Current estimates of carbon (C) storage in peatland systems worldwide
indicate that tropical peatlands comprise about 15 % of the global peat carbon
pool. Such estimates are uncertain due to data gaps regarding organic peat
soil thickness, volume and C content. We combined a set of indirect
geophysical methods (ground-penetrating radar, GPR, and electrical
resistivity imaging, ERI) with direct observations using core sampling and C
analysis to determine how geophysical imaging may enhance traditional coring
methods for estimating peat thickness and C storage in a tropical peatland
system in West Kalimantan, Indonesia. Both GPR and ERI methods demonstrated
their capability to estimate peat thickness in tropical peat soils at a spatial
resolution not feasible with traditional coring methods. GPR is able to
capture peat thickness variability at centimeter-scale vertical resolution,
although peat thickness determination was difficult for peat columns
exceeding 5 m in the areas studied, due to signal attenuation associated
with thick clay-rich transitional horizons at the peat–mineral soil
interface. ERI methods were more successful for imaging deeper peatlands
with thick organomineral layers between peat and underlying mineral soil.
Results obtained using GPR methods indicate less than 3 % variation in
peat thickness (when compared to coring methods) over low peat–mineral soil
interface gradients (i.e., below 0.02
Globally, tropical peatlands are estimated to store 89 PgC, equivalent to
about
Current estimates of C storage in global peatlands range between 528 and 694 Pg
C (Hooijer et al., 2006; Yu et al., 2010). Tropical and subtropical systems
are estimated to comprise about 15 % of the global peat carbon pool, with
Indonesia estimated to contain about 65 % of tropical peat carbon (Page
et al., 2011). However, these estimates are tentative due to uncertainties in
peat thickness, volume and C density at large spatial scales. Estimating peat
carbon storage requires accurate volume measurements calculated from peat
area and thickness. Page et al. (2011) calculated peat volume for Indonesia
using a mean peat depth of 5.5 m, which was based on very few geographically
biased data considering the scale at which the mean depth estimate was
applied: 206 950 km
Near-surface geophysical methods, particularly ground-penetrating radar (GPR), have been used extensively in boreal peatland systems to explore many aspects related to peat development and stratigraphy (Comas and Slater, 2009). Recent studies of peat thickness and peat basin volume using GPR include a variety of field sites and typically indicate discrepancies in peat volume estimates of about 20 % when compared to traditional direct methods such as coring (Rosa et al., 2009; Parsekian et al., 2012; Parry et al., 2014). Electrical resistivity imaging (ERI) has also been used in boreal systems for investigating several aspects of peatland stratigraphy and hydrogeology (Meyer, 1989; Slater and Reeve, 2002; Comas et al., 2004, 2011); however, no studies to our knowledge have focused on peat thickness characterization using ERI.
Although numerous studies have used GPR and ERI methodologies to study peatland attributes in boreal systems, the use of these techniques in tropical systems has not been reported. Although differences in peat types, terrain and/or vegetation cover between boreal and tropical systems must be considered, similarities in peat electromagnetic and electrical properties are anticipated, supporting the use of GPR and ERI methods for mapping tropical peatlands and underlying buried topography.
Here we report the use of a combination of GPR and ERI methods to obtain high-resolution profiles of peat layers in West Kalimantan, Indonesia. The objectives of this study were to (1) test the potential of GPR and ERI for estimating peat thickness in a non-invasive and spatially continuous way at a resolution previously unreported for tropical peatlands and (2) evaluate whether certain information on geological settings and/or peat composition can be drawn from these methods. The ultimate aim of the approach presented here is to demonstrate the applicability of geophysical methods to investigate tropical peat systems, and to highlight potential for improved accuracy of peat C storage estimates relative to estimates derived from traditional coring methods. Advancing this knowledge could help to inform peatland management decisions in Indonesia and improve assessments of peat subsidence and C stock changes.
