Experiments under controlled conditions have established that ecosystem functioning is generally positively related to levels of biodiversity but it is unclear how widespread these effects are in real-world settings and whether they can be harnessed for ecosystem restoration. We used a long-term, field-scale tropical restoration experiment to test how the diversity of planted trees affected recovery of a 500-ha area of selectively logged forest measured using multiple sources of satellite data. Replanting using species-rich mixtures of tree seedlings with higher phylogenetic and functional diversity accelerated restoration of remote sensing estimates of aboveground biomass, canopy cover and Leaf Area Index. Our results are consistent with a positive relationship between biodiversity and ecosystem functioning in the lowland dipterocarp rainforests of SE Asia and demonstrate that using diverse mixtures of species can enhance their initial recovery after logging.

Study system

The Sabah Biodiversity Experiment (http://www.sabahbiodiversityexperiment.org) occupies 500 ha in the southern part of the Malua Forest Reserve, in Sabah, Malaysian Borneo. The Malua Forest Reserve is an area of approximately 35,000 ha of predominantly selectively logged forest that is publicly owned through Yayasan Sabah (The Sabah Foundation), which holds a 100-year concession under its goal to increase socioeconomic standards in the state. Within the wider Yayasan Sabah logging concession is the Innoprise-FACE Foundation Rainforest Rehabilitation project (INFAPRO), a 25,000-ha area dedicated to promoting the rehabilitation of forests through large-scale enrichment planting within logged areas. To help provide practical recommendations, the Sabah Biodiversity Experiment followed INFAPRO enrichment planting techniques. The region experiences an average temperature of 27°C and an annual rainfall of >3000 mm, distributed between two wet seasons. The Malua Forest Reserve area has been logged twice, once in the 1980s and again in 2007. The 500-ha area of the Sabah Biodiversity Experiment itself was spared the second round of selective logging in 2007 due to the establishment of the experiment in 2002 and has therefore been recovering from the initial round of logging for nearly 40 years. Elevation at this site is under 250 m, with 0–20° range in topography. The pre-logging timber volume of this region has been estimated at 193-221 m³ ha^-1, of which dipterocarps account for the vast majority at 180-216 m³ ha^-1.

Study species

The 16 species used in this experiment are native species belonging to the Dipterocarpaceae: Dipterocarpus conformis Slooten, Dryobalanops lanceolata Burck, Hopea ferruginea Parij, Hopea sangal Korth., Parashorea malaanonan (Blanco) Merr., Parashorea tomentella (Blanco) Merr., Shorea argentifolia Sym., Shorea beccariana Bruck, Shorea faguetiana Heim., Shorea gibbosa Brandis., Shorea johorensis Foxw., Shorea leprosula Miq., Shorea macrophylla Ashton, Shorea macroptera King, Shorea ovalis Korth., and Shorea parvifolia Dyer. Generally, dipterocarps in Sabah are emergent tree species, rarely found 1,200 m above sea level. They have an array of characteristics which likely contribute to their dominance in SE Asian forests including their symbiotic ectomycorrhizal associations and wind-dispersed winged fruits. Reproduction takes place largely through ‘mast fruiting’ events that occur between 2–10 years apart, where many or most of the dipterocarp species simultaneously produce fruit. Dipterocarps have recalcitrant seeds and no soil seed bank. Instead, successful recruits form a seedling bank which often suffer from heavy herbivory. Dipterocarps dominate the lowland forests of SE Asia in terms of biomass but have been heavily selectively logged.

