Estimating the bulk density in 0– 20 cm of tilled soils in China’s Loess Plateau using support vector machine modeling

ABSTRACT

Pedotransfer functions (PTFs), as an indirect forecasting method, offer an alternative for labor-intensive bulk density (BD) measurements. In order to improve the forecasting accuracies, support vector machine (SVM) method was first used to develop PTFs for predicting BD. Cross-validation and grid-search methods were used to automatically determine the SVM parameters in the forecasting process. Soil texture and organic matter content were selected as input variables based on results of predecessors, coupled with gray correlation theory. And additional properties were added as inputs for improving PTF's accuracy and reliability. The performance of the PTF established by SVM method was compared with artificial neural network (ANN) method and published PTFs using two indexes: root-mean-square error (RMSE) and coefficient of determination(R²). Results showed that the average RMSE of published PTFs was 0.1053, and the R² was 0.4558. The RMSE of ANN–PTF was 0.0638, and the R² was 0.7235. The RMSE of SVM–PTF was 0.0558, and the R² was 0.7658. Apparently, the SVM–PTF had better performance, followed by ANN–PTF. Additionally, performances could be improved when accumulated receiving water was added as predictor variable. Therefore, the first application of SVM data mining techniques in the prediction of soil BD was successful, improved the accuracy of predictions, and enhanced the function of soil PTFs. The idea of developing PTFs using SVM method for predicting soil BD in the study area could provide a reference for other areas.     

Predicting Cationic Exchange Capacity in Calcareous Soils of East-Azerbaijan Province,Northwest Iran

The aim of this research is to study the efficiency of pedotransfer functions (PTFs) and artificial neural networks (ANNs) for cationic exchange capacity (CEC) prediction using readily available soil properties. Here, 417 soil samples were collected from the calcareous soils located in East-Azerbaijan province, northwest Iran and readily available soil properties, such as particle size distribution (PSD), organic matter (OM) and calcium carbonate equivalent (CCE), were measured. The entire 417 soil samples were divided into two groups, a training data set (83 soil samples) and test data set (334 soil samples). The performances of several published and derived PTFs and developed neural network algorithms using multilayer perceptron were compared, using a test data set. Results showed that, based on statistics of RMSE and R², PTFs and ANNs had a similar performance, and there was no significant difference in the accuracy of the model results. The result of the sensitivity analysis showed that the ANN models were very sensitive to the clay variable (due to the high variability of the clay). Finally, the models tested in this study could account for 85% of the variations in cationic exchange capacity (CEC) of soils in the studied area.

Abbreviations: ANN: arti?cial neural networks; MLP: multilayer perceptron; MLR: multiple linear regression; PTFs: Pedotransfer Functions; RBF: Radial Basis Function; MAE: mean absolute error; MSE: mean square error; CEC: cationic exchange capacity     

Prediction of Soil Cation Exchange Capacity Using Different Soil Parameters by Intelligent Models

ABSTRACT

The traditional methods for the measurement of soil cation exchange capacity (CEC) are time-consuming and laborious. It is also difficult to maintain stability for long-term experiments and projects. Therefore, it is necessary to develop an indirect approach such as pedotransfer functions (PTFs) to estimate this property from more easily available soil data. The aim of this study was to compare multiple linear and nonlinear regression, classification and regression trees (C&RT), artificial neural network (ANN) model included multiple layer perceptron (MLP) and k-nearest neighbors (k-NN) to develop PTFs for predicting soil CEC. Soil samples, 929, were used into two subsets for training and testing of the models. Sensitivity and statistical analyzes were conducted to determine the most and the least influential variables affecting soil CEC. The prediction capability of models was assessed by statistical indicators included the normalized root-mean-square error (NRMSE) and the coefficient of determination (R²). Results of the present investigation showed that the k-NN and ANN models had the ability to estimate soil CEC by computing easily measurable variables with a guarantee of authenticity, reliability, and reproducibility. Therefore, the results of this study provide a superior basis for predicting soil CEC and could be applied to other parts of the world with similar challenges.     

