OPEN ACCESS

Tropical forest carbon assessment integratingsatellite and airborne mapping approachesTo cite this article Gregory P Asner 2009 Environ Res Lett 4 034009

View the article online for updates and enhancements

Related contentMonitoring and estimating tropical forestcarbon stocks making REDD a realityHolly K Gibbs Sandra Brown John ONiles et al

-

Measurement and monitoring needscapabilities and potential for addressingreduced emissions from deforestation andforest degradation under REDD+Scott J Goetz Matthew Hansen Richard AHoughton et al

-

Reference scenarios for deforestation andforest degradation in support of REDD areviewof data and methodsLydia P Olander Holly K Gibbs MarcSteininger et al

-

Recent citationsBuilding Pareto Frontiers under tree-levelforest planning using airborne laserscanning growth models and spatialoptimizationAdriaacuten Pascual

-

Inventory of Forest Attributes to Supportthe Integration of Non-provisioningEcosystem Services and Biodiversity intoForest Planningmdashfrom Collecting Data toProviding InformationThomas Knoke et al

-

Assessment of linear relationshipsbetween TanDEM-X coherence andcanopy height as well as abovegroundbiomass in tropical forestsM Schlund and H -D V Boehm

-

This content was downloaded from IP address 61772279 on 24092021 at 1756

IOP PUBLISHING ENVIRONMENTAL RESEARCH LETTERS

Environ Res Lett 4 (2009) 034009 (11pp) doi1010881748-932643034009

Tropical forest carbon assessmentintegrating satellite and airborne mappingapproachesGregory P Asner

Department of Global Ecology Carnegie Institution for Science 260 Panama Street StanfordCA 94305 USA

E-mail gpastanfordedu

Received 10 April 2009Accepted for publication 25 August 2009Published 7 September 2009Online at stacksioporgERL4034009

AbstractLarge-scale carbon mapping is needed to support the UNFCCC program to reduce deforestationand forest degradation (REDD) Managers of forested land can potentially increase their carboncredits via detailed monitoring of forest cover loss and gain (hectares) and periodic estimatesof changes in forest carbon density (tons haminus1) Satellites provide an opportunity to monitorchanges in forest carbon caused by deforestation and degradation but only after initial carbondensities have been assessed New airborne approaches especially light detection and ranging(LiDAR) provide a means to estimate forest carbon density over large areas which greatlyassists in the development of practical baselines Here I present an integrated satellitendashairbornemapping approach that supports high-resolution carbon stock assessment and monitoring intropical forest regions The approach yields a spatially resolved regional state-of-the-forestcarbon baseline followed by high-resolution monitoring of forest cover and disturbance toestimate carbon emissions Rapid advances and decreasing costs in the satellite and airbornemapping sectors are already making high-resolution carbon stock and emissions assessmentsviable anywhere in the world

Keywords biomass mapping carbon accounting deforestation forest degradation LiDARREDD satellite mapping UNFCCC

1 Introduction

The United Nations Framework Convention on ClimateChange (UNFCCC) is facilitating large-scale forest monitoringfor Reduced Deforestation and Degradation (REDD httpunfcccintmethods scienceredditems4531php) In supportof REDD the Intergovernmental Panel on Climate Change(IPCC 2006) provided guidelines to assist countries indeveloping carbon assessment methodologies Theseguidelines are organized into three lsquoTiersrsquo each providingsuccessively increased accuracy and thus potentially higherfinancial returns for monitoring and verifying carbon stocksand emissions The Tier I approach is the most generalbased on simple nationwide estimates of forest cover andgeneric forest carbon density values (eg tons of carbon per

hectare) Tiers II and III provide increased detail on carbonstocks and emissions at regional and national levels using acombination of plot inventory satellite mapping and carbonmodeling approaches To achieve Tier III levels of accuracyboth aboveground and belowground live and dead carbonstocks must be estimated and modeled (Gibbs et al 2007)

At the national scale many tropical countries will relyinitially on Tier I levels of accuracy based on averagebiomass and soil carbon values assigned for biomes and largegeographic regions which will generate large uncertaintiesand thus potentially lower carbon credits (Gibbs et al 2007Angelson et al 2009) Conservative accounting guidelinesfrom the IPCC require that users of Tier I estimates assign arelatively low carbon stock per hectare compared to what mightbe achieved if more detailed measurements are conducted

1748-932609034009+11$3000 copy 2009 IOP Publishing Ltd Printed in the UK1

Environ Res Lett 4 (2009) 034009 G P Asner

Developing regional and national monitoring capacities aboveTier I accuracies will require improved high-resolution carbonmapping and modeling approaches with the pay-off realizedas increased carbon credit boosted carbon sequestration andimproved ecosystem protection

Gibbs et al (2007) point out that in tropical foreststhe majority of carbon is sequestered in aboveground livetissues (eg trees) with secondary stocks in soils and coarsewoody debris Root and soil carbon stocks average about20 of total carbon stored in tropical forests (Cairns et al1997) and the necromass produced by tree mortality averagessim10 of aboveground live biomass (Brown et al 1995Keller et al 2004) One major exception rests in the largebelowground carbon stocks found in peat swamps of SEAsia Notwithstanding peats the major unknown rests in theaboveground live biomass and thus it is my focus here

To monitor aboveground carbon stocks including carbonlosses and gains caused by deforestation forest degradationand forest recovery a combination of data is required (i) therate of change in forest cover and disturbance and (ii) theamount of carbon stored in the forest (lsquocarbon densityrsquo inunits such as tons of carbon per hectare tons C haminus1)Various satellites measure forest cover canopy loss anddisturbance and metrics of forest structure (Chambers et al2007) but no satellite technology can directly measure carbondensity (GOFC-GOLD 2009) Satellites thus provide anopportunity to monitor changes in forest carbon caused bydeforestation and degradation but only if carbon densitieshave been assessed Traditionally carbon densities havebeen assessed using field-based inventory plots which arevaluable but also expensive time consuming and inherentlylimited in geographic representativeness A survey of themajor plots networks in Amazonia suggests that less thanone-millionth of the region has been measured in the fieldalbeit in an effort to be statistically representative (synthesizedby Phillips et al 1998 Saatchi et al 2007) However thenatural variation in forest structure and biomass combinedwith the enormous geographic extent and rate of forest lossand disturbance makes field plots difficult to use (alone) forassessing tropical forest carbon densities over large areas Suchfield-based monitoring would require thousands of plots andregular revisits using laborious hand measurement techniquesNew approaches are needed to extend field plot networksand to bridge a gap between field measurements and satelliteobservations