Experimental design

Sabah Biodiversity Experiment features several experimental treatments within its replicated, randomised block design. The experiment consists of 124 four-hectare (200 m x 200 m) plots, divided into two blocks separated by an old logging road (60 plots in the north block and 64 to the south). Each plot (apart from the unplanted controls) is enrichment planted with a mixture of seedlings with a controlled species number (richness) and composition. The design ensures at least one replicate plot for each species richness and composition treatment level in each of the two blocks. Each plot contains 20 parallel planting lines, separated by 10 m areas of remnant vegetation left after the prior selective logging. Within each line, seedlings were planted with 3 m spacing, and planting lines were initially cleared of bamboo, lianas, and shrubs up to a maximum of 1 m on either side of the line of planted seedlings. The experiment was primarily designed to manipulate the diversity and composition of enrichment planted dipterocarps but also investigates the forest management practice of liana removal ('climber cutting'). 114 of the plots make up a gradient in the diversity of enrichment-planted tree species comprising mixtures of 1, 4, or 16 species. The remaining 12 plots were left as naturally regenerating unplanted controls (six in each block). The design uses a set of 16 species that were available in the local seedling nursery in sufficient numbers. These 16 species were grown in single-species enrichment planting 'monocultures' and combined together to form enrichment planting 'polycultures' of 4 or 16 species. The plots enrichment planted with only a single species of dipterocarp allow a comparison of individual species identity effects since each species has a replicate in each of the two blocks (a total of 32 1-species plots).

The intermediate 4-species diversity level is comprised of 16 different species compositions that produce two further treatments that are factorially crossed. These two treatments manipulate genus diversity (two levels) and predicted canopy structural complexity (two levels). The genus diversity treatment compares mixtures of four species comprising two or four dipterocarp genera. The canopy structural complexity treatment also features two levels that either combine species with similar predicted mature heights or with a wider range of these predicted values. In total, this factorial manipulation of genus and canopy structural complexity comprises 32 plots (the 2² factorial combination of the 4 treatments, each with 4 replicate species compositions, each replicated in the two blocks).

Sixteen plots of the most diverse (16-species) mixtures underwent two rounds of liana removal (‘climber cutting’), which were compared with 32 plots enrichment-planted with the same number of species but without this local climber cutting restoration strategy. Due to practical constraints, these cuttings took place in two stages. In July 2011 ten plots were cut in the southern block, and in June 2014 these ten plots, as well as six plots in the northern block, underwent a full round of liana removal. Therefore, at the time of the RapidEye satellite remote sensing in 2012, only the ten plots in the southern block had been subjected to the liana removal treatment. Nevertheless, to avoid the risk of missing effects of this treatment, we included it in the statistical analysis.

In line with standard enrichment planting procedure, after the initial cohort of seedlings was planted (between January 2002 and September 2003), a second cohort was planted to replace initial mortalities (cohort 2 planted September 2008 to August 2009). In combination, the two cohorts planted and surveyed a total of 96,369 dipterocarp seedlings. Further details of the Malua reserve and Sabah Biodiversity Experiment can be found in previous publications.

Remote sensing

Landsat Vegetation Continuous Fields (VCF) tree cover, RapidEye and MODIS imagery were selected to estimate variation in canopy structure based on the needs of data accuracy, the size of the study site and plots and the time period of the experiment.

Landsat Vegetation Continuous Fields tree cover

The Landsat Continuous Fields tree cover (Landsat tree cover) estimates the percentage of horizontal ground per 30 m pixel which is covered with vegetation of at least 5 m vertical height. In this study we refer to Landsat tree cover as Landsat vegetation cover, as in Sabah virtually all vegetation detected by Landsat is higher than this minimum. The product is derived from all 7 bands of Landsat-5 Thematic Mapper (TM) and/or Landsat Enhanced Thematic Mapper Plus (ETM+). The partial resolution of the Landsat vegetation cover dataset is 30 m, which is appropriate for the Sabah Biodiversity Experiment’s plot size of 200 m x 200 m, giving c. 44 pixels per plot. This dataset contains three epochs, 2000, 2005, and 2010, each consisting of a composite of several years’ worth of images in order to minimise the effects of cloud cover. The 2000 epoch consists of data from 1999 to 2002, our 2005 epoch contains years 2003 to 2008, and the 2010 epoch ranges 2008 to 2012.