Comparative Analysis of MLR,ANN, and ANFIS Models for Prediction of Field Capacity and Permanent Wilting Point for Bafra Plain Soils

ABSTRACT

Soil hydraulic parameters like moisture content at field capacity and permanent wilting point constitute significant input parameters of various biophysical models and agricultural practices (irrigation timing and amount of irrigation to be applied). In this study, the performance of three different methods (Multiple linear regression – MLR, Artificial Neural Network – ANN and Adaptive Neuro-Fuzzy Inference System – ANFIS) with different input parameters in prediction of field capacity and permanent wilting point from easily obtained soil characteristics were compared. Correlation analysis indicated that clay content, sand content, cation exchange capacity, CaCO₃, and organic matter had significant correlations with FC and PWP (p < .01). Validation results revealed that the ANN model with the greatest R² and the lowest MAE and RMSE value exhibited better performance for prediction of FC and PWP than the MLR and ANFIS models. ANN model had R² = 0.83, MAE = 2.36% and RMSE = 3.30% for FC and R² = 0.81, MAE = 2.15%, RMSE = 2.89% for PWP in training dataset; R² = 0.80, MAE = 2.27%, RMSE = 3.12% for FC and R² = 0.83, MAE = 1.84%, RMSE = 2.40% for PWP in testing dataset. Also, Bayesian Regularization (BR) algorithm exhibited better performance for both FC and PWP than the other training algorithms.     

青海三江源地区土壤容重转换函数的建立与比较

Bulk density (BD) is an important soil physical property and has significant effect on soil water conservation function. Indirect methods, which are called pedotransfer functions (PTFs), have replaced direct measurement and can acquire the missing data of BD during routine soil surveys. In this study, multiple linear regression (MLR) and artificial neuron network (ANN) methods were used to develop PTFs for predicting BD from soil organic carbon (OC), texture and depth in the Three-River Headwater region of Qinghai Province, China. The performances of the developed PTFs were compared with 14 published PTFs using four indexes, the mean error (ME), standard deviation error (SDE), root mean squared error (RMSE) and coefficient of determination (R²). Results showed that the performances of published PTFs developed using exponential regression were better than those developed using linear regression from OC. Alexander (1980)-B, Alexander (1980)-A and Manrique and Jones (1991)-B PTFs, which had good predictions, could be applied for the soils in the study area. The PTFs developed using MLR (MLR-PTFs) and ANN (ANN-PTFs) had better soil BD predictions than most of published PTFs. The ANN-PTFs had better performances than the MLR-PTFs and their performances could be improved when soil texture and depth were added as predictor variables. The idea of developing PTFs or predicting soil BD in the study area could provide reference for other areas and the results could lay foundation for the estimation of soil water retention and carbon pool.     

Using support vector machines to predict cation exchange capacity of different soil horizons in Qingdao City,China

Agricultural, environmental and ecological modeling requires soil cation exchange capacity (CEC) that is difficult to measure. Pedotransfer functions (PTFs) are thus routinely applied to predict CEC from easily measured physicochemical properties (e.g., texture, soil organic matter, pH). This study developed the support vector machines (SVM)‐based PTFs to predict soil CEC based on 208 soil samples collected from A and B horizons in Qingdao City, Shandong Province, China. The database was randomly split into calibration and validation datasets in proportions of 3:1 using the bootstrap method. The optimal SVM parameters were searched by applying the genetic algorithm (GA). The performance of SVM models was compared to those of multiple stepwise regression (MSR) and artificial neural network (ANN) models. Results show that the accuracy of CEC predicted by SVM improves considerably over those predicted by MSR and ANN. The performance of SVM for B horizon (R² = 0.85) is slightly better than that for A horizon (R² = 0.81). The SVM is a powerful approach in the simulation of nonlinear relationship between CEC and physicochemical properties of widely distributed samples from different soil horizons. Sensitivity analysis was also conducted to explore the influence of each input parameter on the CEC predictions by SVM. The clay content is the most sensitive parameter, followed by soil organic matter and pH, while sand content has the weakest influence. This suggests that clay is the most important predictor for predicting CEC of both soil horizons.     

Model Development to Predict CEC Using the Intelligence Data Mining Approaches

ABSTRACT

Using easily measurable soil properties and pedotransfer functions (PTFs) is a time-saving, non-destructive and cost-saving way in the prediction of the cation exchange capacity (CEC). The purpose of this study was to compare and evaluate the regression tree (RT), multiple linear regression (MLR) and Mamdani fuzzy inference system (MFIS) in estimating CEC. For this work, 100 soil samples from unsaturated soil hydraulic database (UNSODA) data-set were used. %Organic matter (OM), bulk density (BD), the geometric mean particle diameter (dg) and fractal dimension of particle size (D) were applied as the input predictive variables. First, the type of relationship between easily measurable soil properties and CEC was investigated and, then used for the development of PTFs and fuzzy membership functions. The results showed that MLR method was developed only based on %OM (r = 0.68, p < .01) and D (r = 0.68, p < .01). While in the RT method, all of the predictive variables were appeared in the tree-like based on their correlation coefficient with CEC. The D and %OM also were considered as input variables in developing fuzzy membership functions. Results also revealed that RT method had a higher performance than MLR and MFIS in the estimation of CEC with the highest coefficient of determination (R² = 0.77), smallest root-mean-square error (RMSE = 5.14 meq/100_gsoil), normalized root-mean-square error (NRMSE = 0.25 meq/100_gsoil) and mean error (ME = ?1.80 meq/100_gsoil). In addition, the MFIS had a higher efficiency than the MLR in the CEC estimation.     

