Abstract:

【Objective】China carried out the second state soil survey from 1979 to 1987 and the soil nutrient investigation of farmland from 2005 to 2017. Both surveys covered the whole country with a huge amount of ground soil samplings. The data generated from the two surveys have become the most detailed spatial-temporal data for soil types and quality in China. The purpose of the study was to test and to evaluate the geostatistical characteristics of the data by geostatistical testing approach, so as to provide the reference for the use of these data to characterize the temporal and spatial distribution of soil features in different disciplines. 【Method】7 testing areas were selected to represent different regions in China. Soil organic matter (SOM) contents of 0-20 cm soil layer from soil profile sampled in 1979-1987 and from plough layer sampled in 2005-2017 were extracted from the corresponding data bases. The ground sampling for soil profiles in 1979-1987 was to give priority to typical soil types firstly and secondly to keep an evenly distributed sampling as possible. 100 000 soil profiles with about 1m soil deep were finally sampled. After integrated data processing and coordinate matching, 60 000 profiles obtained coordinates. Ground sampling for soil plough layer in 2005-2017 was in grid distribution. 10 000 000 plough layer soil samples with GPS positioning coordinates have been completed. For each testing area, the data set contained two groups, about 500-1 300 SOM values from soil profile data and 50 000-250 000 values from plough layer data. The data from two time groups of each testing data set were analyzed by ordinary Kriging approach separately. 80% of the data were randomly selected as the training sample set for modeling and 20% as the verification sample set. The linear regression between the predicted value and the measured value of the validation sample was carried out. R² (coefficient of determination) and RMSE (root mean square error) were calculated to evaluate the reliability and uncertainty of the data sets in expressing the spatial distribution of the soil feature. 【Result】It was showed that the reliability of mapping SOM content by profile data of all of the 7 testing areas reached significant levels. However, the deviation between predicted values and measured values of the test data set was relatively great. The values of R² were low, between 0.223-0.380 and RMSE were relatively high. Testing results by soil plough layer data sampled in 2005-2017 showed that through large sample size and grid sampling, the reliability and prediction accuracy of mapping SOM content were improved greatly, for R² increased and RMSE decreased. The geostatistical test results of two periods with a time interval of 30 years showed that although there were some changes in the contents of soil organic matter, the overall spatial distribution of SOM content in each testing area expressed by the two data groups was similar. 【Conclusion】 The reliability and accuracy of soil maps were much better in terms of characterizing the spatial distribution of soil features, when the soil investigation was by means of a large sample size with grid sampling. It meant that the reliability and accuracy of the original large-scale soil thematic maps, such as maps of soil types, organic matter, pH value, soil nitrogen, phosphorus and potassium nutrient contents from second state soil survey, were better than maps generated by profile data, as these original large-scale soil thematic maps were derived from the large sample size with grid sampling. However, the data of 60 000 soil profiles from second state soil survey, which contained many soil features and could supply reliable soil thematic maps, were also of great importance for understanding spatial characteristics of these soil features. It has been showed that a large sample size was essential for a precise and accurate mapping of soil feature of the whole country. For mapping long-term changing or stable soil features such as soil types, texture and morphological features, it would be difficult to obtain reliable maps by a soil sample size much less than the second state soil survey. Considering the current requirements and the available data resources in China, the soil investigation in the future could be mainly focused in investigating data missing areas as well as some missing soil features for soil functions.

Key words: soil survey, geostatistics, soil organic matter, soil quality, reliability test, digital soil mapping