The Effect Of The Precision Of Input Soil Maps On The Applicability Of Updating Soil Maps

High quality soil information is increasingly required for precise agriculture and environmental pollution management.In the past decades,huge piles of conventional soil maps various in scale and nature have been accumulated.It is a common method to utilize these maps for predicting the soil spatial distribution.Many scholars find that the precision of input soil maps will have great influence on the inference accuracy,but they don't present a quantitative relationship between the two.Therefore,a case study is conducted in a small watershed in Raffelson,Wisconsin of USA to explore the quantitative relationship between the inference accuracy and the precision of input soil maps,and to research the applicability of the predictive soil mapping approach.Main work of this paper includes:(1)Design noise simulation method and obtain soil maps with different precision.Soil maps created through manual surveys are often prone to two types of errors:misplacement of soil boundaries and inclusions.We design two methods to simulate noises and add into the original soil map,so as to obtain a series of soil maps with varying degrees of map accuracy.(2)Select training samples and build model.Based on soil maps with different precision,training samples are generated based on two training sample selection methods.Subsequently,the training samples derived from different input soil maps with diverse precision are applied in three data mining methods respectively to predict the soil spatial distribution and acquire the inference accuracy.Two training sample selection methods are area-proportion method and grading proportion method.Three data mining methods are classification and regression tree,random forest and support vector machine.(3)The quantitative relationship between inference accuracy and input soil map precision.We explore the quantitative relationship between the inference accuracy and the precision of input soil maps,research the applicability of the predictive soil mapping approach,and discuss the sensitively of inference accuracy to various training sample selection methods and data mining methods.Experimental results show that when the precision of input soil map is low,extract correct soil-environmental relationships and updating conventional soil maps is infeasible;when the precision of input soil map is high,the inferred soil maps can extract mostly expert knowledge embedded in the conventional data resource,but it is hard to updating the soil maps.In conclusion,the precision of conventional soil maps will have limitation on the applicability of this method and the effectiveness of updating conventional soil maps.