Study On The Method Of Mapping Soil Particle-size Fractions

Soil particle-size fractions(psf)as basic physical variables need to be accurately predicted for regional hydrological,ecological,geological,agricultural and environmental studies frequently.Some methods such as compositional kriging(CK),log-ratio kriging(Log-ratio OK),log-ratio cokriging(Log-ratio COK)had been proposed to interpolate the spatial distributions of soil psf,but the performance of these methods and the applicability of these methods are still unclear.Four log-ratio transformations,including additive log-ratio(ALR),centered log-ratio(CLR),isometric log-ratio(ILR),and symmetry log-ratio(SLR),combined with OK are named log-ratio OK: ALR_OK,CLR_OK,ILR_OK and SLR_OK);combined with COK are named log-ratio COK: ALR_COK,CLR_COK,ILR_COK,SLR_COK;combined with robust estimator for the variograms in COK are named log-ratio RCOK: ALR_RCOK,CLR_RCOK,ILR_RCOK and SLR_RCOK,and the compositional kriging including CK,Log-ratio CK and Log-ratio RCK were selected to be used and compared the performance of spatial prediction of soil psf in different space scale areas of Heihe River Basin,China,respectively.In addition,ALR and ILR combined with environmental variables through generalized linear models(GLM)and random forest(RF)are introduced to map for the whole Heihe River Basin.Root mean squared error(RMSE),Aitchison's distance(AD),standardized residual sum of squares(STRESS)and right ratio of the predicted soil texture types(RR)were chosen to evaluate the accuracy for different interpolators.The results showed that log-ratio CK had a better accuracy than the four log-ratio kriging methods in Tianlaochi Basin.The RMSE(sand,9.27%;silt,7.67%;clay,4.17%),AD(0.45),STRESS(0.60)of CK were the lowest and the RR(58.65%)was the highest in the five interpolators.ILR_RCOK had the best accuracy with RMSE(sand,5.98%;silt,5.84%;clay,3.85%;AD,0.37;STRESS,0.40)in upper reaches of Heihe River Basin.The ILR combined with different interpolators achieved relatively better performance than the other log-ratio transformed methods.In addition,the interpolators combined with robust estimator have significantly improved the predicting accuracy comparing with the standard estimator for the variograms on mapping soil psf.The CK,log-ratio OK,log-ratio COK and log-ratio RCOK methods system are used in the whole Heihe River Basin.The results showed ILR_COK have the best accuracy with RMSE(sand,11.29%;silt,10.6%;clay,4.79%).In addition,ILR_RF have the best accuracy(RMSE: sand,12.34%,silt,11.78%,clay,6.27%)and have significantly improved the predicting accuracy comparing with the standard estimator and presented reasonable and smooth transition on mapping soil psf.The study gives insights for mapping soil psf accurately by comparing different methods for compositional data interpolation and could guide people to select the appropriate interpolator according to the data characteristics.The further researches of methods could develop log-ratio transformations combined with some new high accuracy interpolators such as HASM to improve mapping performance of compositional data.