Two peatland sites located in the West Kalimantan Province of Indonesia were chosen for this study: Tanjung Gunung (Sejahtera village, North Kayong Regency); and Pelang (Pelang village, Ketapang Regency). Both sites had been previously investigated by USFS (United States Forest Service) collaborators and were known to contain variable peat thickness and multiple landcover types, while providing relatively easy access. The Tanjung Gunung site (hereafter referred to as TG) is adjacent to Gunung Palung National Park and its natural resources have been heavily exploited by the local community for decades. Within the TG site, two areas along the same peat formation were studied: a thinned, degraded forest (TG1) and a mature rubber plantation which is located at the edge of the peat formation (TG2). The physiographic terrain at TG is a 6 km wide swamp peatland known as Mendawai, MDW (RePPProT, Regional Physical Planning Programme for Transmigration, 1990) that is characterized by shallow peat. Kahayan (KHY) peaty alluvial plains are also formed along the seaward edges of MDW (inset in Fig. 1). Although the two selected study sites (TG1 and TG2) are only approximately 1 km apart and are both situated in a transition zone between KHY and MDW ecosystems, differences exist in terms of thickness of peat and organomineral transitional layers and water table depth. While TG1 is characterized by MDW properties (i.e., shallow peat swamps), TG2 is characterized by a mixture of MDW and KHY properties, including landforms such as coalescent estuarine and riverine plains with lithologies that include alluvium and marine sediments.
At the Pelang forest site (hereafter referred to as P), two areas along the
same peat formation were also studied: a thinned, degraded forest occurring
on approximately 4–5 m deep peat (P1), which transitioned to a cleared area
covered in secondary ferns and grasses, and a degraded forest (P2) heavily
used by a local village occurring on very deep peat (> 9 m).
Compared to the Tanjung Gunung sites (TG1 and TG2), Pelang forest sites are
characterized by extensive peatlands over about 20 km
Schematic showing the location of the study sites in West Kalimantan, Indonesia. A total of four sites were investigated: Tanjung Gunung Site 1 (TG1) and Site 2 (TG2), and Pelang forest Site 1 (P1) and Site 2 (P2). Inset shows details about the land system as classified after RePPProT (1990): Kahayan (KHY) mainly characterized by alluvial plains; and Gambut (GBT), Mendawai (MDW) and Klaru (KLR) characterized by swamps. Color scale indicates elevation above sea level.
Ground-penetrating radar (GPR) is a fast, reliable, and inexpensive
geophysical method for non-destructive mapping of shallow subsurface
features in peatlands at scales ranging from kilometers for geological
features influencing peatland hydrology such as eskers (Comas
et al., 2011), to centimeters for determination of bubble distribution in
peat blocks at the laboratory scale (Comas and Slater, 2007). The GPR
technique involves the transmission of short pulses of high frequency
electromagnetic (EM) energy into the ground, and measurement of the energy
reflected from interfaces between subsurface materials with contrasting
electrical properties. In the most common deployment, one antenna (the
transmitter) radiates short pulses of EM waves, and the other antenna (the
receiver) measures the reflected signal as a function of time. Reflections
are primarily caused by changes in water content, which in turn are
determined by sediment type and soil density. Reliable estimates of EM wave
velocity (
Summary of field sites including landcover, peat depth (from direct core measurements) and land system after RePPProT (Regional Physical Planning Programme for Transmigration 1990).
GPR surveys were performed using a MALÅ RAMAC system with 50, 100 and 200 MHz antennas, with the 100 MHz antennas proving the best compromise between depth of investigation and resolution. Malfunctioning of the 50 MHz antennas towards the end of the campaign prevented testing depth of penetration for this frequency at study sites with thicker peat columns. The spacing between traces was 0.2 m and 16 stacks (or replicates) were used for each trace. Two types of surface GPR surveys were performed: (1) common-offset surveys, where both transmitter and receiver antennas are kept at a constant distance as they are moved along transects and (2) common mid-point (CMP) measurements where transmitter and receiver are separated incrementally to larger distances. Common-offset surveys were used for subsurface imaging purposes (since profiles resemble a geological cross-section where depth is expressed as a travel time of the EM wave), whereas CMPs were used for velocity estimation.
ERI is a method for generating images of the variation in electrical resistivity in either two or three dimensions below a line or grid of electrodes placed at the Earth's surface. Data are acquired by measuring the voltage differences between electrode pairs in response to current injection between additional electrode pairs. Numerical methods are used to solve the Poisson equation relating the theoretical voltages at the electrodes to the distribution of resistivity in the subsurface. Inverse methods are used to find a model for the subsurface resistivity structure that is consistent with the recorded field data and also conforms to model constraints imposed (typically the resistivity structure varies smoothly). The resulting resistivity structure describes variations in the ability of subsurface soils and rocks to conduct an electrical current. The resistivity is strongly controlled by water content, chemical composition of the pore water and soil surface area/grain particle size distribution.
