forest-carbon-in-png-julian-fox-et-al-biotropica

Key words: Carbon, Biomass, Sequestration, Forest dynamics, Flux, Degradation, Selective-logging, REDD, ENSO, Deforestation, Secondary... Quantifying forest Carbon C in primary and seco

Title: Forest carbon in Papua New Guinea

Authors: Julian C Fox 1*, Cossey K Yosi 1,2 and Rodney J Keenan 1

Affiliations: 1 Department of Forest and Ecosystem Science, The University of

Melbourne Burnley Campus, 500 Yarra Blvd, Richmond, Victoria 3121 Australia

*Correspondence: Phone: +61 3 9250 6862, Fax: +61 3 9250 6886 email:

jcfox@unimelb.edu.au 2 Papua New Guinea Forest Research Institute, PO Box 314, Lae

411, Morobe Province, PNG

Key words: Carbon, Biomass, Sequestration, Forest dynamics, Flux, Degradation,

Selective-logging, REDD, ENSO, Deforestation, Secondary

Quantifying forest Carbon (C) in primary and secondary tropical forest is one ofthe challenges of climate change mitigation initiatives such as reduced emissions fromdeforestation and degradation (REDD) Papua New Guinea (PNG) has become the focus

of the REDD initiative, but defensible estimates of forest C are lacking Here we present

a methodology for estimating forest C from a large Permanent Sample Plot (PSP)network, and report the first defensible estimates of forest C in undisturbed andselectively-logged (degraded, secondary) forest in Papua New Guinea This paperrepresents the first published account of this large and important census of PNG’s diversetropical forests

Average above ground live biomass in trees greater than 10cm diameter(AGLB>10cm) in 135 selectively-logged 1 hectare plots was 66 Mg C ha-1 with a standarddeviation (SD) of 19, while for 20 undisturbed plots the average was 110 Mg C ha-1 (SD28) By estimating unmeasured above ground C components, total above ground biomass(AGB) was estimated as 92 Mg C ha-1 and 154 Mg C ha-1 for selectively-logged andundisturbed forest respectively Our estimate for undisturbed forest is lower than biomeaverages for tropical equatorial forest; 180 Mg C ha-1 IPCC (2006), 202 Mg C ha-1 Lewis(2009) Our estimate for degraded secondary forest is higher than previous estimates,suggesting that the selective-logging practiced in PNG (targeting high-value speciesabove a 50cm diameter limit) has a lesser impact on forest C than other anthropogenicdisturbances Secondary forests in PNG have previously been assumed to hold little valuefor either timber or carbon, but these higher estimates suggest that they should be valued,and perhaps actively managed for the carbon they contain Provincial averages for

AGLB>10cm in selectively-logged forest varied from Central and Oro Provinces with

averages of c 50 Mg C ha-1 to Western Province with c 80 Mg C ha-1

Observed forest C reported here are the first defensible measurements for PapuaNew Guinea, and represent a critical step toward REDD implementation

Introduction

Papua New Guinea (PNG) along with other rainforest nations have recentlybecome the focus of climate change mitigation efforts with the ‘reducing emissions fromdeforestation and degradation’ (REDD) initiative of the United Nations FrameworkConvention on Climate Change (UNFCCC) Developing tropical countries such as PNGface many challenges in reporting for the REDD initiative Estimating forest C pools indifferent forest stratum such as primary and secondary forest is an important precursor toREDD implementation (Gibbs et al 2007) Here we contribute to REDD implementationfor PNG by quantifying above ground carbon (C) stock in undisturbed (primary) andselectively-logged (secondary, degraded) forest across a PSP network initiated andmaintained by Papua New Guinea Forest Research Institute (PNGFRI) A majority (112)

of the PSPs were established in selectively-logged forests, with undisturbed forest beingrelatively poorly represented (13 plots) Based on PSPs in secondary forest we candetermine defensible provincial averages for above ground biomass (AGB) for thesecondary forest stratum

