Background:Informationonabove-groundbiomass(AGB)isimportantformanagingforestresourceuseatlocallevels,landmanagementplanningatregionallevels,andcarbonemissionsreportingatnationalandinternationallevels.Inmanytropicaldevelopingcountries,thisinformationmaybeunreliableoratascaletoocoarseforuseatlocallevels.ThereisavitalneedtoprovideestimatesofAGBwithquantifiableuncertaintythatcanfacilitatelandusemanagementandpolicydevelopmentimprovements.Model-basedmethodsprovideanefficientframeworktoestimateAGB.Methods:UsingNationalForestInventory(NFI)datafora~1,000,000hastudyareainthemiomboecoregion,Zambia,weestimatedAGBusingpredictedcanopycover,environmentaldata,disturbancedata,andLandsat8OLIsatelliteimagery.Weassesseddifferentcombinationsofthesedatasetsusingthreemodels,asemiparametricgeneralizedadditivemodel(GAM)andtwononlinearmodels(sigmoidalandexponential),employingageneticalgorithmforvariableselectionthatminimizedrootmeansquarepredictionerror(RMSPE),calculatedthroughcross-validation.Wecomparedmodelfitstatisticstoanullmodelasabaselineestimationmethod.Usingbootstrapresamplingmethods,wecalculated95%confidenceintervalsforeachmodelandcomparedresultstoasimpleestimateofmeanAGBfromtheNFIgroundplotdata.Results:Canopycover,soilmoisture,andvegetationindiceswereconsistentlyselectedaspredictorvariables.ThesigmoidalmodelandtheGAMperformedsimilarly;forbothmodelstheRMSPEwas-36.8tonnesperhectare(i.e.,57%ofthemean).However,thesigmoidalmodelwasapproximately30%moreefficientthantheGAM,assessedusingbootstrappedvarianceestimatesrelativetoanullmodel.Afterselectingthesigmoidalmodel,weestimatedtotalAGBforthestudyareaat64,526,209tonnes(+/-477,730),withaconfidenceinterval20timesmoreprecisethanasimpledesignbasedestimate.Conclusions:OurfindingsdemonstratethatNFIdatamaybecombinedwithfreelyavailablesate
