Bałazy, Oleksyn, Zawiła-Niedźwiecki, Bouriaud, Bussotti, Finér, Jaroszewicz, Jucker, Valladares, Jagodzinski, Peri, Gonmadje, Marthy, O’Brien, Martin, Marshall, Rovero, Bitariho, Niklaus, Alvarez-Loayza, Chamuya, Valencia, Mortier, Wortel, Engone-Obiang, Ferreira, Odeke, Vasquez, Lewis and Reich2016; Tilman et al. Reference Tilman, Isbell and Cowles2014). These include impacts upon how efficiently resources are taken up and converted to biomass (Hodapp et al. Reference Hodapp, Hillebrand and Striebel2019). Species-specific pathogen and herbivore impacts are also limited by greater diversity (Parker et al. Reference Parker, Saunders, Bontrager, Weitz, Hendricks, Magarey, Suiter and Gilbert2015). Whilst there has been evidence of diversity-carbon storage links across wet tropical forests, it is less clear whether this is reflected in the seasonally dry tropics (Dexter et al. Reference Dexter, Pennington, Oliveira-Filho, Bueno, Silva de Miranda and Neves2018; Godlee et al. Reference Godlee, Ryan, Bauman, Bowers, Carreiras, Chisingui, Cromsigt, Druce, Finckh, Gonçalves, Holdo, Makungwa, McNicol, Mitchard, Muchawona, Revermann, Ribeiro, Siampale, Syampungani, Tchamba, Tripathi, Wallenfang, Beest, Williams and Dexter2021; Miles et al. Reference Miles, Newton, DeFries, Ravilious, May, Blyth, Kapos and Gordon2006; Murphy and Lugo Reference Murphy and Lugo1986). Drylands are a critical and growing component of the global carbon stock (Ahlström et al. Reference Ahlström, Raupach, Schurgers, Smith, Arneth, Jung, Reichstein, Canadell, Friedlingstein, Jain, Kato, Poulter, Sitch, Stocker, Viovy, Wang, Wiltshire, Zaehle and Zeng2015). Subsequently, they are key to carbon cycling (Fawcett et al. Reference Fawcett, Cunliffe, Sitch, O’Sullivan, Anderson, Brazier, Hill, Anthoni, Arneth, Arora, Briggs, Goll, Jain, Li, Lombardozzi, Nabel, Poulter, Séférian, Tian, Viovy, Wigneron, Wiltshire and Zaehle2022). Despite this, dryland climates often make for productive agricultural lands, leading to their conversion (Lewin et al. Reference Lewin, Murali, Rachmilevitch and Roll2024). Drylands are also expected to be at risk from global temperature rise; this is coupled with lower formal conservation designation than non-drylands (Huang et al. Reference Huang, Yu, Dai, Wei and Kang2017). It is therefore crucial to understand the importance of dryland diversity for ecosystem function, so biodiversity conservation can be prioritised (Hanan et al. Reference Hanan, Milne, Aynekulu, Yu and Anchang2021).
Diversity controls carbon stocks in grass-dominated dry ecosystems in Africa (Godlee et al. Reference Godlee, Ryan, Bauman, Bowers, Carreiras, Chisingui, Cromsigt, Druce, Finckh, Gonçalves, Holdo, Makungwa, McNicol, Mitchard, Muchawona, Revermann, Ribeiro, Siampale, Syampungani, Tchamba, Tripathi, Wallenfang, Beest, Williams and Dexter2021; Pelletier et al. Reference Pelletier, Siampale, Legendre, Jantz, Laporte and Goetz2017). Caatinga dry forest biomass likewise shows diversity-dependent biomass (Gomes Souza et al. Reference Gomes Souza, Sfair, Paula, Barros, Rito and Tabarelli2019). However, more widely across neotropical dry forests, diversity shows little sign of a relationship with carbon stocks (Mora et al. Reference Mora, Jaramillo, Bhaskar, Gavito, Siddique, Byrnes and Balvanera2018). Studies on the tropical islands of Asia are conspicuous in their absence (Becknell et al. Reference Becknell, Kissing Kucek and Powers2012). Asian dry forests are found from India to the Lesser Sunda archipelago of Indonesia (Ratnam et al. Reference Ratnam, Tomlinson, Rasquinha and Sankaran2016). Ecophysiological constraints of drought result in plant community species composition quite distinct from the rest of wet tropical Asia (Kalkman Reference Kalkman1955; Baltzer et al. Reference Baltzer, Davies, Bunyavejchewin and Noor2008; Rustiami et al. Reference Rustiami, Tihurua and Trethowan2020; van Steenis Reference van Steenis1979). The Lesser Sunda archipelago is a chain of small islands, the largest of which is Timor, being only 28,418 km2. The subsequent minimal interior of these small islands leads to a greater proportion of the natural vegetation being exposed to the predominantly coastal human communities. The coastal human communities exploit the vegetation as a source of timber, food products, and indirectly water via the protection of watersheds (particularly important during the months of low precipitation) (Beller et al. Reference Beller, d’Ayala and Hein1990; Pattanayak and Wendland Reference Pattanayak and Wendland2007). It is unclear how land use impacts carbon storage in the Indonesian seasonal forest ecosystem and how sensitive stocks are to changing diversity (Jew et al. Reference Jew, Dougill, Sallu, O’Connell and Benton2016; McNicol et al. Reference McNicol, Ryan and Mitchard2018; Ruiz-Jaen and Potvin Reference Ruiz-Jaen and Potvin2011; Silva Pedro et al. Reference Silva Pedro, Rammer and Seidl2015).
