Mapping Detailed Soil Property Using Small Scale Soil Type Maps and Sparse Typical Samples

Soil type maps at the scale of 1..1 000 000 are used extensively to provide soil spatial distribution information for soil erosion assessment and watershed management models in China. However, the soil property maps produced through conventional direct linking method usually suffer low accuracy as well as the lack of spatial details within a soil type polygon. This paper presents an effective method to produce detailed soil property map based on representative samples which were extracted from each polygon on the 1..1 000 000 soil type map. The representative sample of each polygon is defined as the location that can represent the largest area within the polygon. The representativeness of a candidate sample is determined by calculating the soil-forming environment condition similarities between the sample and other locations. Once the representative sample of each polygon has been chosen, the property values of the existing typical samples are assigned to the corresponding representative samples with the same soil type. Finally, based on these representative samples, the detailed soil property map could be produced by using existing digital soil mapping methods. The case study in XuanCheng City, Anhui Province of China, demonstrated the proposed method could produce soil property map at a higher level of spatial details and accuracy: 1) The soil organic matter (SOM) map produced based on the representative samples can not only depict the detailed spatial distribution of SOM within a soil type polygon but also largely reduce the abrupt change of soil property at the boundaries of two adjacent polygons. 2) The Root Mean Squared Error (RMSE) of the SOM map based on the representative samples is 1.61, and it is 1.37 for the SOM map produced by using conventional direct linking method. Therefore, the proposed method is an effective approach to produce spatial detailed soil property map with higher accuracy for environment simulation models.