Airborne mapping methods may assist in developingcarbon stock estimates in tropical forests (Brown et al 2005)The latest airborne approaches especially light detection andranging (LiDAR) can be used to estimate aboveground carbonstocks over large areas (reviewed by Lefsky et al 2002b)LiDAR mapping when combined with field calibration plotscan yield aboveground carbon maps over thousands of hectaresper day of flying Here I propose that LiDAR-based mapscan be used in a baseline analysis to initiate a long-termmonitoring program that will subsequently rely primarilyon low-cost satellite data and analysis methods Howevergiven the limited work attempting to integrate airbornetechnologies into the carbon mapping process for REDD

there are no operational or even any clearly proposed methodsfor using these technologies How would the new airborneLiDAR mapping approaches be used with satellite and plotmeasurement methods to estimate aboveground carbon stocksAnd what satellite monitoring technology is available today tosupport such a monitoring approach This paper presents aconceptual framework for integrating new satellite monitoringtechnology with infrequent aircraft-based mapping and amodest number of field plots to set baselines and to monitorcarbon stocks losses and recovery in forests at the sub-nationallevel The primary focus here is on tropical forests but many ofthe concepts presented are readily transferable to other foresttypes

2 Overview of technical approach

An approach to support Tier IIIII carbon stock estimationinvolves two tasks develop a spatially resolved regionalstate-of-the-forest carbon baseline and monitor forest coverchange to estimate losses or gains in forest carbon stocksThe first task can be described in four detailed steps(a) satellite analysis of forest cover and condition (b) stratifiedsampling of forest canopy three-dimensional (3D) structureusing airborne LiDAR (c) conversion of LiDAR structuraldata to aboveground carbon density estimates using newLiDAR allometrics along with a limited number of field plotsand (d) integration of the satellite map with the airborneLiDAR data to set a regional high-resolution baseline carbonstock estimate The second major task involves long-termmonitoring of carbon losses and gains using satellite data andmodels This task involves at least three steps (i) continuousmonitoring of forest cover losses and gains using satelliteimagery (ii) conversion of mapped losses in forest coverand degradation to carbon emissions and (iii) modeling ofcarbon gains in intact forest and secondary regrowth Thesemajor tasks and the steps required to achieve them arediscussed and illustrated in the following sections The laststep on modeling carbon gains is only partially covered heresince modeling of carbon accumulation is demonstrated andsynthesized elsewhere (eg Hirsch et al 2004 IPCC 2006)

3 State-of-the-forest carbon baseline

There is much debate in the REDD monitoring arena regardingthe development of baseline carbon assessments (reviewed byOlander et al 2008) This debate is largely political withmost parties focusing on the time period in which to setthe baseline carbon stock Nonetheless forest managementor conservation organizations will need to create a currentbaseline of forest carbon stocks from which to monitor changeinto the future Since no satellites can directly measure forestcarbon a baseline assessment must be developed using acombination of maps that include regional forest cover andcondition plus estimates of forest carbon densities Anillustrated flow diagram depicting the outputs of each step isshown in figure 1 with each step noted in the diagram

2

Environ Res Lett 4 (2009) 034009 G P Asner

Figure 1 Illustrated flow diagram depicting an approach to (a) assess forest cover and disturbance with satellite imagery (b) stratify theforested landscape (c) sample the carbon densities of each major forest type using airborne light detection and ranging (LiDAR) and(d) estimate the baseline carbon stocks throughout the region of interest

31 Mapping forest type cover and condition

A first step to developing a spatially-detailed forest carbonassessment involves the acquisition andor development ofa map depicting forest type deforestation and disturbance(figure 1(a)) In recent years many maps have becomeavailable showing forest type at national and sub-nationallevels These maps typically provide categories such as lsquoterrafirmersquo lsquobamboo-dominatedrsquo lsquoopen woodlandrsquo and othersconsidered appropriate to the region of interest and to themap scale Some of these maps are derived from satellitedata whereas many are created from a combination of aerialphotography and field surveys A map of forest type will aidin partitioning the landscape into categories for subsequentairborne and field measurements as described later Themore detailed the map the more finely the landscape can bepartitioned for biomass sampling

Although laborious to create maps of forest typerarely require complete redevelopment unless a major shift

in forest cover occurs Most maps of forest type areavailable from regional and national government sources non-government organizations or international sources such as theWorld Conservation Monitoring Centre of the United Nations(httpwwwunep-wcmcorgforestglobal maphtm) and theEuropean Commissionrsquos Global Land Cover 2000 (httpbiovaljrceceuropaeuproductsglc2000productsphp) Anexample map developed by the Brazilian government is shownin figure 2

Maps of deforestation and disturbance are more difficultto obtain but are essential to developing an initial regionalcarbon stock estimate Various satellite techniques areavailable with spatial resolutions of lt10 to 1000 m or more(reviewed by DeFries et al 2005 Achard et al 2007 GOFC-GOLD 2009) For regional to national level mapping offorest cover it is advantageous to use moderate-resolutionsatellites such as Landsat (30 m) and SPOT (10ndash20 m)because these systems resolve forest clearings down to about

3

Environ Res Lett 4 (2009) 034009 G P Asner

Figure 1 Illustrated flow diagram depicting an approach to (a) assess forest cover and disturbance with satellite imagery (b) stratify theforested landscape (c) sample the carbon densities of each major forest type using airborne light detection and ranging (LiDAR) and(d) estimate the baseline carbon stocks throughout the region of interest

31 Mapping forest type cover and condition

A first step to developing a spatially-detailed forest carbonassessment involves the acquisition andor development ofa map depicting forest type deforestation and disturbance(figure 1(a)) In recent years many maps have becomeavailable showing forest type at national and sub-nationallevels These maps typically provide categories such as lsquoterrafirmersquo lsquobamboo-dominatedrsquo lsquoopen woodlandrsquo and othersconsidered appropriate to the region of interest and to themap scale Some of these maps are derived from satellitedata whereas many are created from a combination of aerialphotography and field surveys A map of forest type will aidin partitioning the landscape into categories for subsequentairborne and field measurements as described later Themore detailed the map the more finely the landscape can bepartitioned for biomass sampling

Although laborious to create maps of forest typerarely require complete redevelopment unless a major shift

in forest cover occurs Most maps of forest type areavailable from regional and national government sources non-government organizations or international sources such as theWorld Conservation Monitoring Centre of the United Nations(httpwwwunep-wcmcorgforestglobal maphtm) and theEuropean Commissionrsquos Global Land Cover 2000 (httpbiovaljrceceuropaeuproductsglc2000productsphp) Anexample map developed by the Brazilian government is shownin figure 2

Maps of deforestation and disturbance are more difficultto obtain but are essential to developing an initial regionalcarbon stock estimate Various satellite techniques areavailable with spatial resolutions of lt10 to 1000 m or more(reviewed by DeFries et al 2005 Achard et al 2007 GOFC-GOLD 2009) For regional to national level mapping offorest cover it is advantageous to use moderate-resolutionsatellites such as Landsat (30 m) and SPOT (10ndash20 m)because these systems resolve forest clearings down to about