A consensus has yet to be reached on the form of the relationship between diversity and carbon stocks. Linear positive relationships are often assumed where greater diversity results from efficient niche partitioning and resource use that benefits biomass production (Poorter et al. Reference Poorter, van der Sande, Thompson, Arets, Alarcón, Álvarez-Sánchez, Ascarrunz, Balvanera, Barajas-Guzmán, Boit, Bongers, Carvalho, Casanoves, Cornejo-Tenorio, Costa, Castilho, Duivenvoorden, Dutrieux, Enquist, Fernández-Méndez, Finegan, Gormley, Healey, Hoosbeek, Ibarra-Manríquez, Junqueira, Levis, Licona, Lisboa, Magnusson, Martínez-Ramos, Martínez-Yrizar, Martorano, Maskell, Mazzei, Meave, Mora, Muñoz, Nytch, Pansonato, Parr, Paz, Pérez-García, Rentería, Rodríguez-Velazquez, Rozendaal, Ruschel, Sakschewski, Salgado-Negret, Schietti, Simões, Sinclair, Souza, Souza, Stropp, ter Steege, Swenson, Thonicke, Toledo, Uriarte, van der Hout, Walker, Zamora and Peña-Claros2015; Ruiz-Benito et al. Reference Ruiz-Benito, Gómez-Aparicio, Paquette, Messier, Kattge and Zavala2014). Likewise, greater diversity can increase the chance that high-carbon individuals will be present (Carol Adair et al. Reference Carol Adair, Hooper, Paquette and Hungate2018; Cavanaugh et al. Reference Cavanaugh, Gosnell, Davis, Ahumada, Boundja, Clark, Mugerwa, Jansen, O’Brien, Rovero, Sheil, Vasquez and Andelman2014; Loreau and Hector Reference Loreau and Hector2001). Alternatively, at higher diversity, carbon storage may plateau or even decrease, caused by more competitive, low-carbon individuals (Aponte et al. Reference Aponte, Kasel, Nitschke, Tanase, Vickers, Parker, Fedrigo, Kohout, Ruiz-Benito, Zavala and Bennett2020; Conti and Díaz Reference Conti and Díaz2013). Therefore, quadratic effects of diversity upon carbon storage may occur where an initial increase is followed by saturation and/or decline. Despite the often observed parabolic relationship, linearity between biodiversity and carbon is often predicted. Here, we test whether that is the case on the Lesser Sunda island of Flores.
Communities contain a combination of species, their evolutionary history, and their traits (e.g. Chao et al. Reference Chao, Chiu and Jost2014a). These features combine as potential multidimensional controls upon carbon storage (Tilman et al. Reference Tilman, Isbell and Cowles2014). Calculated via species identification, measurements of evolutionary tree topology, and trait (e.g. leaf area, height, etc.) ordination space, they offer complementary information that impacts niche packing, resource use and abundance (Conti and Díaz Reference Conti and Díaz2013; Mensah et al. Reference Mensah, Veldtman, Assogbadjo, Glèlè Kakaï and Seifert2016; Ouyang et al. Reference Ouyang, Xiang, Wang, Zeng, Lei, Deng and Peng2016). How the fluctuation of these three measures of diversity mediates carbon storage across seasonally dry land use types requires study to determine the importance of species for conservation plans primarily focused upon ecosystem function.
Whilst niche packing is invoked as a driver of productivity and biomass stocks via ties to efficiency of resource use in the local environment (Tilman et al. Reference Tilman, Isbell and Cowles2014). Species niche is a function of interactions across guilds, including non-woody species that store minimal carbon. Here, we use all plant species to calculate diversity metrics. This ensures we account for more competing and facilitating species than using tree species alone. These metrics include trait diversity, where we gather reproductive and dispersal-linked trait data, alongside more commonly studied traits such as plant and leaf size, to better encompass niche axes and provide a metric better linked to the niche space occupied by the community (E-Vojtkó et al. Reference E-Vojtkó, Junker, Bello and Götzenberger2022; McFadden et al. Reference McFadden, Fritz, Zimmermann, Pellissier, Kissling, Tobias, Schleuning and Graham2022).
Land use impacts diversity by altering the conditions for species reliant upon minimal human impact (Alroy Reference Alroy2017). With greater human resource use, e.g. logging, species are removed, and species that favour disturbance begin to dominate (Carreño-Rocabado et al. Reference Carreño-Rocabado, Peña-Claros, Bongers, Alarcón, Licona and Poorter2012; Sheil and Burslem Reference Sheil and Burslem2003). As human resource use in a landscape increase toward agriculture more planted species become present (Barlow et al. Reference Barlow, Gardner, Araujo, Ávila-Pires, Bonaldo, Costa, Esposito, Ferreira, Hawes, Hernandez, Hoogmoed, Leite, Lo-Man-Hung, Malcolm, Martins, Mestre, Miranda-Santos, Nunes-Gutjahr, Overal, Parry, Peters, Ribeiro-Junior, Silva, Silva Motta and Peres2007). These changes have knock-on effects upon diversity but also aboveground biomass (Avila et al. Reference de Avila, van der Sande, Dormann, Peña-Claros, Poorter, Mazzei, Ruschel, Silva, Carvalho and Bauhus2018). How variation in diversity and land use changes across Indonesian dry ecosystems is little known, as are their joint effects upon aboveground biomass.
