Sampling Design Research On The Influence Of The Spatial Prediction Accuracy Of Soil Organic Matter Content

Soil organic matter (SOM) plays an important role in maintaining soil fertility and improving the soil quality and crop yield. Information and knowledge about the spatial variability and distribution patterns of SOM are critical for the scientific management practice recommendations and sustainable use of soil resource.So far the soil field sampling and laboratory analysis are still the most important way for obtaining spatial distribution information of SOM content. The soil sampling design method, however, is the first stage that one has to face before the field sample collection was conducted. Under the same prediction accurate request, suitable sampling design needs less sampling point while irrational sample design may cause large errors. Usually the more number of samples, the higher sample costs. Sparse sampling design may lose important spatial information of soil properties, moreover, the spatial predicting method also has great influences on the predictive mapping results. Consequently, the optimal soil sampling design and predicting methods are extremely important for decreasing soil sampling cost and improving mapping precision.Black soil region is the most important grain-producing area of China, and the strategic position of the area for maintaining food supply is critical. A total of270soil samples in Hailun county of Heilongjiang Province were collected in this study, and sequential gaussian simulation, ordinary kriging, regression kriging, four sample design methods(simple random, systematic regular, stratified random and simulated annealing sampling), and nine series of sampling density (100,184,275,365,451,541,625,713,and1000) were applied in this study for exploring the effects of sampling design and spatial prediction method on the spatial predictive mapping of SOM content in Hailun topsoils. The results showed that:1. The spatial distribution patterns of SOM derived from different sampling designs are similar, whereas the abilities for identifying spatial structures are essentially different. With the increase of sampling densities, the identified continuous spatial component appeared as irregular fluctuation. Stratified random sampling model is the optimal design for revealing the spatial variability of SOM, and optimal sampling pattern design is more important than solely increasing sampling density.2. With the sampling numbers increasing, the spatial distribution of soil organic matter content is more detailed by ordinary kriging, while sampling numbers affect remarkably the spatial distribution of soil organic matter content by regression kriging. Whatever the prediction methods are, smooth effect of prediction by systematic regular sampling model is much more than the other three sampling models. On the contrary, the spatial local variation of soil organic matter content by stratified random sampling model is more reflective. The spatial distributed structure of soil organic matter content by simple random sampling model and simulation annealing model is similar.3.The difference between the means of RMSE for the four sampling designs are not significant, the systematic regular sampling model and simulated annealing sampling model are slight worse than the other two sampling designs. With respect to the ranges of RMSE, the minimum RMSE can be achieved by the simulated annealing sampling model and stratified random sampling model. Changes in MAE values were similar with those of RMSEs. Therefore, when field investigation is conducted at county level/scale, the stratified random sampling model and the simulated annealing sampling model are obviously better since prior geographic and spatial variability information were considered. The performance of systematic regular sampling model and simple random sampling model are relatively worse. The means and ranges of RMSE gradually reduced with the sampling densities increased, while the means of RMSE for the former three sampling series decreased more greatly than the later six series, namely, after365samples, the means of RMSE tend to decline steadily. Hence,365samples points are recommended in this study for mapping the spatial distribution patterns of SOM in Hailun County.4. The spatial variability structures represented by the semivariograms of different interpolation methods and sampling designs are all different significantly. Ranges of semivariogram using regression kriging method are smaller than those by using ordinary kriging. Differences in random structure component ratios for simple random sampling model and stratified random sampling model are not significant. While the ratio drop substantially in systematic regular sampling model, even very unstably. Apart from the100-sample series, no matter which sampling densities and designs, the RMSEs using regression kriging were even higher than those using ordinary kriging, indicating that regiession kriging method may not improve the spatial prediction accuracy of SOM in the study area.