3

Environ Res Lett 4 (2009) 034009 G P Asner

Figure 2 Example map of forest and other ecosystem types in the Brazilian Legal Amazon (IBGE 1993)

01 ha resolution Moreover newer analytical approachesuse the same moderate-resolution imagery to resolve muchfiner forest disturbances caused by selective logging small-scale clearing and other land-use processes (Souza et al2005) The Carnegie Landsat Analysis System (CLAS) isone such system developed to map forest disturbances suchas selective logging (Asner et al 2005 2006) RecentlyCLASlite was created to provide automated mapping offorest cover deforestation and disturbance using a widerange of satellite sensors (httpclasliteciwedu) CLASliteis easy to use well tested on standard desktop computersfree of charge and it is currently being disseminated togovernment and non-government agencies in South AmericaCLASlite can be used with a variety of freely availablesatellite imagery such as Landsat ASTER or MODIS or withcommercial SPOT imagery For example Landsat 5 and 7imagery is now completely free of charge from the UnitedStates Geological Survey (httplandsatusgsgov) and fromthe Brazilian National Institute for Space Research (httpwwwdgiinpebr) This is further discussed in the subsequentsection on costs

Typical image results from the CLASlite step are shownusing Landsat 7 imagery over an 8000 km2 sample area inthe eastern Brazilian Amazon (figure 3) This analysis wasfully automated taking 20 min on a standard desktop computerand with no technical training (described in detail by Asneret al 2009b) Forest cover deforested areas and forestdisturbancemdashhere caused primarily by selective loggingmdasharereadily observed in the CLASlite imagery It is important tonote that each CLASlite image pixel reports a percentage offorest canopy dead vegetation and bare soil (0ndash100 for eachof these surface types) These percentages precisely indicatethe amount of forest damage per pixel (Asner et al 2004) Thisis critical to delineating areas of wholesale clearing (gt80

forest cover loss in a pixel) versus forest degradation associatedwith the partial clearing of a pixel

32 Forest stratification and airborne LiDAR sampling

Using the maps derived from CLASlite combined with mapsof forest type the region can be stratified into LiDAR mappingunits (figure 1(b)) These mapping units will vary in size andregional importance providing the user with an opportunityto prioritize them based on a variety of criteria such aspotential biomass biodiversity hotspots risk of deforestationor degradation and other factors The prioritization of LiDARmapping units will be highly dependent upon the specificinterests and needs of the user For carbon assessmentpurposes the map could be prioritized to cover areas ofintact forest likely harboring the highest biomass as well asareas of recent selective logging or secondary regrowth thatwill undergo carbon accumulation in the years following thebaseline assessment (figure 4)

Airborne LiDAR mapping involves flight planning sensorcalibration data acquisition post-flight image calibration and3D data generation phases (reviewed by Lefsky et al 2002b)There is a rapidly growing industry for commercial LiDARmapping available throughout the world that can accommodatethese tasks In fact one market analysis reports that between2005 and 2008 there was a 75 increase in the number ofairborne LiDAR systems and a 53 increase in the volume ofLiDAR operators worldwide (httpwwwcaryandassociatescommarketing toolsmarket reportsrpt4html) As a resultLiDAR operators can provide data acquisition services nearlyanywhere in the world today Research and demonstrationsystems such as the Carnegie Airborne Observatory (CAOhttpcaociwedu) integrate LiDAR with other imagingsensors to facilitate mapping of both forest composition and3D structure simultaneously (Asner et al 2007) Independent

4

Environ Res Lett 4 (2009) 034009 G P Asner

Figure 3 Example CLASlite image results for an 8000 km2 regionof the eastern Brazilian Amazon Top panel shows location of thisarea in South America middle panel shows the entire region bottompanel shows a 1500 km2 zoom image CLASlite imagery provides apercentage (0ndash100) canopy cover of liveforest vegetation dead orsenescent vegetation and bare soils Recently deforested andorburned areas are obvious with high bare soil fractions (redsmagenta) forest regrowth is apparent as large fractions of greencanopy intertwined with exposed soils (yellows) selectively loggedforest (blue patterns in forest) and intact forest (nearly solid green)

Figure 4 Stratification of the forested region using vegetation mapfrom figure 2 and CLASlite output from figure 3 White linesindicate the general delineation between upland (terra firme) andlow-lyingdrainage areas

of whether the source is commercial government or researchairborne LiDAR provides an opportunity to map the 3Dstructure of forest canopies over more than 5000 ha per dayat 10 m spatial resolution

Figure 5 Example relationships between airborne LiDARmeasurements of mean canopy vertical height profiles (MCH) andfield-based estimates of aboveground carbon density (Mg C haminus1)The blue line indicates the average estimated relationship fortemperate and boreal forests (standard error = 4 Mg C haminus1) thegreen line shows the average relationship for lowland to montanetropical forests (standard error = 20 Mg C haminus1) These regressionsare derived from Lefsky et al (2002a) and Asner et al (2009a) Thered line shows an example adjustment made to the tropical forestrelationship (green line) using 10 randomly selected plots from a newforest location This illustrates the ability to rapidly adjust LiDARmetrics for a new region using a modest number of field plots

33 Conversion of LiDAR measurements to abovegroundcarbon density

Following the airborne LiDAR sampling of the stratified forestmap (figure 4) the mapped canopy 3D structural informationderived from LiDAR can be used to estimate abovegroundcarbon density (ACD units of Mg carbon haminus1) This steprequires a set of lsquoLiDAR metricsrsquo Ground-based allometricequations have been used for years to estimate ACD fromfield measurements of tree diameters height and wood density(Brown and Lugo 1984 Chave et al 2005) In contrastLiDAR metrics are new relating airborne measurements offorest 3D structure including height and the vertical canopyprofile to field-based estimates of ACD Whereas the basicLiDAR data are typically collected at a spatial resolution ofabout 1 m LiDAR metrics are usually averaged at the plotlevel then regressed against plot-level ACD estimates (eg 01ndash10 ha) This approach parallels traditional field studies that usemanual measurements of the diameter and height of each treeto estimate plot-level biomass

There are multiple ways to relate LiDAR measurementsto aboveground carbon density Lefsky et al (2002a) derivedLiDAR metrics for three temperate and boreal forest biomesyielding a general equation

ACD (Mg C haminus1) = 0378 times MCH2

r 2 = 084 p lt 0001 (1)

where MCH is the mean vertical canopy profile height derivedfrom the airborne LiDAR (figure 5) This is just one of severalcanopy vertical profile metrics developed from LiDAR shownhere simply as an illustration For a detailed presentation ofthis and other metrics see Lefsky et al (2002a 2002b)