Whilst studies of Indonesia’s seasonally dry plant diversity are growing (Laumonier and Nasi Reference Laumonier and Nasi2018; Simbolon Reference Simbolon1999; Sumadijaya et al. Reference Sumadijaya, Noviady, Kusuma, Sambas, Jakalalana, Rosalina, Mahendra, Windadri and Gardner2024; Sutomo et al. Reference Sutomo, Darma and Iryadi2020, Reference Sutomo, Etten and Iryadi2021; Trethowan Reference Trethowan2021; Wiriadinata et al. Reference Wiriadinata, Girmansyah, Hunter, Hoover and Kartawinata2013), assessments of its relationship with carbon storage are few. Here, we use data from a landscape with a gradient from high to low forest management in western Flores to test two hypotheses (1) aboveground carbon stocks are determined by the interaction of land use and species, phylogenetic and trait diversity and (2) that a linear relationship best fits the interaction effects of diversity.
Methods
Site description, plot establishment and aboveground biomass
One hundred and thirty-three 20 m × 50 m plots were established in the seasonally dry landscape of Mbeliling, West Manggarai district (119.78 E to 120.30 E, −8.53 S to −8.87 S) on the island of Flores (Figure 1(a), Table S1). West Manggarai mean air temperatures range from 26.6 to 28.2 ºC. Relative humidity varies between 71 and 88 %, and annual precipitation ranges from 1700 mm to 2400 mm. We sampled locations with increasing intensities of land use in three broad categories of primary, secondary, or agricultural sites. We removed 13 mangrove and savanna plots. The 35 primary sites were relatively untouched by humans. The 51 secondary sites were forests that had been logged or damaged by trails (38) or forests that had been completely destroyed and begun to regrow – often appearing shrubby (13). The 34 agricultural sites were teak plantations (7) or farmlands planted with non-rice crops and timber species (27; Figure 1(a)).
(a) Map of vegetation sampling sites in west Flores, Indonesia. (b) Mean and standard error aboveground biomass (AGB) across land use types. (c) Diversity and diversity land use interaction effect size range across models of AGB. Linear and quadratic diversity effects presented. (d) Differences in model error, measured by the Akaike Information Criterion (AIC), between models parameterised by quadratic or linear effects of diversity. (e) Predicted AGB from the four models with consistent land use diversity interactions and lowest error rates (AIC). Colours represent plots of land use type included as the parameter in the model, grey points represent plots that are not within that type as specified in that model. For example, the left-hand panel shows secondary forests in red and grey represent primary and agricultural plots.
All trees >35 cm diameter at breast height (1.3 m, dbh) were recorded from each 20 m × 50 m plot. Nested within each of these plots, all trees >20 cm dbh were recorded in a 20 m × 20 m subplot. All trees >10 cm dbh were recorded in a 10 m × 10 m subplot. All individuals >2 cm dbh and height >1.5 cm were recorded in a 5 m × 5 m subplot, and all individuals irrespective of diameter were recorded in a 2 m × 2 m subplot. This method (Hairiah et al. Reference Hairiah, Sitompul, van Noordwijk and Palm2001) allowed us to sample across all guilds that contribute to seasonal plant communities (Devecchi et al. Reference Devecchi, Lovo, Moro, Andrino, Barbosa-Silva, Viana, Giulietti, Antar, Watanabe and Zappi2020). Most individuals (93%) were assigned to a species or named morphospecies. The few unidentified individuals were removed from all analyses, resulting in 126 named species and morphospecies. We calculated aboveground biomass
$AGB$
, our measure of carbon storage, for all trees >10 cm dbh using hypsometer-measured height
$H$
, diameter at breast height
$D$
and wood density
$WD$
acquired from previously published datasets (Chave et al. Reference Chave, Réjou-Méchain, Búrquez, Chidumayo, Colgan, Delitti, Duque, Eid, Fearnside and Goodman2014) using the biomass package (Réjou-Méchain et al. Reference Réjou-Méchain, Tanguy, Piponiot, Chave and Hérault2017). For this, we discarded bamboo, palm, and liana species as their carbon stocks are poorly predicted by allometric equations (Chave et al. Reference Chave, Réjou-Méchain, Búrquez, Chidumayo, Colgan, Delitti, Duque, Eid, Fearnside and Goodman2014). Aboveground biomass (AGB) calculation used the following equation from Chave et al. (Reference Chave, Réjou-Méchain, Búrquez, Chidumayo, Colgan, Delitti, Duque, Eid, Fearnside and Goodman2014):
$$AGB = 0.0673 + {\left( {WD\ {\rm{*}}\ H\ {\rm{*}}\ {D^2}} \right)^{0.976}}$$
To calculate AGB for the entire 20 m × 50 m plots, we multiplied the trees sampled in subplots: those >10 cm dbh and <20 cm dbh in 10 m × 10 m subplots by 10, and those >20 cm dbh and <35 cm dbh in 20 m × 20 m subplots by 2.5.