Secondary forest can be defined as forest which has been disturbed and is at somestage of regeneration, and have been estimated to comprise 40% of all tropical forest(Brown and Lugo, 1990) Considering that 40% of terrestrial biomass is stored in tropical

forests (Phillips et al 1998), secondary tropical forest thus represents a very significant

global C pool (c 20%) that has considerable (but unverified) potential for C flux into the

future (Brown et al 1996, Fehse et al 2002) Secondary tropical forest remains a poorlyunderstood resource relative to primary forest (Sierra et al 2007b) Many studies fail toadequately distinguish between primary and secondary forest (Houghton et al 2001), andmerge estimates of forest C over the two stratum Consistent with this, secondary forest

in PNG is a large and poorly understood resource considered to hold little value for either

timber or carbon (PNGFA pers comm.) Using statistics from 2002, PNG Forest Authority estimated that undisturbed forest covers an area of c 29.7 million hectares whilst secondary forest covers c 3.3 million hectares (PNGFA pers comm.) However,

recent studies indicate that the area of secondary forest may be rapidly expanding (seeShearman et al 2009) Assessment of this large and expanding forest resource is apriority and could potentially facilitate it’s inclusion in climate change mitigation efforts

Many previous studies of tropical forest C and C flux have been plagued bymethodological problems that limit the veracity of the estimates (Clark et al 2001b,Phillips et al 2002) Several important problems have been identified; small plot sizes ofless than 1 ha (0.25 ha is common) limit the representativeness of measured forest and islikely to result in overestimates generated by larger trees being over-represented (Brownand Lugo 1992); There is lack of replication in both time and space which will again limitthe representativeness of measured forest and will result in results skewed toward patches

of forest with the highest biomass (Clark et al 2001a, 2001b, Phillips et al 2002, 2004).These methodological problems are exaggerated for tropical forests, due to large spatialvariations in structure, productivity, and the presence of large trees There is clearly a

need for studies with large plots that sample widely across both time and space (Clark et

al 2001a, 2001b, Phillips et al 2004) The PSPs used in this study overcome many ofthese methodological issues, and provide a sound basis for the estimation of forest C and

C flux; plots are large (1 ha) and are replicated through both space and time We willestimate above ground live biomass (AGLB>10cm) for each PSP measurement and examinenational and provincial averages

Another methodological issue is introduced when tree variables measured fortimber inventory purposes are used to estimate biomass and C (Lindner and Karjalainen2007) Two methods are commonly used; the first converts tree volumes to biomass using

a biomass expansion factor (Segura 2006); the second uses previously developedallometric equations to estimate biomass per tree as a function of tree parameters such astree diameter, tree height and wood density (Brown et al 1989, Clark et al 2001, Chave

et al 2003, Baker et al 2004) These allometric equations will have been derived frombiomass harvesting studies (Brown et al 1989, Chambers et al 2001, Chave et al 2001),and their application is dependent on the availability of equations for a similar forest, andalso for similarly sized trees (Chave et al 2005) Many allometric equations use onlydiameter to predict tree biomass, however, including wood density and height canimprove the accuracy of tree level predictions (Chave et al 2004, Chave et al 2005),particularly considering the variation in tree architecture and wood density in tropicalforests Despite possible errors in estimating AGLB, Baker et al (2004) found thatestimated AGLB flux was unaffected by to the type of allometric equation used Giventhe absence of allometrics for PNG we are forced to convert measured tree parameters totree biomass using allometrics derived from other equatorial tropical forest In doing this,

it is important to include drivers of tree architecture and the physiological characteristicsthat determine C composition such as diameter, height, and wood density (Chave et al.2005).

The only previous estimate of forest C in PNG known to the authors is Edwardsand Grubb (1977); in a study of lower montane rainforest (at 2500m) they estimated totalbiomass of 175 Mg C ha-1 consisting of 151 Mg C ha-1 in tree AGLB In a majorcompilation of biomass measurements in primary tropical forests, Clark et al (2001a)revealed that above ground biomass (AGB) varied from as low as 40 to as high as 250

Mg C ha-1 Despite many assessments of primary tropical forest biomass, comparativelyfew studies differentiate between primary and secondary forest (Houghton et al 2001).However, this is changing with the realisation of the increasing importance of secondaryforests in the composition of tropical landscapes For example, Sierra et al (2007a)explicitly compare C stock in primary and secondary Colombian rainforest (AGB 111 Mg

C ha-1 and 21 Mg C ha-1 respectively) They found that the AGB was most sensitive toanthropogenic disturbance, with significant differences in estimates for primary andsecondary forest

The objectives of this study are to quantify above ground C pools in secondaryand primary forest in PNG This required a sound methodology with considered errorcorrection techniques and the development of appropriate tree allometrics Results willelucidate the role of degraded secondary forests as a C pool Assessment of the Ccontained in these forests may facilitate their potential inclusion in REDD negotiations

