交叉验证对土壤制图模型的影响研究——以亳州市微量元素预测为例
|
摘要
以亳州市谯城区农田土壤的有效铁与有效铜含量预测为例,探讨了基于10%、20%验证点的Holdout验证与留一交叉验证在数字土壤制图中的具体应用,旨在为现代土壤属性模拟提供更合理的验证依据。研究结果表明:1)Holdout验证方式在执行一次的情况下,很难准确度量建立模型的质量,需要重复多次构建训练集并构建相应的预测模型,以提升模型的预测精度;2)模型在高值区的预测精度差异较大,这些区域应是补充采样的重点区域;3)预测模型在制图过程中的稳定性不尽相同,在使用过程中应对比分析。
Based on the spatial prediction of soil available iron(AFe) and available copper(ACu) concentrations in arable soil,this study investigated the effect of cross validation on soil mapping application to reveal the limitation of traditional holdout validation and the uncertainty involved by different predictive models.The results showed that 1)Holdout validation would be inferior to leave-one-out cross validation(LOOCV) when running one time,and would generate similar results with LOOCV when running many times.Thus,it is necessary to perform many times running of models that were validated by holdout method; 2) High uncertainty was found in the areas at high AFe or ACu concentrations; and 3) Different models showed different stability in the digital soil mapping,therefore the contrastive analysis of models should be used.
Based on the spatial prediction of soil available iron(AFe) and available copper(ACu) concentrations in arable soil,this study investigated the effect of cross validation on soil mapping application to reveal the limitation of traditional holdout validation and the uncertainty involved by different predictive models.The results showed that 1)Holdout validation would be inferior to leave-one-out cross validation(LOOCV) when running one time,and would generate similar results with LOOCV when running many times.Thus,it is necessary to perform many times running of models that were validated by holdout method; 2) High uncertainty was found in the areas at high AFe or ACu concentrations; and 3) Different models showed different stability in the digital soil mapping,therefore the contrastive analysis of models should be used.
引文
[1]CAMBULE A H,ROSSITER D G,STOORVOGEL J J.A methodology for digital soil mapping in poorly-accessible areas[J].Geoderma,2013,192:341-353.
[2]ZHANG C,TANG Y,XU X,et al.Towards spatial geochemical modelling:Use of geographically weighted regression for mapping soil organic carbon contents in Ireland[J].Applied Geochemistry,2011,26(7):1239-1248.
[3]GRUNWALD S.2009.Multi–criteria characterization of recent digital soil mapping and modeling approaches[J].Geoderma,152(3-4):195-207.
[4]LIU F,ZHANG G L,SUN Y J,et al.Mapping the Three-Dimensional Distribution of Soil Organic Matter across a Subtropical Hilly Landscape[J].Soil Science Society of America Journal,2013,77(4):1241-1253.
[5]Liu F,Rossiter D G,Song X D,et al.A similarity-based method for three-dimensional prediction of soil organic matter concentration[J].Geoderma,2016,263:254-263.
[6]KEMPEN B,BRUS D J,STOORVOGEL J J.Three-dimensional mapping of soil organic matter content using soil type-specific depth functions[J].Geoderma,2011,162(1-2):107-123.
[7]ZHAO Y,XU X,HAI N,et al.Uncertainty assessment for mapping changes in soil organic matter using sparse legacy soil data and dense new-measured data in a typical black soil region of China[J].Environmental Earth Sciences,2015,73(1):197-207.
[8]LIEβM,GLASER B,HUWE B.Uncertainty in the spatial prediction of soil texture Comparison of regression tree and Random Forest models[J].Geoderma,2010,170:70-79.
[9]ZHU A X,QI F,MOORE A,et al.Prediction of soil properties using fuzzy membership values[J].Geoderma,2010,158(3-4):199-206.
[10]孙福军,雷秋良,刘颖,等.数字土壤制图技术研究进展与展望[J].土壤通报,2011,42(6):1502-1507.
[11]孙孝林,赵玉国,刘峰,等.数字土壤制图及其研究进展[J].土壤通报,2013(3):752-759.
[12]SONG X D,BRUS D J,LIU F,et al.Mapping soil organic carbon content by geographically weighted regression:A case study in the Heihe River Basin,China[J].Geoderma,2016,261:11-22.
[13]ZHAO M S,ROSSITER D G,LI D C,et al.Mapping soil organic matter in low-relief areas based on land surface diurnal temperature difference and a vegetation index[J].Ecological Indicators,2014,39:120-133.
[14]SONG X D,ZHANG G L,LIU F,et al.Modeling spatio-temporal distribution of soil moisture by deep learning-based cellular automata model[J].Journal of Arid Land,2016,8(5):734-748.
[15]KUMAR S,LAL R,LIU D.A geographically weighted regression kriging approach for mapping soil organic carbon stock[J].Geoderma,2012,189-190(6):627-634.
[16]SCHMIDT K,BEHRENS T,DAUMANN J,et al.A comparison of calibration sampling schemes at the field scale[J].Geoderma,2014,232-234:243-256.
[17]宋效东,郭卢,赵安,等.皖北亳州烟区土壤肥力定量评价[J].土壤通报,2016,47(1):8-14.