Environmental covariate
To ensure our results were not a function of environmental variation across the landscape, we accounted for it across models (Corona-Núñez et al. Reference Corona-Núñez, Campo and Williams2018). For each plot, we downloaded fifteen ‘bioclim’ variables from Worldclim (Fick and Hijmans Reference Fick and Hijmans2017). Eight climatic variables were log-transformed (mean annual temperature, maximum temperature of the warmest month, minimum temperature of the coldest month, mean temperature of the wettest quarter, mean temperature of the driest quarter, mean temperature of the warmest quarter, mean temperature of the coldest quarter, precipitation seasonality) and seven were not transformed (annual precipitation, precipitation of the wettest month, precipitation of the driest month, precipitation of the wettest quarter, precipitation of the driest quarter, precipitation of the warmest quarter, precipitation of the coldest quarter). Non-log-transformed elevation data from the Shuttle Radar Topography Mission was used (Farr et al. Reference Farr, Rosen, Caro, Crippen, Duren, Hensley, Kobrick, Paller, Rodriguez, Roth, Seal, Shaffer, Shimada, Umland, Werner, Oskin, Burbank and Alsdorf2007). Bulk density of fine earth fraction, volumetric fraction of coarse fragments, pH, sand in fine earth and soil organic carbon from SoilGrids were included (Hengl et al. Reference Hengl, Jesus, Heuvelink, Gonzalez, Kilibarda, Blagotić, Shangguan, Wright, Geng, Bauer-Marschallinger, Guevara, Vargas, MacMillan, Batjes, Leenaars, Ribeiro, Wheeler, Mantel and Kempen2017). There was a single principal component axis of environmental variation between sites that explained 93 % of the variation in the scaled (z-scores) environmental data. This was retained as an environmental covariate across all models.
Trait data
We gathered traits for 126 species from taxonomic publications, compiling data on plant, leaf and fruit size, plant habit and flower size, and colour. We had five dichotomous plant habit variables: tree, shrub, climbing, herb and epiphyte. Leaf shapes were designated as either obovate, oblanceolate, palmately lobed, ovate, elliptic, lanceolate, linear, orbicular or oblong (Schrader et al. Reference Schrader, Shi, Royer, Peppe, Gallagher, Li, Wang and Wright2021). Other leaf shape designations were placed within this framework, e.g. sublanceolate became lanceolate, and subovate became obovate. We used the largest recorded measurements for plant size (herb, shrub, tree height, or liana length) and for the two leaf size categories, length and width (Trethowan et al. Reference Trethowan, Arvidsson and Bramley2022). Leaf area was calculated as leaf length multiplied by leaf width multiplied by a leaf shape correction factor (Schrader et al. Reference Schrader, Shi, Royer, Peppe, Gallagher, Li, Wang and Wright2021). For fruit size, we used the largest recorded measurement of either length or width (Trethowan et al. Reference Trethowan, Jennings, Bramley, Clark, Dawson, Moore, Pearce, Puglisi, Riwu-Kaho, Trias-Blasi, Utteridge and Rustiami2023). Flower size was the largest recorded measurement of either calyx, corolla or stamen length (Trethowan et al. Reference Trethowan, Jennings, Bramley, Briggs, Clark, Dawson, Fontaine, Heatubun, Lewis, Lucas, Moore, Pearce, Puglisi, Utteridge, Wanma and Rustiami2024b). Flower colour was split into seven dichotomous variables: red, yellow, white, purple, pink, orange and brown (Trethowan et al. Reference Trethowan, Jennings, Bramley, Briggs, Clark, Dawson, Fontaine, Heatubun, Lewis, Lucas, Moore, Pearce, Puglisi, Utteridge, Wanma and Rustiami2024b). This data was not collected in the field, and intraspecific variation could lead to trait differences on the ground in Flores. All continuous data were log-transformed and scaled as z-scores prior to trait imputation, co-occurrence modelling, and trait diversity calculation.
To fill continuous and categorical trait data gaps, we ran random forests incorporating the first ten selected phylogenetic eigenvectors to aid imputation accuracy (Carmona et al. Reference Carmona, Pavanetto and Puglielli2024; Penone et al. Reference Penone, Davidson, Shoemaker, Di Marco, Rondinini, Brooks, Young, Graham and Costa2014; Stekhoven and Bühlmann Reference Stekhoven and Bühlmann2012). Missing data were far below the 70 % threshold where diversity indices are expected to be affected post gap-filling (Stewart et al. Reference Stewart, Carmona, Clements, Venditti, Tobias and González-Suárez2023). Imputation models were iterated 15 times, and species mean values were calculated for each continuous trait, and the most frequently predicted value for categorical traits was selected.
The phylogeny used for imputation was built for all species in the dataset and assumed a Brownian motion model of evolution (Li et al. Reference Li, Dinnage, Nell, Helmus and Ives2020a). To construct the phylogeny, dated phylogenetic data were acquired from published phylogenies (Jin and Qian Reference Jin and Qian2019). When only genera were present in the backbone phylogeny, species were added as a polytomy at the root node (i.e., most recent common ancestor) of the genus. Genera not present in the backbone phylogeny were bound to the halfway point between the family root node and the basal node. When the family branch length was longer than two-thirds from the family root node to the tip, the genus was bound to the upper third of the whole family branch length (Jin and Qian Reference Jin and Qian2019; Qian and Jin Reference Qian and Jin2016). This followed methods implemented previously using different software, i.e., Phylomatic and BLADJ (Webb et al. Reference Webb, Ackerly and Kembel2008).