5

Environ Res Lett 4 (2009) 034009 G P Asner

Developing generic LiDAR metrics for tropical forests hasbeen more challenging Although equations were developedby Drake et al (2003) for moist and wet tropical forests theyemployed a LiDAR technology unique to NASArsquos researchprogram that is they used technology that is not widely used inthe commercial sector Commercial LiDAR will be needed tosupport carbon mapping for REDD To address this gap Asneret al (2009a) derived equations relating airborne commercial-grade LiDAR metrics to carbon densities for lowland tomontane tropical forest

ACD (Mg C haminus1) = 0844 times MCH2

r 2 = 080 p lt 001 (2)

Although equations (1) and (2) are shown as comparableexamples it is notable that equation (2) scales carbon densitiesfrom LiDAR measurements at a rate that is 22 times greaterthan for temperate and boreal forests from Lefsky et al (2002a)This is caused by forest-dependent differences in the wooddensity stem diameter and the number of trees per unit area(eg boreal conifer versus lowland tropical) As a result fieldplots are needed to adjust the slope of the relationship betweenMCH2 and aboveground carbon density for different foresttypes Field plots can be set up anywhere within the LiDARcoverage area but should span a range of apparent biomasslevels from low to high A relatively small number of plotsspanning the biomass levels found within each forest type (egfewer than 10) can increase the accuracy of the LiDAR-to-biomass conversion equations by up to 40 (figure 5) Thisissue remains an intense focus of research but it is becomingclear that the traditional notion that hundreds of plots arerequired per vegetation type is not supported by the currentLiDAR literature thereby decreasing the need for very largeforest inventory measurement programs The LiDAR becomesthe regional sampling approach whereas field plots help tocalibrate the LiDAR to forest conditions A major advantageof this approach is that LiDAR can map thousands of hectaresof forest per day whereas field plots cannot achieve the samegeographic coverage

Output from this LiDAR mapping step includes estimatesof aboveground carbon density at 01ndash10 ha resolution alongwith error values derived from uncertainty in the LiDAR-to-biomass conversion equations employed An example 5000 hamap of forest aboveground carbon densities at 01 ha resolutionis shown in figure 6 for a gradient of lowland to montanerain forest types on Hawaii Island The resulting distributionof aboveground carbon densities is shown in the lower leftpanel Current estimated uncertainties in the relationshipbetween airborne LiDAR forest structure and ACD range fromabout 10ndash15 for temperate and boreal forests (Lefsky et al2002a Popescu et al 2004 and many others) to 8ndash25 intropical forests (Drake et al 2003 Asner et al 2009a) It isimportant to note that these uncertainties are similar to the errorranges reported for plot-based estimates of ACD For exampleBrown and Lugo (1984) estimate a 20 uncertainty in stand-level biomass estimates derived from forest volumes Chaveet al (2005) estimate standard errors in stand biomass of 12ndash20 and Keller et al (2001) reported up to 50 uncertaintyin a very detailed large-area biomass study in the central

Brazilian Amazon In short the LiDAR-to-biomass estimationapproaches can be as certain as the measurement-to-biomass(allometric) approaches used at the plot level

34 Integration of LiDAR and satellite data

Following LiDAR survey the statistical distribution of thesampled carbon densities can be applied to a satellite imageto derive a map of aboveground carbon stocks along withuncertainty bounds A simulated regional image of forestcarbon stock is shown in figure 7 The map depicts theregional distribution of forest cover deforested lands and forestdisturbed through selective logging There are also areas ofsecondary forest regrowth The distributions of carbon densityare shown in the bottom panels and color-coded to match thedetailed forest stratification

The baseline aboveground carbon map simulated infigure 7 covers an area of 8000 km2 in the Brazilian AmazonThe amount of airborne LiDAR mapping required to applydistributions of carbon stocks to the satellite image dependsupon the relative proportions of each forest stratum theintensity and diversity of land-use activities within each foreststratum and the relative emphasis placed on these areas forcarbon accounting The original forest type and forest coverstratification process illustrated in figures 2ndash4 determines thepartitioning of LiDAR coverage Within a forest stratumLiDAR coverage requirements are dictated by the need toachieve a robust statistical distribution of estimated carbondensities In the central Amazon Keller et al (2001) showedthat 13 of a terra firme forest required surveying to achievea statistical representation of a 392 ha area Estimates from a5000 ha tropical rain forest suggest that less than a 1 LiDARsampling of each forest stratum (eg lowland intact forest bogforest montane forest etc) yields carbon density distributionsthat are statistically similar to those of each forest type (derivedfrom Asner et al 2009a)

Following aboveground carbon mapping it is possibleto estimate belowground carbon densities as well as carbondensities for coarse woody debris and other surface organiclitter The IPCC (2006) provides scaling factors which canbe applied to the image of aboveground carbon densities Forexample belowground carbon is estimated at 037 times theaboveground carbon density for tropical rain forest (Mokanyet al 2006) providing support for Tier II level analysisThe modest number of field plots deployed in the LiDARcalibration step described above can also be used to improveestimates of belowground and surface litter carbon densities(IPCC 2006) if time and funding permit to aim for Tier IIItype accuracies

4 Monitoring losses and gains in forest carbon

Following a baseline carbon mapping assessment two ongoingmeasurement efforts are required to track carbon losses andgains into the future Carbon losses can be monitored throughthe use of satellite technology Forest clearing and degradationcan be mapped at high resolution on an annual basis ormore often depending upon needs and availability of satellite

6

Environ Res Lett 4 (2009) 034009 G P Asner

Figure 6 Aboveground biomass map (Mg C haminus1) at 01 ha resolution derived from airborne LiDAR This 4000 ha image was collectedalong the eastern slopes of Mauna Kea Volcano Hawaii Island The estimated distribution of aboveground biomass in largeold growthsmallerintact and opendisturbed rain forests are shown in the lower left panel

imagery (DeFries et al 2005 GOFC-GOLD 2009) Forexample deforestation and degradation maps derived fromCLASlite can be applied to the baseline carbon stock map toestimate carbon emissions to the atmosphere (figure 8) Pitfallshere include uncertainty in converting satellite measurementsof forest degradation to carbon loss although the IPCC (2006)provides guidelines on ways to do this accounting

A conservative quantitative approach to monitoringcarbon emissions following a baseline assessment is tomultiply the fractional canopy loss within each CLASlitepixel by the carbon density value derived from the baselinecarbon map For example a pixel might transition from 100forest cover (CLASlite baseline) with 50 Mg Cpixel (LiDARbaseline) to 50 forest cover following selective logging

resulting in an estimated 25 Mg C emitted to the atmosphereAccuracy can be increased via rapid field assessment of actualcarbon losses

