简介：Background:LeafAreaIndex(LAI)isanimportantparameterusedinmonitoringandmodelingofforestecosystems.Theaimofthisstudywastoevaluateperformanceoftheartificialneuralnetwork(ANN)modelstopredicttheLAIbycomparingtheregressionanalysismodelsastheclassicalmethodinthesepureandeven-agedCrimeanpineforeststands.Methods:OnehundredeighttemporarysampleplotswerecollectedfromCrimeanpineforeststandstoestimatestandparameters.EachsampleplotwasimagedwithhemisphericalphotographstodetecttheLAI.ThepartialcorrelationanalysiswasusedtoassesstherelationshipsbetweenthestandLAIvaluesandstandparameters,andthemultivariatelinearregressionanalysiswasusedtopredicttheLAIfromstandparameters.DifferentartificialneuralnetworkmodelscomprisingdifferentnumberofneuronandtransferfunctionsweretrainedandusedtopredicttheLAIofforeststands.Results:ThecorrelationcoefficientsbetweenLAIandstandparameters(standnumberoftrees,basalarea,thequadraticmeandiameter,standdensityandstandage)weresignificantatthelevelof0.01.Thestandage,numberoftrees,siteindex,andbasalareawereindependentparametersinthemostsuccessfulregressionmodelpredictedLAIvaluesusingstandparameters(/?;adj=0.5431).AscorrespondingmethodtopredicttheinteractionsbetweenthestandLAIvaluesandstandparameters,theneuralnetworkarchitecturebasedontheRBF4-19-1withGaussianactivationfunctioninhiddenlayerandtheidentityactivationfunctioninoutputlayerperformedbetterinpredictingLAI(SSE(12.1040),MSE(0.1223),RM5E(0.3497),AIC(0.1040),BIC(-777310)andR2(0.6392))comparedtotheotherstudiedtechniques.Conclusion:TheANNoutperformedthemultivariateregressiontechniquesinpredictingLAIfromstandparameters.TheANNmodels,developedinthisstudy,mayaidinmakingforestmanagementplanninginstudyforeststands.