应用rioja软件包建立有壳变形虫-环境因子转换函数

Building Transfer Functions Between Testate amoeba and Environmental Variables with‘rioja’Package

应用R语言rioja软件包的加权平均(Weighted Averaging,WA)和加权平均偏最小二乘(Weighted AveragingPartial Least Squares,WA-PLS)模型建立了长白山区泥炭藓泥炭地有壳变形虫与水位埋深(depth to water table,DWT)、pH和泥炭湿度的转换函数,为古环境定量重建奠定了基础,也提供了rioja软件包应用的实例和参考。结果表明水位埋深以WA-PLS模型最佳(预测均方根误差RMSEP为7.39 cm,R2=0.74);对于pH和泥炭湿度,WA-PLS第一分量和WA.inv都产生了最小的RMSEP和较高的R2值。pH的RMSEP为0.18,R2为0.72。泥炭湿度的RMSEP为1.95%,R2为0.62。如果泥炭剖面的有壳变形虫种类组成与本研究的训练样本集相同,水位埋深、pH和泥炭湿度可以分别以±7.39 cm、±0.18和±1.95%的平均误差进行重建。

R language,as an open source programming language and software environment,is widely used in statistics for its free availability.The‘rioja’package of R specially deals with the analysis of Quaternary science data,containing functions for constrained clustering,transfer functions and plotting stratigraphic data.Testate amoebae are a group of unicellular protists living in terrestrial habitats.Their decayed resistant and morphologically diagnostic shells（tests） allow them to be extensive used as proxy in peat based paleoenvironmental reconstruction.This study aimed to： ① Present an example of application of‘rioja’package;② build Testate amoeba-based transfer functions for quantitatively reconstructing paleoenvironment changes in Changbai Mountains with peat archive.The training set was constituted by 75 samples collected from four peatlands,Hani（42°12′50″N,126°31′05″E）,Jinchuan（42°20′47″N,126°21′35″E）,Chichi（42°03′16″N,128°03′22″E） and Yuanchi（42°01′55″N,128°25′58″E）,in Changbai Mountains,northeast China.Three factors,depth to water table（DWT）,pH and peat moisture,were selected as the target environmental variables.The models of Weighted Averaging（WA） and Weighted Averaging Partial Least Squares（WA-PLS） were used to build transfer functions.Leave-one-out was chosen as cross validation method.The results showed that the second component of WA-PLS is the best models for DWT producing a RMSEP of 7.39 and R2 of 0.74.For pH and peat moisture,both first component of WA-PLS and WA with inverse deshrinking could be regarded as the best models for they have the lowest RMSEP and relatively higher R2.The RMSEP of pH is 0.18 and R2 is 0.72,while for peat moisture RMSEP is1.95% and R2is 0.62.The performances of the transfer function were comparable with other studies in the world.DWT,pH and peat moisture could be quantitive reconstructed with the mean errors of ±7.39 cm,±0.18 and ±1.95%,respectively,if Testate amoebae assemblage of profiles was the same as the training set in this study.