Soil erodibility and its prediction in semi-arid regions

ABSTRACT

Pedotransfer functions (PTFs) have been used to save time and cost in predicting certain soil properties, such as soil erodibility (K-factor). The main objectives of this study were to develop appropriate PTFs to predict the K-factor, and then compare new PTFs with Universal Soil Loss Equation (USLE) and the Revised Universal Soil Loss Equation (RUSLE) K-factor models. The K-factor was measured using 40 erosion plots under natural rainfall in Simakan Watershed, an area of 350 km² in central of Iran. The Regression Tree (RT) and Multiple Linear Regression (MLR) were used to develop PTFs for predicting the K-factor. The result showed that the mean of measured K was 0.01 t h MJ^?1 mm^?1. The mean K value predicted by USLE and RUSLE was 2.08 and 2.84 times more than the measured K, respectively. Although calcium carbonate was not considered in the original USLE and RUSLE K-factors, it appeared in the advanced PTFs due to its strong positive significant impact on aggregate stability and soil infiltration rate, resulting in decreased K-factor. The results also showed that the RT with R² = 0.84 had higher performance than developed MLR, USLE and RUSLE for the K estimation.     

Soil water-retention prediction from pedotransfer functions for some Indian soils

This paper discusses the development of pedotransfer functions (PTFs) and uses a multiple nonlinear regression technique to validate point and parametric PTFs for the estimation of a water-retention curve from basic soil properties such as particle-size distribution, bulk density and organic matter content. One hundred soil samples were collected at different depths from different locations in the Pavanje river basin that lies within the southern coastal region of Karnataka, India. Prediction accuracies were evaluated using the coefficient of determination (R ²), root mean square error (RMSE) and mean error (ME) between measured and predicted values. Overall, both point and parametric methods predicted water contents at selected water potentials with considerable accuracy. However, prediction of the soil water-retention curve using PTFs by point estimation method was relatively more successful (best case R ² = 0.983) for the sampled soils. F-tests were also conducted for all cases. For one regression equation, the p-value was zero and for other equation, values were close to zero. Critical comparative analysis was carried out on the performances of the point and parametric methods. Use of the developed PTFs is suggested for the prediction of a water-retention curve for loamy sand and sandy loam soils in this area of the coastal region of southern India.     

Support Vector Machine and Nonlinear Regression Methods for Estimating Saturated Hydraulic Conductivity

Pedotransfer functions (PTFs) are widely used for hydrological calculations based on the known basic properties of soils and sediments. The choice of predictors and the mathematical calculus are of particular importance for the accuracy of calculations. The aim of this study is to compare PTFs with the use of the nonlinear regression (NLR) and support vector machine (SVM) methods, as well as to choose predictor properties for estimating saturated hydraulic conductivity (K_s). K_s was determined in direct laboratory experiments on monoliths of agrosoddy-podzolic soil (Umbric Albeluvisol Abruptic, WRB, 2006) and calculated using PTFs based on the NLR and SVM methods. Six classes of predictor properties were tested for the calculated prognosis: K_s-1 (predictors: the sand, silt, and clay contents); K_s-2 (sand, silt, clay, and soil density); K_s-3 (sand, silt, clay, soil organic matter); K_s-4 (sand, silt, clay, soil density, organic matter); K_s-5 (clay, soil density, organic matter); and K_s-6 (sand, clay, soil density, organic matter). The efficiency of PTFs was determined by comparison with experimental values using the root mean square error (RMSE) and determination coefficient (R²). The results showed that the RMSE for SVM is smaller than the RMSE for NLR in predicting K_s for all classes of PTFs. The SVM method has advantages over the NLR method in terms of simplicity and range of application for predicting K_s using PTFs.     