Multidimensional diversity calculation
Species, phylogenetic and trait diversity metrics followed the Hill number framework (Chao et al. Reference Chao, Chiu and Jost2014a). Hill numbers obey the replication principle in that when two assemblages have zero species in common and identical relative abundance distributions, when summed, they have twice as great diversity. This is not the case for classic abundance weighted diversity measures, e.g. Shannon entropy and Gini-Simpson (Chao et al. Reference Chao, Chiu and Jost2014a; Jost et al. Reference Jost, DeVries, Walla, Greeney, Chao and Ricotta2010).
For all diversity metric calculations, we used species abundances. We extrapolated the Hill species diversity metric to give 98.5 % species coverage for plots without abundance weighting (q = 0) following published recommendations (Chao et al. Reference Chao, Gotelli, Hsieh, Sander, Ma, Colwell and Ellison2014b; Hsieh et al. Reference Hsieh, Ma and Chao2016) and repeated this method for phylogenetic and trait diversity. Phylogenetic extrapolations were of total branch length from the root node, and trait diversity from the summed number of functional entities, which were those entities that exceeded the threshold of the mean distance between any two individuals randomly selected from the pooled assemblage (Chao et al. Reference Chao, Gotelli, Hsieh, Sander, Ma, Colwell and Ellison2014b, Reference Chao, Chiu, Hsieh, Davis, Nipperess and Faith2015). The trait distances were Gower distances with each flower colour and habit down-weighted so that the total weights for flower colours and habits were equal to the continuous trait variables of plant, flower and fruit size and leaf area. Leaf area was retained rather than leaf shape, length, and width, as it is more tightly linked to plant function (Li et al. Reference Li, Reich, Schmid, Shrestha, Feng, Lyu, Maitner, Xu, Li, Zou, Tan, Su, Tang, Guo, Feng, Enquist and Wang2020b). The community phylogeny used for total branch length calculation was identical to that used for trait imputation above.
Statistics
To avoid spatial autocorrelation, we assessed our hypotheses with a series of spatial simultaneous autoregressive lag models of AGB (Bivand et al. Reference Bivand, Hauke and Kossowski2013; Bivand and Piras Reference Bivand and Piras2015). Here, predictions are also dependent upon the value of neighbouring plots, ensuring results account for plots that are closer together, more likely to have similar AGB and limiting potential inflation of effect score estimates (Darmofal Reference Darmofal2015). Neighbours were determined by a spatial covariance matrix resulting from the scaled exponential distances between plots (French Reference French2018). The general statistical model structure was as follows:
$$AGB = \alpha \gamma + \delta $$
The land use parameter
$\gamma $
was a dichotomous variable for either agriculture, secondary or primary forest (hypothesis 1). Models were iterated for each land-use type, for instance, the
$\gamma $
parameter for agriculture models is a categorical variable of either ‘agriculture’ or ‘non-agriculture’. We iterated in this fashion as we were not confident of the order in the magnitude of effect between agriculture, secondary and primary forest upon AGB. Diversity
$\alpha $
was either species, phylogenetic or trait diversity (hypothesis 2). This gave us nine parameter combinations. We also ran models with quadratic diversity effects and calculated the Akaike Information Criterion (AIC) to compare model error between quadratic and linear effects (hypothesis 2). A total of 18 models were built to cover all parameter combinations and quadratic or linear effects of diversity
$\alpha $
. The environmental covariate
$\delta $
was identical across all models.
Results
Land use impacts upon AGB
Models showed that primary forest plots had greater AGB (mean effect upon AGB (effect score) = 0.88, mean standard error (SE) = 0.29) than non-primary plots (Figure 1(b)). Agricultural plots had lower AGB (effect score = −0.34, SE = 0.24, Figure 1(b)), as did secondary forests (effect score = −0.42, SE = 0.35, Figure 1(b)).
Joint land use and diversity effects upon AGB
Diversity measures’ direct effects upon AGB were inconsistent, in terms of both direction and size, in addition to frequently having large standard errors (Supplementary Material Table 1). The interaction between land use and diversity, however, often had consistent effect directions across models, irrespective of the form of the effects of diversity upon AGB (Figure 1(c)). For instance, primary forest interaction with trait diversity had consistently negative effects upon AGB with both linear (effect score = −0.02) and quadratic diversity effects (effect score = −0.58, Figure 1(c)).
All models except for that parameterised by phylogenetic diversity and secondary forest status (and their interaction) had reduced lower prediction error with quadratic diversity effects (Figure 1(d)). The greatest improvement was in the model parameterised by trait diversity and primary forest status, and their interaction (Figure 1(d)).
Amongst the models with consistent effects of the interaction between land use and diversity (Figure 1(c), where effect score bar do not approach zero), all models had reduced model error with quadratic effects (Figure 1(c)). The models in question were two parameterised by phylogenetic diversity and its interaction with primary forest and agriculture, two with species diversity and its interaction with primary and secondary forest parameters, but no models parameterised by trait diversity (displayed in Figure 1(e)). Quadratic diversity effects were overall negative (Table S2) indicative of the concave relationship, e.g. peak followed by decline, with AGB.