We also intend to provide provincial averages of secondary forest C for specificapplication within PNG

Materials and Methods

We estimate AGLB>10cm for each PSP measurement Consistent with previousstudies, AGLB>10cm will be reported in megagrams of carbon per hectare (Mg C ha-1) The

C content of biomass will be reported assuming that dry biomass is 50% C (Clark et al.2001a, Houghton et al 2001, Malhi et al 2004) This is an acceptable approximation;however, the wood C fraction does exhibit some small variation across species and treeages (Elias and Potvin 2003)

PNGFRI’s PSP database

Over the last 20 years PNGFRI has established and remeasured over 135 PSPsacross PNG covering all major forest types A map of PNG showing provincialboundaries and PSP locations is shown in Figure 1 Each PSP plot is one hectare in sizeand is divided into 25 sub-plots of 20 m x 20 m The spatial location, diameter, height,and crown characteristics are recorded for all trees over 10cm The PSP databaserepresents a strong basis for the estimation of biomass and C in these forests For furtherdetails of the PSP data refer to Yosi et al (2009)

PSPs were measured following a field procedure (PNGFRI 1994) Despite thiswell developed and uniformly applied field procedures, problems arise in large databasesdue to measurement and transcription errors (Baker et al 2004) To identify potentialerrors the distribution of diameter increments was examined, and those less than -0.2, orgreater than 2.6 cm yr-1 were flagged for investigation This represented approximately1% on each tail of the increment distribution, and 2% of all increments in total These

values are similar to those used by Chave et al (2003) and Baker et al (2004) to flagerroneous measurements Examination of diameters for flagged trees often revealedtranscription errors, such as an extra zero, or a missing zero These were corrected on atree by tree basis, with careful adjustment of the erroneous diameter measurement When

it was clear a measurement error had occurred, the erroneous diameter was correctedusing a species specific diameter growth model This methodology was followed to avoidthe significant biases introduced if erroneous records are removed (Chave et al 2003),and to improve on previous error corrections using interpolation (e.g Chave et al 2003),stand-level (e.g Baker et al 2004) or species-level averages (e.g Rice et al 2004).Measurement errors tend to most prevalent in larger buttressed trees, and removal ofthese records can have a significant effect on biomass estimates

Species-specific increment model for error correction

First we needed to identify a diameter-diameter increment base model that is mostappropriate for observed tree growth The relationship between diameter and diameterincrement is non-linear and sigmoidal; the curve starts at the origin and rises to amaximum diameter increment at an inflection point before falling to an asymptoticdiameter increment (Zeide 1993, Huang and Titus 1995) Several base models thatcharacterise this relationship were examined; the Box-Lucas function (1) (Box and Lucas1959), and a simplified model (2) from Huang and Titus (1995);

(e bD e aD)

b a

Where Dincr is diameter increment in cm yr-1, D is measured diameter at breast height over bark, and a and b are parameters to be estimated.

Species-specific models were initially fitted using NLIN in SAS, and the Lucas base function was found to provide superior fit for species represented on PSPs.However, model fitting for individual trees within PSPs is affected by a nesteddependence structure; Diameter-diameter increment relationships for the same specieswithin a PSP will be more similar than that between each PSP, as trees on the same plotwill be subject to the same local environmental conditions, and will be of a similar foresttype (Fox et al 2001) We can explicitly account for this nested dependence using a non-linear mixed model, with a separate random parameter for each PSP To facilitate this,SAS’s Proc NLMixed was used to fit the non-linear Box-Lucas mixed model (Wolfingerand O’Connell 1993, Davidian and Giltinan 2003) This will ensure correct statisticalinference within and between PSP plots, as well as plot localised increment predictionsthat can replace erroneous measures Fitted Box-Lucas models for four species are shown

Box-in Figure 1 Horsfieldia spp and Celtis latiflia are both pioneer species, and their curves

have a maximum diameter increment at a small diameter of approximately 25cm, and

then approach zero increment for the larger diameters Pometia pinnata and Celtis phillippensis are climax species with higher diameter increments across the full range or

diameters Model (1) parameters for the 50 most common species on PSPs can be found

in Table S2 of supplementary material

Species-specific predictions of diameter increment were applied whenmeasurement errors had occurred; less than -0.2, or greater than 2.6 cm yr-1 and with no

obvious transcription errors Some pioneer species (Macaranga spp., Spondias spp.,

