English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

. Soil texture and soil particle size fractions (PSFs) playan increasing role in physical, chemical, and hydrological processes. Manyprevious studies have used machine-learning and log-ratio transformationmethods for soil texture classification and soil PSF interpolation toimprove the prediction accuracy. However, few reports have systematicallycompared their performance with respect to both classification and interpolation. Here,five machine-learning models – K-nearest neighbour (KNN), multilayerperceptron neural network (MLP), random forest (RF), support vector machines(SVM), and extreme gradient boosting (XGB) – combined with the original data and three log-ratio transformation methods – additive log ratio (ALR), centred log ratio (CLR), andisometric log ratio (ILR) – were applied to evaluate soil texture andPSFs using both raw and log-ratio-transformed data from 640 soil samples in the Heihe River basin(HRB) in China. The results demonstrated that the log-ratio transformationsdecreased the skewness of soil PSF data. For soil textureclassification, RF and XGB showed better performance with a higher overallaccuracy and kappa coefficient. They were also recommended to evaluate theclassification capacity of imbalanced data according to the area under theprecision–recall curve (AUPRC). For soil PSF interpolation, RFdelivered the best performance among five machine-learning models with thelowest root-mean-square error (RMSE; sand had a RMSE of 15.09 %, silt was 13.86 %, andclay was 6.31 %), mean absolute error (MAE; sand had a MAD of 10.65 %, silt was 9.99 %, and clay was 5.00 %), Aitchison distance (AD; 0.84), and standardizedresidual sum of squares (STRESS; 0.61), and the highest Spearman rankcorrelation coefficient (RCC; sand was 0.69, silt was 0.67, and clay was 0.69). STRESSwas improved by using log-ratio methods, especially for CLR and ILR. Predictionmaps from both direct and indirect classification were similar in the middle andupper reaches of the HRB. However, indirect classification maps using log-ratio-transformed data provided more detailed information in the lower reaches ofthe HRB. There was a pronounced improvement of 21.3 % in the kappacoefficient when using indirect methods for soil texture classification comparedwith direct methods. RF was recommended as the best strategy among the fivemachine-learning models, based on the accuracy evaluation of the soil PSFinterpolation and soil texture classification, and ILR was recommended forcomponent-wise machine-learning models without multivariate treatment,considering the constrained nature of compositional data. In addition, XGBwas preferred over other models when the trade-off between the accuracy and runtime wasconsidered. Our findings provide a reference for future works with respect to thespatial prediction of soil PSFs and texture using machine-learning modelswith skewed distributions of soil PSF data over a large area.

hydrology and earth system sciences

Systematic comparison of five machine-learning models in classification and interpolation of soil particle size fractions using different transformed data