Discussion
We have shown that land use impacts above-ground carbon stocks in the dry tropical forests of Indonesia (hypothesis 1) but that whilst this is mediated by diversity (hypothesis 2), it is best predicted by the quadratic effects of diversity, which suggests that whilst accounting for the effects of land use, when diversity increases, carbon storage increases to a point but with ever increasing diversity carbon stocks will plateau or even decline in the seasonally dry forests of Indonesia.
In the west Flores landscape, we show that phylogenetic and species diversity are more influential for carbon stocks than trait diversity. For phylogenetic diversity, there are contrasting effects for agricultural and primary forests. Agricultural landscapes’ carbon stocks benefit from more lineages, but primary landscapes do not. Primary landscapes also have a negative relationship with species diversity. Interestingly, species diversity had a positive interaction with secondary forests. Across guild species, diversity is also slightly greater in secondary forests than in primary. This suggests that more species and lineages do not guarantee a more carbon-dense community in primary landscapes but can for those with greater human influence, where diversity may likewise be greatest (Flynn et al. Reference Flynn, Mirotchnick, Jain, Palmer and Naeem2011; Mensah et al. Reference Mensah, Salako and Seifert2020; Tolmos et al. Reference Tolmos, Guerrero-Ramirez, Ameztegui, Barajas Barbosa, Craven and Kreft2024). We found little interaction between trait diversity and land use in shaping carbon stocks. Our focus upon across guild diversity may be linked to this; where previous evidence of the importance of trait diversity was linked to tree species (Conti and Díaz Reference Conti and Díaz2013), we found that land use effects on carbon were independent of trait diversity for a larger cross-section of the plant community.
Our results show that the impacts of secondary forest effects on carbon stocks interact positively with species diversity, and to a lesser extent, phylogenetic diversity. These secondary forests have been shaped by disturbance and succession, both of which may exacerbate diversity impacts upon carbon storage (Lasky et al. Reference Lasky, Uriarte, Boukili, Erickson, John Kress and Chazdon2014). During succession, fast-growing ruderal species may accumulate carbon and, for these taxa, greater phylogenetic and species diversity may result from their survival despite the risk of or prior to being outcompeted as succession proceeds (Lasky et al. Reference Lasky, Uriarte, Boukili, Erickson, John Kress and Chazdon2014).
We present a study that incorporates fruit and flower traits into trait diversity metrics. These traits are representative of niche axes linked to dispersal and reproduction. These traits can influence the overall energy budget of plants with potential consequences for biomass (Salguero-Gómez Reference Salguero-Gómez2017). Plus, they shape the ability to recruit new individuals locally or the likelihood of replenishment from a wider metapopulation, which may determine survival and thus carbon stock retention of individual species (Fricke et al. Reference Fricke, Cook-Patton, Harvey and Terrer2025). There are however, other traits often used as single parameters of productivity and biomass, such as specific leaf area, tree maximum height and wood density (Finegan et al. Reference Finegan, Peña-Claros, de Oliveira, Ascarrunz, Bret-Harte, Carreño-Rocabado, Casanoves, Díaz, Eguiguren Velepucha, Fernandez, Licona, Lorenzo, Salgado Negret, Vaz and Poorter2015; Ruiz-Jaen and Potvin Reference Ruiz-Jaen and Potvin2011). The quadratic effects of trait diversity that we found may not hold if those traits are examined; however our results encompass trait diversity across guilds, providing a different and complementary insight from studies focused solely upon trees.
We show that diversity and carbon storage do not consistently covary across land use types in west Flores. This adds to the evidence base that conservation plans across landscapes of varying human impact will require separate priorities for both biodiversity and carbon (Ferreira et al. Reference Ferreira, Lennox, Gardner, Thomson, Berenguer, Lees, Mac Nally, Aragão, Ferraz, Louzada, Moura, Oliveira, Pardini, Solar, Vieira and Barlow2018; Thomas et al. Reference Thomas, Anderson, Moilanen, Eigenbrod, Heinemeyer, Quaife, Roy, Gillings, Armsworth and Gaston2013; Zhu et al. Reference Zhu, Hughes, Zhao, Zhou, Ma, Shen, Li, Liu, Xu and Watson2021). It has been suggested that conservation of secondary forests will be key to these strategies, and our results suggest that is the case in Flores. Although if, as expected, secondary forests are more accessible and in regions of greater human population density, then their land value will be such that it may be more cost-effective to conserve primary forests (Malhi et al. Reference Malhi, Riutta, Wearn, Deere, Mitchell, Bernard, Majalap, Nilus, Davies, Ewers and Struebig2022; Phelps et al. Reference Phelps, Webb and Adams2012; Van de Perre et al. Reference Van de Perre, Willig, Presley, Bapeamoni Andemwana, Beeckman, Boeckx, Cooleman, de Haan, De Kesel, Dessein, Grootaert, Huygens, Janssens, Kearsley, Kabeya, Leponce, Van den Broeck, Verbeeck, Würsten, Leirs and Verheyen2018). Remotely sensed carbon stocks can easily be incorporated, but plant diversity layers are still required to inform evidence-based spatial conservation plans. Our results suggest that high multidimensional diversity landscapes may not be carbon priorities, yet these may be diversity source regions and may be more stable when exposed to novel climates (Gallagher et al. Reference Gallagher, Hughes and Leishman2013).