Electrical resistivity imaging was conducted using a four-electrode Wenner configuration with both 1 and 2 m electrode spacing. This spacing provided maximum imaged depths of about 16 m. The imaging depth was estimated from the model resolution matrix (Menke, 1989; see Binley and Kemna, 2005 for further details) that depicted relatively good resolution within this region when compared with the rest of the modeling domain. Measurements were performed using an ARES (Automatic Resistivity System) G4 2A resistivity meter with a 48-multi-electrode switch box. Inversion and forward simulations were performed with R2 software written by Andrew Binley (Lancaster University). R2 uses an iterative finite-element method to estimate resistivity values at user-specified element locations in a finite-element mesh. The regularization was based on the popular smoothness constrained approach used to solve for the minimum structure resistivity model that satisfies the data constraints.
A triangular mesh with a characteristic length one-quarter of the spacing at the electrodes and growing larger toward the edges (to account for decaying model resolution) was built using Gmsh, a three-dimensional finite-element mesh program (Geuzaine and Remacle, 2009). R2 requires an estimate of the error associated with each data point for convergence to be evaluated. For this purpose, it is best practice to collect reciprocal data (a companion data set where current and potential electrodes are reversed) to gain an informed estimate of the errors associated with ERI measurements (Slater et al., 2000), since underestimating these errors can produce image artifacts in the final ERI result which can mistakenly be interpreted as real structures. In lieu of reciprocal data, we employed a 2 % error model as input to R2 given the low electrical noise expected in our remote field sites and stacking errors (recorded on the instrument) of less than 1.1 %.
A total of nine core samples were obtained along the linear transects
established for geophysical surveys using an Eijkelkamp Russian-style peat
auger inserted vertically into the peat layer. Representative 5 cm peat soil
subsamples were taken at 0–30, 30–50 and 50–100 cm depth intervals and each
subsequent 100 cm interval until mineral substrate was reached. After
extraction of core samples, water tables were directly measured using a
measuring tape. The length of the sampling device was 9 m, so detection of
any deeper boundaries below 9 m using direct methods was not possible. Peat
layers were described in the field as “peat”, “transitional” (a mixing
horizon of peat and mineral soil) and “mineral soil” (mostly marine-derived fine silt and clay), which represented underlying mineral substrate.
The 5 cm subsamples were oven dried at 60
A set of geophysical surveys combined with direct sampling at each study site consisted of (1) one or more GPR common-offset transects between 30 and 100 m long to identify the peat–mineral soil reflector and other stratigraphic features (such as presence of layers rich in woody debris or buried buttressed trees) within the peat soil reflection record; (2) one or more GPR common mid-point surveys to estimate EM wave velocity along the peat column and convert two-way travel time into depth for common-offset profiles; (3) one or more electrical resistivity transects between 48 and 144 m long to provide additional information related to (a) peat thickness in regions where GPR was anticipated to fail due to thicknesses being greater than the GPR penetration depth and/or excessive GPR attenuation associated with high electrical conductivity and (b) variations in the lithology of the sub-peat mineral deposits; and (4) one or more direct soil cores in order to confirm depth of the peat–mineral soil interface and to obtain samples for subsequent C analysis at selected locations. Since not every core collected was analyzed for C content, Table 2 presents a summary of cores collected including average C percent and content along the peat column.
A set of two orthogonal common-offset profiles were collected at Site TG1 at
the 0 m distance in Line 1 (Fig. 2a) crossing Line 2 (Fig. 2b) at 24 m
along the profile. An average EM wave velocity of 0.04 m ns
Direct coring at two locations (shown in Fig. 2a and b) confirms a total peat
thickness of 4 m with a 0.1–0.2 m sandy clay transition (also containing
some organics) into a clayey mineral soil at about 4.2 m depth. Direct
coring also detected the presence of (1) a water table at 0.5 m depth
coinciding with the presence of a distinctive reflector in the GPR record
(particularly clear in Fig. 2b); (2) a woody area between 2 and 3 m depth
(indicated in Fig. 2) resulting in isolated points of core refusal that
coincide with the presence of hyperbolic diffractions in the reflection
record. Extracted core samples showed an average of 58.5 % C and C
content of 2311.0 Mg ha
Electrical resistivity imaging results for Line 1 and Line 2 at Site TG1 are shown in Fig. 3a and b, respectively. Direct cores as shown in Fig. 2 are superimposed for comparison. The resistivity inversion shows a relatively conductive (resistivity less than 100 ohm m) upper layer, underlain by a more resistive unit of undetermined thickness. The upper layer (showing a progressive increase in resistivity with depth between 60 and 200 ohm m) correlates with the terrestrial peat deposit as confirmed from direct sampling and GPR. The underlying resistive layer (ranging between 200 and 300 ohm m) includes both a transition layer composed of a mixture of sand and clay (with some organics) and a clayey mineral soil as confirmed from coring. Although lower resistivities are typical for clayey mineral sediments that are usually found below peat, in this case the higher resistivities are attributed to a sandy mineral soil matrix as confirmed from coring in the transition layer.