Hernandia spp., palaquium spp., melanolepis spp., antiarus spp., litsea spp., trichospermum spp., artocarpus spp., sterculia spp., Trema spp., Elaeocarpus spp., Labula spp., Endospermum spp, Octomeles spp) were found to have valid growth rates

that exceeded 2.6 cm yr-1 Trema spp and Macaranga spp appear capable of

extraordinary growth rates of up to 6 cm yr-1 These exceptional growth rates were notaltered From the total of 153900 tree records, 326 (0.2%) were obvious transcriptionerrors that were manually corrected, and 3418 (2%) were erroneous measurements thatwere corrected using modeled diameter increment

A large number of PSP plots were affected by the El Niño-Southern Oscillation(ENSO) event of 1997/1998 and these plots were set aside, as the high rates of treemortality resulted in declines in forest C and a negative flux that skewed analysis ofunaffected remeasurements

Estimating above ground living biomass

The first step in quantifying forest C is to estimate AGLB in standing trees Therehas been much recent work on the development of allometric equations for estimatingbiomass for tropical forests from tree inventory information Typically, they are modelsderived from destructively sampled trees and easily measured biometric variables such asdiameter and height (Liddell et al 2007) In an extensive study of allometric models fortropical forests, Chave et al (2005) found the most important predictors of AGB werediameter, wood specific gravity, total height, and forest type (dry, moist, or wet) Theydeveloped a model (3) for wet tropical forests that was used to estimate AGB for trees onPSPs;

[ 2 ]0 940

*0776

Where D i is diameter in centimeters, H i is total height in meters, and q i is wood specific

gravity in grams per cubic meter for tree i The resulting AGLB i estimated from the

equation is the total biomass of the stem, crown and leaves for tree i in kilograms Chave

et al (2005) found that locally, the error on the estimation of a tree’s biomass was in theorder of ±5% Average AGBL (and other tree statistics) for the 50 most common species

on PSPs can be found in Table S1 of supplementary material

Total AGLB>10cm was quantified by summing tree level AGLBi estimates for all jtrees on the one hectare plots (4)

Plot level AGLB>10cm estimates are in Mg C ha-1

Estimating total above ground biomass (AGB)

It is appropriate and reasonable to estimate total AGB from the measuredcomponent (AGLB>10cm) based on previously established relationships and literaturereviews (Gibbs et al 2007) AGB can be estimated as the sum of AGLB>10cm and AGLB

in trees less than 10cm (AGLB<10cm) and other plants, and non living biomass (NLB)components (5);

NLB AGLB

AGLB

AGB= >10cm + <10cm + (5)

AGLB<10cm consists of small trees (<10cm), palms, shrubs, vines, and herbaceous plants.

In undisturbed forest the biomass in this unmeasured component has been estimated to be

between 3 and 7% of the AGLB >10cm (Chave 2003, Baker et al 2004) However, insecondary or disturbed forest this fraction may rise to as much as 30% depending on the

degree of disturbance and the age of the secondary forest (Brown and Lugo 1990, Lugoand Brown 1992) Palms are often observed on PSPs, but are not included in treeinventories, and in some of the more heavily disturbed PSPs, small regrowth, shrubs andherbaceous plants can be plentiful Based on field observations, and communicationswith measurement staff, we estimate that this component is on average 20% ofAGLB>10cm

NLB consists of necromass in coarse woody debris (CWD) consisting of standingdead trees and fallen trees, and fine litter (FL) consisting of remaining dead plant material(fruits, leaves, flowers, and small branches) on the forest floor FL in tropical forests hasbeen estimated to be 5% of AGLB>10cm (Brown and Lugo 1982) CWD is potentially avery large C pool, particularly in disturbed forest, and may constitute 10-40% of aboveground biomass (Uhl and Kauffman 1990) Based on field observation we estimate CWD

to be 15% of the AGLB>10cm Therefore, we estimate total NLB as 20% of AGLB>10cm.This is at the upper end of previous estimates [10-20% (Houghton et al 2001, Achard et

al 2002)], but this appropriate given the large quantity of CWD in secondary forest

Following the estimated contribution of AGLB<10cm and NLB, AGB can beestimated as follows (6);