MODIS MCD15A3H

MODIS MCD15A3H Leaf Area Index (LAI) is widely used in forest monitoring and exhibits very high accuracy. However, the spatial resolution of 500 m means that each plot does not even have a single complete pixel. Instead, a comparison of the entire SBE sites with the surrounding re-logged area is reported elsewhere.

RapidEye imagery

This study used a RapidEye satellite image of the Sabah Biodiversity Experiment site for August 2012. RapidEye imagery uses a higher-spatial resolution of 5 m and a temporal resolution of 5.5 days. This multi-spectral scanner of the RapidEye satellites acquires data in five bands. The blue (0.44–0.51 µm), green (0.52–0.59 µm), red (0.63–0.68 µm), and near-infrared (0.76–0.85 µm) are very similar to that of the Landsat Spectral band equivalents, while also having an additional red-edge band (0.69–0.73 µm). This band allows RapidEye satellite images to provide greater sensitivity to spatiotemporal changes in vegetation.

Vegetation metrics inversion from RapidEye image

A FLAASH atmospheric correction model was applied to the RapidEye image, and vegetation cover, LAI, and aboveground biomass (AGB) were calculated using empirical formulae develop in Pfeifer et al. (Eqs. S1, S2, and S3, respectively), which used RapidEye imagery of the nearby SAFE landscape. Although these equations were not derived for Malua (where SBE is located), the SAFE landscape is close by, and significantly more so than all other options. This provided us with high-resolution (5 m for RapidEye) estimates of LAI, vegetation cover and AGB using a method developed and validated for lowland dipterocarp forests in the same part of Sabah. Further details of the inversion methodology used can be found in the previous publication.

Eqn. S1.

Vegetation cover = 2.66 - 0.66 ∙ Red + 0.3 ∙ RedEdge - 0.08 ∙ NearIR - 0.17 ∙ DissB3 + 1.48 ∙ DissB4 - 0.42 ∙ DissB5 

Eqn. S2.

LAI = 0.9 – 0.59 ∙ Red + 0.41 ∙ RedEdge – 0.11 ∙ NearIR – 0.53 ∙ DissB3 + 1.08 ∙ DissB4 – 0.36 ∙ DissB5

Eqn. S3. 

AGB = 19.45 - exp(MSAV12) - 2.39 ∙ Green + 1.08 ∙ RedEdge + 2.65 ∙ DissB2 - 0.28 ∙ DissB3 + 0.09 ∙ DissB4 - 0.13 ∙ DissB5

Where MSAV12 is the Modified Soil-Adjusted Vegetation Index 2. Green, Red, RedEdge, and NearIR all correspond to the RapidEye bands of the same name, and DissB2, DissB3, DissB4, and DissB5 are the grey-level dissimilarities of green band, red band, near-infrared band, and red-edge band, respectively. Satellite imagery wa pre-processed using ArcGIS and overlaid with the SBE plot layout based on GPS coordinates collected for the perimeter of each block.

Phylogenetic and functional diversity

To investigate the effects of a broader range of aspects of diversity we calculated measures of both phylogenetic and functional diversity. A phylogenetic tree was created using information specified in two papers detailing recent advancements in dipterocarp phylogeny. We calculated Faith’s phylogenetic diversity (PD), defined as the total branch length of the minimum spanning tree from each node to the tree root. We also calculated functional diversity (FD), a functional equivalent of Faith’s PD, by the sum of branch lengths on a functional dendrogram. Out of 46 total trait measurements available, we opted a priori to use leaf nitrogen, leaf phosphorus, leaf thickness, dry weight, wood density, and specific leaf area as these measurements have been used to estimate FD in the literature most widely, and there are current associations with the trade-off between rapid resource acquisition and faster growth and enhanced environmental tolerance and reduced growth. This also avoided using a large number of partially correlated variables. Both PD and FD measurements are dependent on species richness, and so contain information of phylogenetic and functional diversity both among and within species richness levels.

R and RStudio are recommended to access this data, both of which are open source. No other programmes are required to access this data. Any CSV viewer would also give access to this dataset.