Summary of cores including coordinates, landcover, peat depth (from
direct coring), C stock along the peat profile (in Mg ha
GPR common-offset profile using a MALÅ GPR system with 100 MHz
antennae along Line 1
GPR common-offset profiles at Site TG2 (Figs. 4 and 5) identified a variable
peat column ranging between 0.1 and 3.4 m along the profiles. An average EM
wave velocity of 0.038 m ns
Inverted images of
Geophysical surveys constrained with direct coring at Pelang forest contrast
with those previously described at Tanjung Gunung with greater peat
thicknesses, ranging between 5 m at Site P1 up to 9 m at Site P2. GPR and
electrical resistivity surveys at Site P1 were collected at different
locations separated by about 1 km since GPR transects at this site were not
accessible with heavy resistivity instrumentation. Similar to Site TG1, an
average EM wave velocity of 0.04 m ns
GPR common-offset profile using a MALÅ GPR system with 100 MHz antennae at Site P1. Location of core sample P1.1 and inferred units and water table position are also shown. Larger white arrow indicates the center of a depressional feature within the reflection record centered between 10 and 35 m along the profile and 3–5 m depth. Smaller white arrow indicates the presence of a diffraction hyperbola.
Electrical resistivity imaging results at Site P1 (Fig. 7) show an interface
at about 5 m depth (as confirmed from coring) between an upper resistive
layer with a resistivity ranging between 150 and 300 ohm m interpreted as
peat, underlain by a conductive unit (as low as 30 ohm m) interpreted as
clay and confirmed from coring. These resistivity values are consistent with
those previously shown for Site TG2 in Fig. 4b. Although boundaries are not
clear, a transitional layer along the column between the peat and clay units
shows intermediate resistivity values (around 100 ohm m) and is coincident
with the mixture of sand, clay and organics, with a thickness of about 2.5 m
identified in the coring. Although not directly confirmed from coring, it
appears the interface between the peat and the sandy clay is variable across
the profile in Fig. 7, indicating undulating peat thickness between 5 m
(i.e., at core location at 22 m along the line, and at 70, 105 or 120 m
along the line based on ERI alone) and 7.5–8 m (i.e., at 12, 90 or 130 m
along the profile). The ERI profile also shows a strong lateral resistivity
variation in the deeper mineral soil (i.e., below 10 m depth) varying
between 30 and 100 ohm m from the SE to the NW. Cores P1.1 and P1.2
averaged 50.8 % C with a C content of 2677.1 Mg ha
Variability in peat thickness at Site P2 (Fig. 8) is similar to that
described for Site P1 (Fig. 7) and is confirmed at three coring locations (at
10, 50 and 100 m along the profile) resulting in total peat thicknesses of
9 m or more, 8.7 and 8.8 m, respectively. Since topography can be
considered flat at the scale of measurement used in this profile, these
results confirm that the interface between the peat and the underlying sandy
clay transition is undulating and that resistivity values for the peat
(between 100 and 185 ohm m) and transitional layer (below 100 ohm m) are
consistent with those shown in Fig. 7. The clay layer imaged with the
resistivity profile in Fig. 7 (and confirmed from coring in that figure) is
also visible in Fig. 8 just below the transitional layer and at approximate
depths between 10 and 14 m. For cores P2.1 and P2.2 the soils averaged
57.0 % C with a C content of 5892.3 Mg ha
Inverted image of resistivity survey at Site P1 using a four-electrode Wenner-type array with 2 m electrode spacing. Note that the resistivity profile does not coincide with the location of GPR profile shown in Fig. 6. Location of core sample P1.2 and inferred units (depicted in Fig. 6) are also shown.