Long-term assessment of carbon gains requires that theregion be continuously monitored for the onset of secondaryregrowth following clearing multi-temporal satellite imagingcan greatly enhance this capability A second requirement isto estimate the accumulation of carbon stock on a per-pixelbasis Many models and equations are available to supportthis approach and these models may take into considerationforest and soil type as well as climate (Angelson et al 2009)The IPCC (2006) provides guidelines for incrementing carbonpools The production of stored wood surface woody debrisand other litter can be tracked using either simple book-

7

Environ Res Lett 4 (2009) 034009 G P Asner

Figure 7 Simulated carbon landscape derived from CLASlite imagery taken over an 8000 km2 region of the eastern Brazilian AmazonHistograms simulate distributions of aboveground carbon density for each land cover stratum applied for this demonstration using airborneLiDAR data collected in Hawaii

keeping approaches (eg Houghton et al 2000) or detailedprocess-oriented models (eg Huang et al 2008) Independentof the modeling approach a program of continuous satellitemonitoring of clearing logging or other disturbances aswell as recovery should be considered a core activity withmodeling and field verification ongoing A modest investmentin satellite monitoring effectively decreases the requirement forlabor-intensive field inventories and it decreases reliance onforest growth models

5 Costs

The cost to implement this proposed method for high-resolution baseline assessment is rapidly decreasing Satellite

data are decreasing in cost and the major data sources arenow free of charge to end users Although satellites areexpensive to develop deploy and maintain the governmentsdoing so are making the data available on the WorldWide Web Landsat 5 and 7 and the ChinesendashBrazilCBERS-1 and 2 are good examples The next generationLandsat called the Landsat Data Continuity Mission (LDCMhttplandsatusgsgov) is being built now and will launch in2012 and the next line CBERS sensors are in development(httpwwwcbersinpebr) Other commercial operatorssuch as SPOT Image Corporation are making their dataincreasingly available although not free of charge andthey have mature plans for new SPOT-6 and 7 satellites(httpwwwspotcom) Several other sensors are also available(GOFC-GOLD 2009)

8

Environ Res Lett 4 (2009) 034009 G P Asner

Developing generic LiDAR metrics for tropical forests hasbeen more challenging Although equations were developedby Drake et al (2003) for moist and wet tropical forests theyemployed a LiDAR technology unique to NASArsquos researchprogram that is they used technology that is not widely used inthe commercial sector Commercial LiDAR will be needed tosupport carbon mapping for REDD To address this gap Asneret al (2009a) derived equations relating airborne commercial-grade LiDAR metrics to carbon densities for lowland tomontane tropical forest

ACD (Mg C haminus1) = 0844 times MCH2

r 2 = 080 p lt 001 (2)

Although equations (1) and (2) are shown as comparableexamples it is notable that equation (2) scales carbon densitiesfrom LiDAR measurements at a rate that is 22 times greaterthan for temperate and boreal forests from Lefsky et al (2002a)This is caused by forest-dependent differences in the wooddensity stem diameter and the number of trees per unit area(eg boreal conifer versus lowland tropical) As a result fieldplots are needed to adjust the slope of the relationship betweenMCH2 and aboveground carbon density for different foresttypes Field plots can be set up anywhere within the LiDARcoverage area but should span a range of apparent biomasslevels from low to high A relatively small number of plotsspanning the biomass levels found within each forest type (egfewer than 10) can increase the accuracy of the LiDAR-to-biomass conversion equations by up to 40 (figure 5) Thisissue remains an intense focus of research but it is becomingclear that the traditional notion that hundreds of plots arerequired per vegetation type is not supported by the currentLiDAR literature thereby decreasing the need for very largeforest inventory measurement programs The LiDAR becomesthe regional sampling approach whereas field plots help tocalibrate the LiDAR to forest conditions A major advantageof this approach is that LiDAR can map thousands of hectaresof forest per day whereas field plots cannot achieve the samegeographic coverage

Output from this LiDAR mapping step includes estimatesof aboveground carbon density at 01ndash10 ha resolution alongwith error values derived from uncertainty in the LiDAR-to-biomass conversion equations employed An example 5000 hamap of forest aboveground carbon densities at 01 ha resolutionis shown in figure 6 for a gradient of lowland to montanerain forest types on Hawaii Island The resulting distributionof aboveground carbon densities is shown in the lower leftpanel Current estimated uncertainties in the relationshipbetween airborne LiDAR forest structure and ACD range fromabout 10ndash15 for temperate and boreal forests (Lefsky et al2002a Popescu et al 2004 and many others) to 8ndash25 intropical forests (Drake et al 2003 Asner et al 2009a) It isimportant to note that these uncertainties are similar to the errorranges reported for plot-based estimates of ACD For exampleBrown and Lugo (1984) estimate a 20 uncertainty in stand-level biomass estimates derived from forest volumes Chaveet al (2005) estimate standard errors in stand biomass of 12ndash20 and Keller et al (2001) reported up to 50 uncertaintyin a very detailed large-area biomass study in the central

Brazilian Amazon In short the LiDAR-to-biomass estimationapproaches can be as certain as the measurement-to-biomass(allometric) approaches used at the plot level

34 Integration of LiDAR and satellite data

Following LiDAR survey the statistical distribution of thesampled carbon densities can be applied to a satellite imageto derive a map of aboveground carbon stocks along withuncertainty bounds A simulated regional image of forestcarbon stock is shown in figure 7 The map depicts theregional distribution of forest cover deforested lands and forestdisturbed through selective logging There are also areas ofsecondary forest regrowth The distributions of carbon densityare shown in the bottom panels and color-coded to match thedetailed forest stratification

The baseline aboveground carbon map simulated infigure 7 covers an area of 8000 km2 in the Brazilian AmazonThe amount of airborne LiDAR mapping required to applydistributions of carbon stocks to the satellite image dependsupon the relative proportions of each forest stratum theintensity and diversity of land-use activities within each foreststratum and the relative emphasis placed on these areas forcarbon accounting The original forest type and forest coverstratification process illustrated in figures 2ndash4 determines thepartitioning of LiDAR coverage Within a forest stratumLiDAR coverage requirements are dictated by the need toachieve a robust statistical distribution of estimated carbondensities In the central Amazon Keller et al (2001) showedthat 13 of a terra firme forest required surveying to achievea statistical representation of a 392 ha area Estimates from a5000 ha tropical rain forest suggest that less than a 1 LiDARsampling of each forest stratum (eg lowland intact forest bogforest montane forest etc) yields carbon density distributionsthat are statistically similar to those of each forest type (derivedfrom Asner et al 2009a)