Our findings replicate studies across other seasonally dry landscapes showing how carbon storage shifts with land use and biodiversity (Becknell et al. Reference Becknell, Kissing Kucek and Powers2012; Godlee et al. Reference Godlee, Ryan, Bauman, Bowers, Carreiras, Chisingui, Cromsigt, Druce, Finckh, Gonçalves, Holdo, Makungwa, McNicol, Mitchard, Muchawona, Revermann, Ribeiro, Siampale, Syampungani, Tchamba, Tripathi, Wallenfang, Beest, Williams and Dexter2021). Other ecosystem functions and services, such as pollination and the availability of useful plants, are largely unknown at fine scales across the islands from Sumatra to New Guinea (Cámara-Leret et al. Reference Cámara-Leret, Raes, Roehrdanz, De Fretes, Heatubun, Roeble, Schuiteman, van Welzen and Hannah2019; Pironon et al. Reference Pironon, Ondo, Diazgranados, Allkin, Baquero, Cámara-Leret, Canteiro, Dennehy-Carr, Govaerts, Hargreaves, Hudson, Lemmens, Milliken, Nesbitt, Patmore, Schmelzer, Turner, Andel, Ulian, Antonelli and Willis2024; van der Plas et al. Reference van der Plas, Ratcliffe, Ruiz-Benito, Scherer-Lorenzen, Verheyen, Wirth, Zavala, Ampoorter, Baeten, Barbaro, Bastias, Bauhus, Benavides, Benneter, Bonal, Bouriaud, Bruelheide, Bussotti, Carnol, Castagneyrol, Charbonnier, Cornelissen, Dahlgren, Checko, Coppi, Dawud, Deconchat, De Smedt, De Wandeler, Domisch, Finér, Fotelli, Gessler, Granier, Grossiord, Guyot, Haase, Hättenschwiler, Jactel, Jaroszewicz, Joly, Jucker, Kambach, Kaendler, Kattge, Koricheva, Kunstler, Lehtonen, Liebergesell, Manning, Milligan, Müller, Muys, Nguyen, Nock, Ohse, Paquette, Peñuelas, Pollastrini, Radoglou, Raulund-Rasmussen, Roger, Seidl, Selvi, Stenlid, Valladares, Keer, Vesterdal, Fischer, Gamfeldt and Allan2018). As such, we do not yet have clear baseline knowledge of the interactions between pollinators, plant diversity and carbon, which will be key to understanding climate change impacts upon these megadiverse islands and the knock-on effect for urban and rural livelihoods (Dubois et al. Reference Dubois, Oppo, Galy, Mohtadi, Kaars, Tierney, Rosenthal, Eglinton, Lückge and Linsley2014; Feng et al. Reference Feng, Porporato and Rodriguez-Iturbe2013; Struebig et al. Reference Struebig, Wilting, Gaveau, Meijaard, Smith, Abdullah, Abram, Alfred, Ancrenaz, Augeri, Belant, Bernard, Bezuijen, Boonman, Boonratana, Boorsma, Breitenmoser-Würsten, Brodie, Cheyne, Devens, Duckworth, Duplaix, Eaton, Francis, Fredriksson, Giordano, Gonner, Hall, Harrison, Hearn, Heckmann, Heydon, Hofer, Hon, Husson, Anwarali Khan, Kingston, Kreb, Lammertink, Lane, Lasmana, Liat, Lim, Lindenborn, Loken, Macdonald, Marshall, Maryanto, Mathai, McShea, Mohamed, Nakabayashi, Nakashima, Niedballa, Noerfahmy, Persey, Peter, Pieterse, Pilgrim, Pollard, Purnama, Rafiastanto, Reinfelder, Reusch, Robson, Ross, Rustam, Sadikin, Samejima, Santosa, Sapari, Sasaki, Scharf, Semiadi, Shepherd, Sykes, van Berkel, Wells, Wielstra, Wong, Fischer, Metcalfe and Kramer-Schadt2015; Wolff et al. Reference Wolff, Zeppetello, Parsons, Aggraeni, Battisti, Ebi, Game, Kroeger, Masuda and Spector2021). Our work complements ongoing efforts to assess American, African and continental Asian dry forests (Chaturvedi et al. Reference Chaturvedi, Raghubanshi and Singh2011; Godlee et al. Reference Godlee, Ryan, Bauman, Bowers, Carreiras, Chisingui, Cromsigt, Druce, Finckh, Gonçalves, Holdo, Makungwa, McNicol, Mitchard, Muchawona, Revermann, Ribeiro, Siampale, Syampungani, Tchamba, Tripathi, Wallenfang, Beest, Williams and Dexter2021; Portillo-Quintero et al. Reference Portillo-Quintero, Sanchez-Azofeifa, Calvo-Alvarado, Quesada and Espirito Santo2015; Yang et al. Reference Yang, Yu, Wu, Zhang, Kosugi, Restrepo-Coupe, Huete, Zhang, Liu, Zhang, Liu, Zhao, Zeng, Song, Chen, Song and Tan2023). There are further dry tropical ecosystems in the Lesser Sunda Islands and Papua that still require field-based diversity and carbon data (Laumonier and Nasi Reference Laumonier and Nasi2018; Trethowan et al. Reference Trethowan, Brambach, Cámara-Leret, Laumonier, Sheil, Ferry Slik, Webb, Murdjoko, Bastian, Kartawinata, Mansor, Mansur, Mirmanto, Nurtjahya, Permana, Rozak, Wilkie, Rahmad, Arifiani, Putu