In general, peat thickness estimates using GPR and ERI were consistent across sites although several differences between methodologies are noted. GPR was particularly effective for characterizing peat thickness for shallow peat columns (i.e., TG1 and TG2 in Figs. 2 and 5b, respectively) and able to quantify depth of the peat–mineral soil interface at centimeter-scale resolution both vertically and laterally from a strong reflector that matched closely with coring results. This reflector resembles the peat–mineral soil interface as typically detected with GPR in boreal peatlands in North America and Europe, exemplified in several studies for those higher-latitude systems (Warner et al., 1990; Jol and Smith, 1995; Slater and Reeve, 2002; Parsekian et al., 2012; Comas et al., 2013). However, the GPR method, as used with antenna frequencies available for this study, was limited for imaging deep (i.e., 9 m or more) peat columns (i.e., Sites P1 and P2) in this study. We attribute these limitations to (1) thicker peat columns that excessively attenuate the GPR signal, and/or (2) attenuation due to the presence of clay-rich transition layers with high electrical conductivities as depicted by the low resistivity values in P1 and P2 (Figs. 7 and 8). Attenuation in clay-rich areas was to be expected since it is well known than the effectiveness of GPR in peatlands is compromised when electrical conductivity of peat is high due to high electrical fluid conduction or high percent of clay fractions (Theimer et al., 1994).
Inverted image of resistivity survey at Site P2 using a four-electrode Wenner-type array with 2 m electrode spacing. Location of core sample P2.1, P2.2 and one additional location and inferred units (depicted in Fig. 6) are also shown.
Electrical resistivity imaging also proves useful for detecting changes in peat thickness across sites and for estimating the depth of interface between peat and mineral soil. When compared to GPR, electrical resistivity shows similar imaging capabilities for estimating both shallow and deep peat columns in the study areas (due to larger depths of investigation), although resolution (both vertical and lateral) is lower than that of GPR, particularly as depth increases. The boundaries between the upper resistive layer corresponding to the peat and the underlying conductive materials corresponding to the clay and transitional layer are not clear and are depicted by a gradual increase in conductivity (see Figs. 4b, 7, and 8). These results are consistent with previous studies in northern peatlands which demonstrate that electrical conductivity is not an accurate indicator of peat thickness when peat is underlain by a conductive layer (Slater and Reeve, 2002). The results presented here also confirm the same issue when peat is underlain by a resistive material (Fig. 3), which is not uncommon in Indonesia. For example, sandy mineral soils below the organic sediments of other peatlands in Central Kalimantan have been reported (Shimada et al., 2001). Despite these limitations, a good correspondence exists between the limit of the uppermost high resistivity values at sites TG2, P1 and P2 (depicted in red and orange in Figs 4b, 7, and 8) and the peat layer interface.
Although GPR and ERI data sets presented here are limited in terms of areal extent and scale of measurement, our intent was to test and demonstrate the potential of the methods for estimating peat thickness in tropical peatlands at better resolution than traditional methods (i.e., coring). Therefore, geophysical surveys were developed at plot-level scales with average profiles of 100 m, with the aim of upscaling measurements in subsequent studies. Furthermore, the ultimate aim of this work is to increase the accuracy of peat C storage estimates by using methods able to quantify peat thickness at high lateral resolution (i.e., reaching cm for GPR) when compared to coring. It is important to consider that GPR or ERI as applied here detects interfaces representing contrasts in physical properties which can be used to obtain highly accurate estimates of peat volume. When combined with sampling of representative peat soils for C density determination, total peat carbon storage estimates can be undertaken largely at the site level.
Comparison of peat thickness estimated from the
The profile from Site TG-2 in Fig. 5 can be used to investigate how subtle
changes in peat thickness as detected from GPR (representing a maximum
gradient below 0.02
Error bars in the GPR data (
The results presented here also demonstrate potential for using GPR and ERI
methods to improve the understanding of processes associated with peatland
formation. Differences in the GPR reflection record and contrasts in
electrical conductivity between the two study sites (TG and P) are
interpreted as differences in peat ecosystem type and developmental history
between sites. First, there is a sharp difference between the profiles at TG1
and TG2, as the resistivity profile increases with depth at TG1 (i.e., higher
resistivity at the bottom of the profile, Fig. 3), whereas it decreases at
TG2 (i.e., lower resistivity at the bottom of the profile, Fig. 3). Second,
the interface between peat and mineral soil at TG1 and TG2 is characterized
by a set of 2–3 sharp reflectors in the GPR record (see Figs. 2, 4 and 5),
which are absent at Site P where reflectors are sharply attenuated when
reaching depths corresponding to the transition zone between peat and clay.