Following aboveground carbon mapping it is possibleto estimate belowground carbon densities as well as carbondensities for coarse woody debris and other surface organiclitter The IPCC (2006) provides scaling factors which canbe applied to the image of aboveground carbon densities Forexample belowground carbon is estimated at 037 times theaboveground carbon density for tropical rain forest (Mokanyet al 2006) providing support for Tier II level analysisThe modest number of field plots deployed in the LiDARcalibration step described above can also be used to improveestimates of belowground and surface litter carbon densities(IPCC 2006) if time and funding permit to aim for Tier IIItype accuracies

4 Monitoring losses and gains in forest carbon

Following a baseline carbon mapping assessment two ongoingmeasurement efforts are required to track carbon losses andgains into the future Carbon losses can be monitored throughthe use of satellite technology Forest clearing and degradationcan be mapped at high resolution on an annual basis ormore often depending upon needs and availability of satellite

6

Environ Res Lett 4 (2009) 034009 G P Asner

Figure 6 Aboveground biomass map (Mg C haminus1) at 01 ha resolution derived from airborne LiDAR This 4000 ha image was collectedalong the eastern slopes of Mauna Kea Volcano Hawaii Island The estimated distribution of aboveground biomass in largeold growthsmallerintact and opendisturbed rain forests are shown in the lower left panel

imagery (DeFries et al 2005 GOFC-GOLD 2009) Forexample deforestation and degradation maps derived fromCLASlite can be applied to the baseline carbon stock map toestimate carbon emissions to the atmosphere (figure 8) Pitfallshere include uncertainty in converting satellite measurementsof forest degradation to carbon loss although the IPCC (2006)provides guidelines on ways to do this accounting

A conservative quantitative approach to monitoringcarbon emissions following a baseline assessment is tomultiply the fractional canopy loss within each CLASlitepixel by the carbon density value derived from the baselinecarbon map For example a pixel might transition from 100forest cover (CLASlite baseline) with 50 Mg Cpixel (LiDARbaseline) to 50 forest cover following selective logging

resulting in an estimated 25 Mg C emitted to the atmosphereAccuracy can be increased via rapid field assessment of actualcarbon losses

Long-term assessment of carbon gains requires that theregion be continuously monitored for the onset of secondaryregrowth following clearing multi-temporal satellite imagingcan greatly enhance this capability A second requirement isto estimate the accumulation of carbon stock on a per-pixelbasis Many models and equations are available to supportthis approach and these models may take into considerationforest and soil type as well as climate (Angelson et al 2009)The IPCC (2006) provides guidelines for incrementing carbonpools The production of stored wood surface woody debrisand other litter can be tracked using either simple book-

7

Environ Res Lett 4 (2009) 034009 G P Asner

Figure 7 Simulated carbon landscape derived from CLASlite imagery taken over an 8000 km2 region of the eastern Brazilian AmazonHistograms simulate distributions of aboveground carbon density for each land cover stratum applied for this demonstration using airborneLiDAR data collected in Hawaii

keeping approaches (eg Houghton et al 2000) or detailedprocess-oriented models (eg Huang et al 2008) Independentof the modeling approach a program of continuous satellitemonitoring of clearing logging or other disturbances aswell as recovery should be considered a core activity withmodeling and field verification ongoing A modest investmentin satellite monitoring effectively decreases the requirement forlabor-intensive field inventories and it decreases reliance onforest growth models

5 Costs

The cost to implement this proposed method for high-resolution baseline assessment is rapidly decreasing Satellite

data are decreasing in cost and the major data sources arenow free of charge to end users Although satellites areexpensive to develop deploy and maintain the governmentsdoing so are making the data available on the WorldWide Web Landsat 5 and 7 and the ChinesendashBrazilCBERS-1 and 2 are good examples The next generationLandsat called the Landsat Data Continuity Mission (LDCMhttplandsatusgsgov) is being built now and will launch in2012 and the next line CBERS sensors are in development(httpwwwcbersinpebr) Other commercial operatorssuch as SPOT Image Corporation are making their dataincreasingly available although not free of charge andthey have mature plans for new SPOT-6 and 7 satellites(httpwwwspotcom) Several other sensors are also available(GOFC-GOLD 2009)

8

Environ Res Lett 4 (2009) 034009 G P Asner

Figure 6 Aboveground biomass map (Mg C haminus1) at 01 ha resolution derived from airborne LiDAR This 4000 ha image was collectedalong the eastern slopes of Mauna Kea Volcano Hawaii Island The estimated distribution of aboveground biomass in largeold growthsmallerintact and opendisturbed rain forests are shown in the lower left panel

imagery (DeFries et al 2005 GOFC-GOLD 2009) Forexample deforestation and degradation maps derived fromCLASlite can be applied to the baseline carbon stock map toestimate carbon emissions to the atmosphere (figure 8) Pitfallshere include uncertainty in converting satellite measurementsof forest degradation to carbon loss although the IPCC (2006)provides guidelines on ways to do this accounting

A conservative quantitative approach to monitoringcarbon emissions following a baseline assessment is tomultiply the fractional canopy loss within each CLASlitepixel by the carbon density value derived from the baselinecarbon map For example a pixel might transition from 100forest cover (CLASlite baseline) with 50 Mg Cpixel (LiDARbaseline) to 50 forest cover following selective logging

resulting in an estimated 25 Mg C emitted to the atmosphereAccuracy can be increased via rapid field assessment of actualcarbon losses

Long-term assessment of carbon gains requires that theregion be continuously monitored for the onset of secondaryregrowth following clearing multi-temporal satellite imagingcan greatly enhance this capability A second requirement isto estimate the accumulation of carbon stock on a per-pixelbasis Many models and equations are available to supportthis approach and these models may take into considerationforest and soil type as well as climate (Angelson et al 2009)The IPCC (2006) provides guidelines for incrementing carbonpools The production of stored wood surface woody debrisand other litter can be tracked using either simple book-

7

Environ Res Lett 4 (2009) 034009 G P Asner

Figure 7 Simulated carbon landscape derived from CLASlite imagery taken over an 8000 km2 region of the eastern Brazilian AmazonHistograms simulate distributions of aboveground carbon density for each land cover stratum applied for this demonstration using airborneLiDAR data collected in Hawaii

keeping approaches (eg Houghton et al 2000) or detailedprocess-oriented models (eg Huang et al 2008) Independentof the modeling approach a program of continuous satellitemonitoring of clearing logging or other disturbances aswell as recovery should be considered a core activity withmodeling and field verification ongoing A modest investmentin satellite monitoring effectively decreases the requirement forlabor-intensive field inventories and it decreases reliance onforest growth models