Gede, Damayanto, Puglisi, Asmarayani, Benu, Bramley, Dharma, Heatubun, Hidayat, Kuswandi, Mathews, Rustiami, Santika, Santoso, Setiawan, Triono, Wambrauw, Wanma, Hery Yanto, Siregar, Sadili, Dewi Atikah, Kintamani, Tjoa, Culmsee, Girmansyah, Gunawan, Pitopang, Utteridge, Morley, Schrader and Brearley2024a). Field-based carbon stock data collation is likewise warranted from all the tropical islands of Malesia and will help address a major gap in global analysis (Anitha et al. Reference Anitha, Verchot, Joseph, Herold, Manuri and Avitabile2015; Qie et al. Reference Qie, Lewis, Sullivan, Lopez-Gonzalez, Pickavance, Sunderland, Ashton, Hubau, Abu Salim, Aiba, Banin, Berry, Brearley, Burslem, Dančák, Davies, Fredriksson, Hamer, Hédl, Kho, Kitayama, Krisnawati, Lhota, Malhi, Maycock, Metali, Mirmanto, Nagy, Nilus, Ong, Pendry, Poulsen, Primack, Rutishauser, Samsoedin, Saragih, Sist, Slik, Sukri, Svátek, Tan, Tjoa, van Nieuwstadt, Vernimmen, Yassir, Kidd, Fitriadi, Ideris, Serudin, Abdullah Lim, Saparudin and Phillips2017; Slik et al. Reference Slik, Aiba, Brearley, Cannon, Forshed, Kitayama, Nagamasu, Nilus, Payne, Paoli, Poulsen, Raes, Sheil, Sidiyasa, Suzuki and van Valkenburg2010; Sullivan et al. Reference Sullivan, Talbot, Lewis, Phillips, Qie, Begne, Chave, Cuni-Sanchez, Hubau, Lopez-Gonzalez, Miles, Monteagudo-Mendoza, Sonké, Sunderland, ter Steege, White, Affum-Baffoe, Aiba, Almeida, Oliveira, Alvarez-Loayza, Dávila, Andrade, Aragão, Ashton, Aymard, Baker, Balinga, Banin, Baraloto, Bastin, Berry, Bogaert, Bonal, Bongers, Brienen, Camargo, Cerón, Moscoso, Chezeaux, Clark, Pacheco, Comiskey, Valverde, Coronado, Dargie, Davies, De Canniere, Marie Noel Djuikouo, Doucet, Erwin, Espejo, Ewango, Fauset, Feldpausch, Herrera, Gilpin, Gloor, Hall, Harris, Hart, Kartawinata, Kho, Kitayama, Laurance, Laurance, Leal, Lovejoy, Lovett, Lukasu, Makana, Malhi, Maracahipes, Marimon, Junior, Marshall, Morandi, Mukendi, Mukinzi, Nilus, Vargas, Camacho, Pardo, Peña-Claros, Pétronelli, Pickavance, Poulsen, Poulsen, Primack, Priyadi, Quesada, Reitsma, Réjou-Méchain, Restrepo, Rutishauser, Salim, Salomão, Samsoedin, Sheil, Sierra, Silveira, Slik, Steel, Taedoumg, Tan, Terborgh, Thomas, Toledo, Umunay, Gamarra, Vieira, Vos, Wang, Willcock and Zemagho2017).
We offer evidence that land use and diversity together impact dry ecosystem aboveground carbon stocks. However, diversity effects follow a quadratic form, potentially indicative of declining facilitation toward greater competition with negative consequences for carbon. Although models of carbon stocks through time would be required to determine this in more detail. Overall, multidimensional diversity has complex relationships with carbon across land use types and may require distinct conservation strategies across the insular dry forests of Asia.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/S0266467425100400
Data availability statement
The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request. Rmd file to run analyses and build manuscript is available at https://zenodo.org/records/17647569.
Acknowledgements
We thank Bonefasius Haryanto, Stefanus Landing, Martinus N. Sanus, Stefanus K. Dakosmas, Maria E. Mau, Maximus Abun, Abdul Majid, Thomas Nikasius, Fransiskus B. Yosafat, Fransiskus Teri Bugis, Anselmus Guansi and Imelda Intan for field support.
Authors contributions
Wahyu C. Adinugroho, Liam A. Trethowan, Francis Q. Brearley and Rinaldi Imanuddin designed the study, implemented it and contributed to the development of the manuscript. Wahyu C. Adinugroho and Rinaldi Imanuddin led study site selection, field sampling and data collection. Wahyu C. Adinugroho, Rinaldi Imanuddin, Asep Ayat and Lalu Abdi Wirastami participated in field data collection. Renata Borosova, Marie Briggs, Osanna Chu, Sally Dawson, Grace Flanagan, Laura Jennings, Cecilia Lee-Grant, Alison Moore, Shawn O’Donnell, Carmen Puglisi and Mariana H. Santana gathered trait data. Liam A. Trethowan led the writing and analysis of the manuscript, supported by Francis Q. Brearley and all authors read and approved the final paper.
Competing interests
The authors declare none.
Financial support
Fieldwork was funded by Burung Indonesia, Perhimpunan Pelestarian Burung Liar Indonesia, Bird Life International. LAT was supported by NERC/Newton fund grant NE/S007059/1.