[18]张甘霖,龚子同.土壤调查实验室分析方法[M].北京:科学出版社,2012.
[19]BRUNSDON C F,FOTHERINGHAM A S,CHARLTON M E.Geographically weighted regression:a method for exploring spatial non-stationarity[J].Geographical Analysis,1996,28(4):281-298.
[20]BREIMAN L.Random forests[J].Machine Learning,2001,45(1):5-32.
[21]CORTES C,VAPNIK V.Support-vector networks[J].Machine Learning,1995,20(3):273-297.
[22]BRUS D J,KEMPEN B,GBM H.Sampling for validation of digital soil maps[J].European Journal of Soil Science,2011,62(3):394-407.
[23]SONG X D,ZHANG G L,LIU F,et al.Characterization of the spatial variability of soil available zinc at various sampling densities using grouped soil type information[J].Environmental Monitoring&Assessment,2016,188(11):600.
[24]VASENEV V I,STOORVOGEL J J,VASENEV I I,et al.How to map soil organic carbon stocks in highly urbanized regions[J]?Geoderma,2012,226-227:103-115.
[2]ZHANG C,TANG Y,XU X,et al.Towards spatial geochemical modelling:Use of geographically weighted regression for mapping soil organic carbon contents in Ireland[J].Applied Geochemistry,2011,26(7):1239-1248.
[3]GRUNWALD S.2009.Multi–criteria characterization of recent digital soil mapping and modeling approaches[J].Geoderma,152(3-4):195-207.
[4]LIU F,ZHANG G L,SUN Y J,et al.Mapping the Three-Dimensional Distribution of Soil Organic Matter across a Subtropical Hilly Landscape[J].Soil Science Society of America Journal,2013,77(4):1241-1253.
[5]Liu F,Rossiter D G,Song X D,et al.A similarity-based method for three-dimensional prediction of soil organic matter concentration[J].Geoderma,2016,263:254-263.
[6]KEMPEN B,BRUS D J,STOORVOGEL J J.Three-dimensional mapping of soil organic matter content using soil type-specific depth functions[J].Geoderma,2011,162(1-2):107-123.
[7]ZHAO Y,XU X,HAI N,et al.Uncertainty assessment for mapping changes in soil organic matter using sparse legacy soil data and dense new-measured data in a typical black soil region of China[J].Environmental Earth Sciences,2015,73(1):197-207.
[8]LIEβM,GLASER B,HUWE B.Uncertainty in the spatial prediction of soil texture Comparison of regression tree and Random Forest models[J].Geoderma,2010,170:70-79.
[9]ZHU A X,QI F,MOORE A,et al.Prediction of soil properties using fuzzy membership values[J].Geoderma,2010,158(3-4):199-206.
[10]孙福军,雷秋良,刘颖,等.数字土壤制图技术研究进展与展望[J].土壤通报,2011,42(6):1502-1507.
[11]孙孝林,赵玉国,刘峰,等.数字土壤制图及其研究进展[J].土壤通报,2013(3):752-759.
[12]SONG X D,BRUS D J,LIU F,et al.Mapping soil organic carbon content by geographically weighted regression:A case study in the Heihe River Basin,China[J].Geoderma,2016,261:11-22.
[13]ZHAO M S,ROSSITER D G,LI D C,et al.Mapping soil organic matter in low-relief areas based on land surface diurnal temperature difference and a vegetation index[J].Ecological Indicators,2014,39:120-133.
[14]SONG X D,ZHANG G L,LIU F,et al.Modeling spatio-temporal distribution of soil moisture by deep learning-based cellular automata model[J].Journal of Arid Land,2016,8(5):734-748.
[15]KUMAR S,LAL R,LIU D.A geographically weighted regression kriging approach for mapping soil organic carbon stock[J].Geoderma,2012,189-190(6):627-634.
[16]SCHMIDT K,BEHRENS T,DAUMANN J,et al.A comparison of calibration sampling schemes at the field scale[J].Geoderma,2014,232-234:243-256.
[17]宋效东,郭卢,赵安,等.皖北亳州烟区土壤肥力定量评价[J].土壤通报,2016,47(1):8-14.
[18]张甘霖,龚子同.土壤调查实验室分析方法[M].北京:科学出版社,2012.
[19]BRUNSDON C F,FOTHERINGHAM A S,CHARLTON M E.Geographically weighted regression:a method for exploring spatial non-stationarity[J].Geographical Analysis,1996,28(4):281-298.
[20]BREIMAN L.Random forests[J].Machine Learning,2001,45(1):5-32.
[21]CORTES C,VAPNIK V.Support-vector networks[J].Machine Learning,1995,20(3):273-297.
[22]BRUS D J,KEMPEN B,GBM H.Sampling for validation of digital soil maps[J].European Journal of Soil Science,2011,62(3):394-407.
[23]SONG X D,ZHANG G L,LIU F,et al.Characterization of the spatial variability of soil available zinc at various sampling densities using grouped soil type information[J].Environmental Monitoring&Assessment,2016,188(11):600.
[24]VASENEV V I,STOORVOGEL J J,VASENEV I I,et al.How to map soil organic carbon stocks in highly urbanized regions[J]?Geoderma,2012,226-227:103-115.