5 Costs

The cost to implement this proposed method for high-resolution baseline assessment is rapidly decreasing Satellite

data are decreasing in cost and the major data sources arenow free of charge to end users Although satellites areexpensive to develop deploy and maintain the governmentsdoing so are making the data available on the WorldWide Web Landsat 5 and 7 and the ChinesendashBrazilCBERS-1 and 2 are good examples The next generationLandsat called the Landsat Data Continuity Mission (LDCMhttplandsatusgsgov) is being built now and will launch in2012 and the next line CBERS sensors are in development(httpwwwcbersinpebr) Other commercial operatorssuch as SPOT Image Corporation are making their dataincreasingly available although not free of charge andthey have mature plans for new SPOT-6 and 7 satellites(httpwwwspotcom) Several other sensors are also available(GOFC-GOLD 2009)

8

Environ Res Lett 4 (2009) 034009 G P Asner

Figure 7 Simulated carbon landscape derived from CLASlite imagery taken over an 8000 km2 region of the eastern Brazilian AmazonHistograms simulate distributions of aboveground carbon density for each land cover stratum applied for this demonstration using airborneLiDAR data collected in Hawaii

keeping approaches (eg Houghton et al 2000) or detailedprocess-oriented models (eg Huang et al 2008) Independentof the modeling approach a program of continuous satellitemonitoring of clearing logging or other disturbances aswell as recovery should be considered a core activity withmodeling and field verification ongoing A modest investmentin satellite monitoring effectively decreases the requirement forlabor-intensive field inventories and it decreases reliance onforest growth models

5 Costs

The cost to implement this proposed method for high-resolution baseline assessment is rapidly decreasing Satellite

data are decreasing in cost and the major data sources arenow free of charge to end users Although satellites areexpensive to develop deploy and maintain the governmentsdoing so are making the data available on the WorldWide Web Landsat 5 and 7 and the ChinesendashBrazilCBERS-1 and 2 are good examples The next generationLandsat called the Landsat Data Continuity Mission (LDCMhttplandsatusgsgov) is being built now and will launch in2012 and the next line CBERS sensors are in development(httpwwwcbersinpebr) Other commercial operatorssuch as SPOT Image Corporation are making their dataincreasingly available although not free of charge andthey have mature plans for new SPOT-6 and 7 satellites(httpwwwspotcom) Several other sensors are also available(GOFC-GOLD 2009)

8

Environ Res Lett 4 (2009) 034009 G P Asner

Figure 8 Long-term monitoring using annual CLASlite mapping of deforestation and degradation from 1999 to 2002 for a 8000 km2 regionin the eastern Brazilian Amazon A zoom image (upper left black) shows detail of forest degradation color-coded similarly to that of thedeforestation map revealing small-scale sub-pixel forest canopy loss from selective logging These maps can be used to estimate carbonemissions from the baseline map in figure 7

The cost to analyze the satellite data for forest coverdeforestation and degradation is also rapidly diminishingThe Carnegie Institution is making CLASlite availablefor free to governments and NGOs throughout the Ama-zon region (httplandsatgsfcnasagovnewsnews-archivenews 0182html) The Brazilian Space Research Institute(INPE httpwwwinpebr) and the Brazilian organizationIMAZON (httpimazonorgbr) are making their forest coveranalysis software available as well There is a small cost ofinitial training and the need for an internet connection but themethods are now highly automated and thus can be operatedby an entry-level technician

LiDAR is a powerful airborne imaging technology thatlike aerial photography in the 1970sndash1980s is rapidlyexpanding throughout the world for use across a range ofenvironmental sectors Airborne LiDAR is now widely usedfor mapping infrastructure including power lines dams andcities as well as terrain and forestry As such the commercialbase is very strong and it is supporting both an explosion ofLiDAR technologies into many countries (and all continents)and a teaching of the next generation of remote sensingspecialists Such deep commercial support is unusual in remotesensing and it speaks for the efficacy of the technology Thereare now many airborne LiDAR mapping companies spread

across the Americas Europe Africa Asia Australia and thePacific (httpwwwairbornelasermappingcom)

Today the Carnegie Airborne Observatory can operate itsLiDAR process the data and provide maps of forest structureat a cost of less than $020 per hectare not including the costof ferrying the aircraft and sensors to the theater of operationThis suggests that government or commercial LiDAR groupscan operate within their own countries at similar or lower-cost levels These costs are yet decreasing further with anew generation of fully automated LiDAR processing methods(httpwwwncalmorg) The methods could be disseminatedin a manner parallel to that of the automated satellite mappingapproaches that are supporting REDD baseline assessmentsand monitoring I believe these advances in automationwill decrease the dependence on specialists just as wehave seen with satellite deforestation mapping resulting inLiDAR mapping costs of as little as $005 per hectareEven beyond airborne LiDAR new government-supportedspaceborne LiDAR technology intended to measure foreststructure worldwide may become available in late 2015(httpdesdynijplnasagov) Spaceborne LiDAR will providethe much added benefit of allowing continual reassessment ofcarbon densities in a spatially explicit and contiguous manner

9

Environ Res Lett 4 (2009) 034009 G P Asner

The costs of well-maintained field inventory plots remainshighly variable but likely not any lower than that of LiDARTake into account the fact that airborne LiDAR can cover farmore forest in a day (gt5000 ha) than can ever be measured onthe ground by the same number of personnel combined withthe fact that field inventory methods of aboveground biomassappear no more accurate than LiDAR-based approaches(discussed earlier) and it seems that the methods are becominginterchangeable or at least highly compatible Field plots arestill needed to calibrate and validate LiDAR measurements tobiomass but the number of plots is greatly reduced since thereis less need to sample the forest with plots but rather focus theplots on calibrating the LiDAR The LiDAR then becomes theforest sampling (or mapping) step in place of the plots

Forest carbon monitoring at IPCC Tiers IIIII is currentlyconsidered technically challenging potentially labor-intensiveand thus an expensive undertaking Here I have proposedthat a combination of free satellite monitoring technology withinfrequent aircraft-based mapping can provide baseline carbonestimates and can improve the monitoring of carbon stockslosses and recovery in forests These tools are available nowand can be implemented in any region of the world makingrapid Tier IIIII mapping a reality

Acknowledgments

I thank S Brown G Paez-Acosta G Powell R Martin and twoanonymous reviewers for critical comments that improved thispaper Thanks to M Lefsky for additional guidance CLASlitetraining and capacity building is ongoing with government andnon-government agencies of the Andes-Amazon region Thiswork is supported by the Gordon and Betty Moore Foundation

References

Achard F Defries R Eva H Hansen M Mayaux P andStibig H-J 2007 Pan-tropical monitoring of deforestationEnviron Res Lett 2 045022