4.1

10 ×

= AGLB> cm

Species-specific height-diameter models

Biometric modeling of tree heights is required to generate AGB estimates for alltrees on PSPs Tree heights are only measured when PSPs are first established, and fornew ingrowth Height-diameter (HD) relationships have been found to vary for different

species in tropical rainforest, for example, Yamada et al (2006) found that parameters for

a HD model varied significantly among species for a Bornean rainforest Chave et al.(2003) used species-specific height-diameter models to predict height for all trees for thepurpose of biomass estimation A similar methodology will be followed here

To achieve specific height-diameter models we fitted several specific non-linear HD relationships that were found to perform well for tropical forests

species-in the study of Fang and Bailey (1998); the two parameter log-lspecies-inear model (Alexandrosand Burkhart 1992) (7); the two parameter hyperbolic model (Huang and Titus 1992) (8);the three parameter Exponential model (Fang and Bailey 1998) (9);

bLogD

a

)/(b D

prediction on PSPs An example of the fitted hyperbolic HD model for Myristica spp.

(Nutmeg) is shown in Figure 2 Model (9) parameters for the 50 most common species onPSPs can be found in Table S2 of supplementary material

compiled for tree species on PSPs The range of wood densities was apparent with low

density wood species such as Cananga Odorata having a density of 0.275 and high density species such as Xanthostemon with a density of 0.85 Because wood density has a multiplicative effect on biomass and C Xanthostemon will contain three times as much woody biomass (and C) as a similarly sized Cananga Odorata For species for which no

wood density information is available, an average value across all species on PSPs wasused (Brown, 1997, Chave et al 2003) The average value across species present on PSPswas 0.477 This is similar, although marginally smaller than the average value of 0.54 forthe 50ha plot on Barro Colorado Island (Chave et al 2003) This may be because theaverage of Chave et al (2003) was based on primary tropical forest, whereas ours isbased on secondary forest with a higher representation of lower wood density pioneerspecies Wood density for the 50 most common species on PSPs can be found in Table S1

of supplementary material

Summarising AGLB across PSPs

Following the estimation of AGLB>10cm and AGB for each PSP remeasurement,the following methodology was followed to generate overall and provincial averages.Initially, remeasurements that were affected by ENSO induced fires in 1997/1998 wereseparated from the main analysis PSPs in selectively-logged and undisturbed forest werealso analysed seperately Averages for AGLB>10cm and AGB were estimated across allremeasurements of PSPs This assumes that PSP remeasurements are a representativesample of the growth stages of selectively logged forest The sample size of selectively-

logged and undisturbed PSPs, and the number that were affected by ENSO fires is shown

in Table 1

Results

Above ground live biomass

Average AGLB>10cm for 341 non fire-affected measurements in selectively-loggedforest between 1992 and 2008 is 66 Mg C ha-1 (SD 19) Average AGLB>10cm for 20 nonfire-affected measurements in undisturbed forest is 110 Mg C ha-1 (SD 28) Fromequation (4) we can estimate average AGB for selectively-logged forest as 92 Mg C ha-1,and undisturbed forest as 154 Mg C ha-1

Figure 3 and Table 2 show the average AGLB>10cm in selectively-loggedmeasurements for each province of Papua New Guinea Averages vary from Oro and

Central Provinces with c 50 Mg C ha-1 to West Sepik, Western and East New BritainProvinces with 75-80 Mg C ha-1 These differences reflect differences in forestcharacteristics and productivity across the provinces Central and Oro Provinces havedrier, less productive forests, while Western Sepik, Western, and East New BritainProvince have highly productive wet forests Table 3 indicates that the number ofremeasurements in each province varies from a minimum of 9 in Manus to 67 in MorobeProvince More confidence should be placed in estimates for provinces with manymeasurements such as Morobe, East New Britain, West New Britain, and Madang Table

2 also details comparisons of provincial AGLB>10cm to undisturbed AGLB>10cm Theaverage change (ΔAGLB>10cm) is also included as a percentage

