| 摘 要： |
Over the past decades, conventional soil maps of various scales have been produced and become available in digital form. Efforts have been made to update these maps through various data mining methods to provide more detailed and precise information on soil spatial patterns. Key questions that remain unclear are: (1) How does the accuracy of legacy soil maps impact the update results; (2) Is the accuracy of inferred soil maps always improved regardless of the accuracy of the legacy maps. The current study aims to investigate these questions. Two noise production simulation methods were developed to simulate errors caused by inclusion and boundary displacement in the conventional maps, to generate a series of source maps with different accuracies and spatial patterns. Moreover, the impacts of two training sample selection methods and three data mining models on the accuracies and spatial patterns of the inferred soil maps were also evaluated. A case study was conducted in a small region, Raffelson study area, a typical ridge and valley terrain in La Crosse County, Wisconsin, USA. Results indicated that if the accuracies of the source soil maps ranged from 35% to 75%, the inferred soil map accuracies would be improved. These findings have important implications for updating conventional soil maps through data mining methods and understanding the situation in which the method is effective. |
在线留言
联系我们
所长信箱