Third, resistivity results do not show marked differences in terms of
electrical conductivity between sites along the peat–clay interface,
although coring results show a marked increase in thickness of the transition
zone (mostly corresponding to mixtures of clay and sand) with averages
between 0.1 and 0.2 m for Sites TG1 and TG2 and averages reaching 2.5 m for
Site P1. These differences may be attributed to two related issues: (1) the
developmental history of peatland initiation and formation at each specific
site and (2) the differences in site location as related to physiographic
type of terrain and the characteristics of peat ecosystems at each site. As
shown in Fig. 1, sites TG1 and TG2 correspond to MDW or shallow peat swamp
ecosystems, while sites P1 and P2 are characterized by GBT or large
ombrotrophic peat swamp ecosystems. Coastal peat swamps in Kalimantan have
been described as being the result of peat accumulation developed on marine
clay and mangrove deposits of river deltas and coastal plains during the mid- to late
Holocene (
Finally, the spatial resolution provided by GPR common-offset profiles also
shows the potential for better understanding the nature and internal
structure of the peat matrix. For example, referring to the presence of
hyperbolic diffractions in the GPR record, Figs. 2a, b and 5 show the
presence of several areas with a high density of diffractions. These
diffractions are particularly abundant in Fig. 2a between 10 and 20 m
distance along the profile and at 2.5–3 m depth, or in Fig. 5 between 70
and 85 m distance along the profile and between 2 and 3 m depth (white
arrows in Fig. 5). Diffractions are associated with the presence of objects
that may act as isolated reflector points such as cobbles and boulders (Neal,
2004). In this case, we associate hyperbolic diffractions in GPR common
offsets with the presence of buried woody debris (as further confirmed
through coring). Other investigations in northern peatlands have also related
GPR diffractions to the presence of wood (Slater and Reeve, 2002; Comas et al., 2008). Such
features are absent at P1 (Fig. 6), where more laterally continuous
reflections (i.e., at 3, 4 and 4.5 m depth between 40 and 90 m along the
profile) are present. Previous studies in the Kalimantan region have also
consistently shown layers with large quantities of undecomposed woody
fragments heterogeneously distributed within the peat column (Shimada et al.,
2001). Furthermore, some of these laterally continuous reflectors generate a
depressional feature between 10 and 30 m along the profile of P1 (center
point indicated by a white arrow in Fig. 6) as depicted by a sharp reflector
at depths between 3.5 and almost 6 m that tilts 13 and 9
This study demonstrates the feasibility of using GPR and ERI for non-invasive
mapping of the subsurface of peatlands in Indonesia, at a spatial resolution
previously unreported in tropical peatland systems, which are traditionally
assessed using coring methods. The results presented highlight the
opportunity to use the reflection record from GPR to improve peat thickness
estimates while providing information on certain attributes of the peat
matrix such as presence of wood layers, buttressed trees or tip-up pools or peat soil
origins related to peatland ecosystem type (i.e., mangrove vs. freshwater
peat). While in general GPR is able to predict peat thickness with centimeter
resolution some limitations emerged (i.e., signal attenuation) for peat
columns exceeding 5 m thick. Although the vertical resolution of ERI is more
limited, peat thickness determination shows comparable results for either
shallow or deep peat columns. A comparison between peat thickness estimates
from GPR, ERI and coring showed a variability exceeding 2 % in peat
surface area (or 1191 kg of C assuming C contents of 170 kg C m
This work was supported by the US Agency for International Development (USAID). We are indebted to Kent Elliot (US Forest Service) for all his help with logistics and fieldwork during this study. We are also thankful to Sofyan Kurnianto (CIFOR, Oregon State University) and Ophelia Wang (USAID-IFACS) for their help in the field, and to all local field assistants and guides involved in this research for their support at the study sites. We thank R. Dommain for helpful discussions and revisions in the materials presented in this study. We also thank four anonymous reviewers and the editor for helpful comments which improved an earlier version of this manuscript. Edited by: A. Ito