Angelson A Brown S Loisel C Peskett L Streck C andZarin D 2009 Reducing Emissions from Deforestation andForest Degradation (REDD) An Options Assessment ReportPrepared for the Government of Norway by the MeridianInstitute March 2009 Available for download at httpwwwREDD-OARorg ISBN 978-0-615-28518-4

Asner G P Broadbent E N Oliveira P J C Keller M Knapp D E andSilva J N M 2006 Condition and fate of logged forests in theBrazilian Amazon Proc Natl Acad Sci USA 103 12947ndash50

Asner G P Hughes R F Varga T A Knapp D E andKennedy-Bowdoin T 2009a Environmental and biotic controlsover aboveground biomass throughout a rainforest Ecosystems12 261ndash78

Asner G P Keller M Pereira R Zweede J C and Silva J N M 2004Canopy damage and recovery after selective logging inAmazonia field and satellite studies Ecol Appl 14 S280ndash98

Asner G P Knapp D E Balaji A and Paez-Acosta G 2009bAutomated mapping of tropical deforestation and degradationCLASlite J Appl Remote Sens 3 033543

Asner G P Knapp D E Broadbent E N Oliveira P J C Keller M andSilva J N M 2005 Selective logging in the Brazilian AmazonScience 310 480ndash2

Asner G P Knapp D E Jones M O Kennedy-Bowdoin TMartin R E Boardman J and Field C B 2007 Carnegie airborne

observatory in-flight fusion of hyperspectral imaging andwaveform light detection and ranging for three-dimensionalstudies of ecosystems J Appl Remote Sens 1 013536

Brown I F Martinelli L A Thomas W W Moreira M ZFerreira C A and Victoria R A 1995 Uncertainty in the biomassof Amazonian forests an example from Rondonia Brazil ForestEcol Manag 75 175ndash89

Brown S and Lugo A E 1984 Biomass of tropical forests a newestimate based on forest volumes Science 223 1290ndash3

Brown S Pearson T Slaymaker D Ambagis S Moore NNovelo D and Sabido W 2005 Creating a virtual tropical forestfrom three-dimensional aerial imagery to estimate carbon stocksEcol Appl 15 1083ndash95

Cairns M A Brown S Helmer E H and Baumgardner G A 1997 Rootbiomass allocation in the worldrsquos upland forests Oecologia111 1ndash11

Chambers J Q Asner G P Morton D C Anderson L O Saatchi S SEspirito-Santo D B Palace M and Souza C Jr 2007 Regionalecosystem structure and function ecological insights fromremote sensing of tropical forests Trends Ecol Evol 22 414ndash23

Chave J et al 2005 Tree allometry and improved estimation of carbonstocks and balance in tropical forests Oecologia 145 87ndash99

DeFries R Asner G P Achard F Justice C Laporte N Price KSmall C and Townshend J 2005 Monitoring tropicaldeforestation for emerging carbon markets TropicalDeforestation and Climate Change ed P Moutinho andS Schwartzman (Belem Amazon Institute for EnvironmentalResearch)

Drake J B Knox R G Dubayah R O Clark D B Condit R Blair J Band Hofton M 2003 Above-ground biomass estimation in closedcanopy neotropical forests using lidar remote sensing factorsaffecting the generality of relationships Glob Ecol Biogeogr12 147ndash59

Gibbs H K Brown S Niles J O and Foley J A 2007 Monitoring andestimating tropical forest carbon stocks making REDD a realityEnviron Res Lett 2 045023

GOFC-GOLD 2009 Reducing greenhouse gas emissions fromdeforestation and degradation in developing countries asourcebook of methods and procedures for monitoringmeasuring and reporting GOFC-GOLD Report VersionCOP14-2 GOFC-GOLD Project Office Natural ResourcesCanada Alberta Canada

Hirsch A I Little W S Houghton R A Scott N A and White J D2004 The net carbon flux due to deforestation and forestre-growth in the Brazilian Amazon analysis using aprocess-based model Glob Change Biol 10 908ndash24

Houghton R A Skole D L Nobre C A Hackler J L Lawrence K Tand Chomentowski W H 2000 Annual fluxes of carbon fromdeforestation and regrowth in the Brazilian Amazon Science403 301ndash4

Huang M Asner G P Keller M and Berry J A 2008 An ecosystemmodel for tropical forest disturbance and selective loggingJ Geophys Res 113 G01002

IBGE 1993 Mapa de Vegetacao do Brasil (Vegetation Map of Brazil)2nd edn Instituto Brasiliero de Geografia e Estatistica(Brazilian Institute of Geography and Statistics)

IPCC 2006 IPCC Guidelines for National Greenhouse GasInventories ed H S Eggleston L Buendia K Miwa T Ngara andK Tanabe Prepared by the National Greenhouse GasInventories Programme IGES Japan

Keller M Asner G P Silva N and Palace M 2004 Sustainability ofselective logging of upland forests in the Brazilian Amazoncarbon budgets and remote sensing as tools for evaluatinglogging effects Working Forests in the Neotropics ConservationThrough Sustainable Management ed D J ZarinJ R R Alavalapati F E Putz and M Schmink (New YorkColumbia University Press) pp 41ndash63

Keller M Palace M and Hurtt G 2001 Biomass estimation in theTapajos National Forest Brazil examination of sampling andallometric uncertainties Forest Ecol Manag 154 371ndash82

10

Environ Res Lett 4 (2009) 034009 G P Asner

Lefsky M A Cohen W B Harding D J Parker G G Acker S A andGower S T 2002a LiDAR remote sensing of above-groundbiomass in three biomes Glob Ecol Biogeogr 11 393ndash9

Lefsky M A Cohen W B Parker G G and Harding D J 2002b LiDARremote sensing for ecosystem studies BioScience 52 19ndash30

Mokany K Raison J R and Prokushkin A S 2006 Critical analysis ofrootshoot ratios in terrestrial biomes Glob Change Biol12 84ndash96

Olander L P Gibbs H K Steininger M K Swensen J J andMurray B C 2008 Reference scenarios for deforestation andforest degradation in support of REDD a review of data andmethods Environ Res Lett 3 025011

Phillips O L et al 1998 Changes in the carbon balance of tropicalforests evidence from long-term plots Science 282 439ndash42

Popescu S C Wynne R H and Scrivani J A 2004 Fusion ofsmall-footprint lidar and multispectral data to estimateplot-level volume and biomass in deciduous and pine forests inVirginia USA Forest Sci 50 551ndash65

Saatchi S S Houghton R A Dos Santos Alvala R C Soares J V andYu Y 2007 Distribution of aboveground live biomass in theAmazon basin Glob Change Biol 13 816ndash37

Souza C Roberts D A and Cochrane M A 2005 Combining spectraland spatial information to map canopy damages fromselective logging and forest fires Remote Sens Environ98 329ndash43

11