Forest timber inventory data has been widely used to estimate forest C and has theadvantage of being an extensive and representative sample of the forest, often at anational level (Brown and Gaston 1995, Phillips et al 1998, Baker et al 2004, Lindnerand Karjalainen 2007) This is the case with PNGFRI’s PSP network However, timberinventory data does have important shortcomings for estimating C; the sample may bebiased toward productive forest that contains current or future merchantable timber,measured tree parameters may be insufficient for estimating tree biomass or C accurately,non-merchantable trees such as palms are not sampled, trees smaller than 10cm DBH arenot sampled, and other important C sinks (understorey, underground, and necromass) arenot sampled (Chave 2003) Data used in this study has these shortcomings, but wecontend that the application of appropriate allometrics has resulted in sound estimates ofAGLB Despite this, the development of localised allometrics for the forests of PNG may

be warranted The models of Chave et al (2005) incorporated the biomass harvesting data

of Edwards and Grubb (1977) from the montane forests (2400m) of PNG Given thecurrent interest in forest C, perhaps it is timely to supplement this with a biomassharvesting study in lowland forests The destructive sampling of 2-3 large trees may beall that is required to check and validate estimates from existing allometric equations(Gibbs et al 2007) Beyond these shortcomings, timber inventory data has beenextensively used for studying C and C flux (Baker et al 2004), and is a practical andaccurate method for estimating the C balance of tropical forests (Chave 2003) Theseminal work of Phillips et al (1994, 1998) and Lewis et al (2009) that detectedincreased turnover rates, and increasing biomass in primary Amazonian forest (although

contentions exist; Clark 2002, Wright 2005) was also based on forest timber inventorydata.

PSP plots are often located in proximity to roads or villages which hasimplications for anthropogenic disturbances from gardening and cultural burning throughthe census period This may have implications for our undisturbed PSPs, as it is possiblethat they may have been subject to some degree of previous disturbance This mayexplain why undisturbed AGB (154 Mg C ha-1) is less than biome averages for tropicalequatorial forest compiled by Gibbs and Brown (2007b) (164 Mg C ha-1) the IPCC (2006)(180 Mg C ha-1), and most recently to the compilation of pan-tropical plots in Lewis et al.(2009) (202 Mg C ha-1) However, they fall in the middle of the range identified by Clark

et al (2001a); 40 – 250 Mg C ha-1 The veracity of our AGB estimate for undisturbed islimited by our small sample size This sample in undisturbed forest needs to be increased,and future work will look at the feasibility of this Recently we established 2 PSPs in

undisturbed high altitude (3000m) Nothofagus forest in Simbu Province An increased

sample of undisturbed forest could facilitate valid within province averages for thisstratum, at present we are forced to average across all available plots

Average secondary forest AGB was 92 Mg C ha-1; this is higher than secondaryforest resulting from other land uses such as shifting agriculture (Sierra et al 2007a; AGB

21 Mg C ha-1) and more intensive selective-logging practices in other regions (Pinard andPutz 1996; AGB 68 Mg C ha-1) This may be because selective-logging as practiced inPNG (targeting high-value species above a 50cm diameter limit) has a lesser impact onforest C Despite intentions of randomly selecting forest for census from withinsecondary forest, the PSP network is susceptible to plot selection bias; plots may have

been positioned in areas that contained future merchantable timber, with heavilydegraded areas or areas with no potential for future timber stock being avoided There is

no way to evaluate this plot selection bias, and we need to remain mindful of thispotential bias that may have inflated our estimates of secondary forest AGB

Secondary forests in PNG have previously been assumed to hold little value for

either timber or carbon (PNGFA pers comm.), but the higher estimates reported here

suggest that they should be valued, and perhaps actively managed for the C they contain.Secondary forest in PNG is dominated by species that are not highly valued for timber,

however, these same species have equal value to the highly prized Kwila (Intsia bijuga)

when valued for the C they contain It has been suggested that accessible primary forestmay soon be exhausted of its high value timber species (Shearman et al 2008) If this isthe case, commercial logging operators will turn their attention to secondary forest, andthe secondary timber species it contains If re-entry of secondary forests by commerciallogging operators occurs, then these forests could potentially be included in REDDnegotiations The analysis of secondary forest C we have reported here suggests thatthese forests should be valued for the significant C resource they contain, and perhapsexplicitly managed to that end

Clark (2002) and Phillips et al (2002) compiled an extensive dossier of themeasurement errors that may affect tropical forest census From this list, and our personalexperience, we believe the most significant source of error for the PSPs are themeasurement errors associated with buttressing Buttressing is very common in rainforesttrees, and becomes more prevalent as the trees get larger, with stem deformity oftenreaching several metres up the bole from the base These deformations make accurate