Modeling water retention curves of sandy soils using neural networks

《Modeling water retention curves of sandy soils using neural networks 》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、WATERRESOURCESRESEARCH,VOL.32,NO.10,PAGES3033-3040,OCTOBER1996ModelingwaterretentioncurvesofsandysoilsusingneuralnetworksMarcelG.SchaapandWillemBoutenLandscapeandEnvironmentalResearchGroup,UniversityofAmsterdam,Amsterdam,NetherlandsAbstract.Weusedneuralnetworks(NNs)t

2、omodelthedryingwaterretentioncurve(WRC)of204sandysoilsamplesfromparticle-sizedistribution(PSD),soilorganicmattercontent(SOM),andbulkdensity(BD).Neuralnetworkscanrelatemultiplemodelinputdatatomultiplemodeloutputdatawithouttheneedofanapriorimodelconcept.Inthiswayahighp

3、erformanceblack-boxmodeliscreated,whichisveryusefulinadataexplorationefforttoassessthemaximumobtainablepredictionaccuracy.WeusedaseriesofNNmodelswithanincreasingparametrizationofinputandoutputvariablestogetabetterinterpretabilityofmodelresults.Inthefirsttwomodelsweus

4、edtheninePSDfractions,BD,andSOMasinput,whilewepredictedtheninepointsofthewaterretentioncurve.TheseNNshad12inputand9outputvariables,predictingWRCswithanaverageroot-mean-squareresidual(RMSR)watercontentof0.020cm3cm-3.Afterafewintermediarymodelswithincreasingparametriza

5、tionofPSDandWRCusing(adapted)vanGenuchten[1980]equationswearrivedatafinalNNmodelthatusedsixinputvariablestopredictthreevanGenuchten[1980]parametersresultinginaRMSRof0.024cm3cm-3.WefoundsaturatedandresidualwatercontentstobeunrelatedtothePSD,BD,orSOM,thereforethesatura

6、tedwatercontentwasconsideredtobeanindependentinputvariable,whiletheresidualwatercontentwassettozero.SensitivityanalysesshowedthatthePSDhadamajorinfluenceontheshapeoftheWRC,whileBDandSOMwerelessimportant.Onthebasisofthesesensitivityanalysesweestablishedmoreexplicitequ

7、ationsthatdemonstratedsimilarityrelationsbetweenPSDandWRCandincorporatedeffectsofSOMandBDinanempiricalway.Despitethefactthatweconsideredalargenumberoflinearandnonlinearvariantstheseequationshadaweakerperformance(RMSR:0.029cm3cm-3)thantheNNmodels,provingthemodelingpow

8、erofthattechnique.1.Introductiontheyusetheshapesimilaritybetweenpore-sizeandparticle-sizedistributionsbutalsorequireempiricalparame