Using quantile regression forest to estimate uncertainty of digital soil mapping products

详细信息    查看全文

• 作者：Ké ; vin Vaysse^a ; ^b ; Philippe Lagacherie^b ; ^{lagache@supagro.inra.fr}
• 关键词：Digital soil mapping ; GlobalSoilMap ; Quantile regression forest ; Uncertainties ; Regional scale
• 刊名：Geoderma
• 出版年：2017
• 出版时间：1 April 2017
• 年：2017
• 卷：291
• 期：Complete
• 页码：55-64
• 全文大小：1450 K
• 卷排序：291

文摘

Two methods of uncertainty estimation of 3 GSM products were tested in South France. 100 validations sets were built by iterative sampling of 25% of the sites. Accuracy plots were proposed for validating uncertainty predictions. Quantile Regression Forests outperformed Regression Kriging in mapping uncertainty. Quantile Regression Forests is recommended in situations of sparse soil data.