沿海滩涂农田耕层土壤饱和导水率间接方法的评价

为筛选和构建适合苏北沿海滩涂围垦农田耕层土壤饱和水力传导率间接估算的土壤转换函数,在典型地块实测土壤饱和导水率和相关土壤基本性质的基础上,分析了11种根据基本土壤性质预测饱和导水率的转换函数方法的适用性,同时探讨了基于人工神经网络方法的土壤转换函数的预测效果。结果表明:滩涂围垦农田耕层土壤平均饱和导水率为10.04 cm/d,属低透水强度;在现有的土壤饱和导水率转换函数中,Vereecken函数是最适合滩涂围垦农区土壤、具有最佳预测精度的转换函数,其预测均方根误差为8.154,其次是Li、Campbell和Rawls函数;以砂粒、粘粒、容重和有机质作为输入因子,基于人工神经网络的土壤转换函数较Vereecken函数其预测均方根误差降至7.920,在输入因子中增加土壤盐分指标可进一步提高饱和导水率的预测精度,其预测均方根误差降为7.634。本文的研究结果显示利用人工神经网络方法建立的转换函数可有效提高滩涂盐渍农田土壤饱和导水率的预测精度。     

Estimation of bulk density of waterlogged soils from basic properties

Pedotransfer functions (PTFs) to predict bulk density (BD) from basic soil data are presented. Available data pertaining to seasonally impounded shrink–swell soils of Jabalpur district in the Madhya Pradesh state of India were used for the study. The data included horizon-wise information of 41 soil profiles in the study area covering nearly 5 million ha. Six independent variables, namely textural data (sand, silt and clay), field capacity (FC), permanent wilting point (PWP) and organic carbon content (OC) were used as input in hierarchical steps to establish dependencies, with bulk density as the dependent variable, using statistical regression and artificial neural networks. The PTFs derived using neural networks [average root mean square error (RMSE) 0.05] were relatively better than statistical regression PTFs (average RMSE > 0.1). The best-performing PTFs required input data on sand, silt content, FC and PWP, with lowest prediction errors (RMSE 0.01, maximum absolute error (MAE) 0.01) and highest values of index of agreement (d, 0.95) and R ² (0.65). Use of measures of structure, as well as information on pore structure, was found to be essential to derive acceptable PTFs. Inclusion of OC as an input variable showed relatively better fitting to the training data set, implying an underlying relationship between OC and BD, but the neural networks could not mimic the relationship when tested against subset.     

Global mapping of volumetric water retention at 100, 330 and 15 000 cm suction using the WoSIS database

Present global maps of soil water retention (SWR) are mostly derived from pedotransfer functions (PTFs) applied to maps of other basic soil properties. As an alternative, ‘point-based’ mapping of soil water content can improve global soil data availability and quality. We developed point-based global maps with estimated uncertainty of the volumetric SWR at 100, 330 and 15 000 cm suction using measured SWR data extracted from the WoSIS Soil Profile Database together with data estimated by a random forest PTF (PTF-RF). The point data was combined with around 200 environmental covariates describing vegetation, terrain morphology, climate, geology, and hydrology using DSM. In total, we used 7292, 33 192 and 42 016 SWR point observations at 100, 330 and 15 000 cm, respectively, and complemented the dataset with 436 108 estimated values at each suction. Tenfold cross-validation yielded a Root Mean Square Error (RMSE) of 6.380, 7.112 and 6.485 10⁻²cm³cm⁻³, and a Model Efficiency Coefficient (MEC) of 0.430, 0.386, and 0.471, respectively, for 100, 330 and 15 000 cm. The results were also compared to three published global maps of SWR to evaluate differences between point-based and map-based mapping approaches. Point-based mapping performed better than the three map-based mapping approaches for 330 and 15 000 cm, while for 100 cm results were similar, possibly due to the limited number of SWR observations for 100 cm. Major sources or uncertainty identified included the geographical clustering of the data and the limitation of the covariates to represent the naturally high variation of SWR.     

Application of VNIR and machine learning technologies to predict heavy metals in soil and pollution indices in mining areas

Purpose

Soil pollution indices are an effective tool in the computation of metal contamination in soil. They monitor soil quality and ensure future sustainability in agricultural systems. However, calculating a soil pollution index requires laboratory measurements of multiple soil heavy metals, which increases the cost and complexity of evaluating soil heavy metal pollution. Visible and near-infrared spectroscopy (VNIR, 350–2500 nm) has been widely used in predicting soil properties due to its advantages of a rapid analysis, non-destructiveness, and a low cost.

Methods

In this study, we evaluated the ability of the VNIR to predict soil heavy metals (As, Cu, Pb, Zn, and Cr) and two commonly used soil pollution indices (Nemerow integrated pollution index, NIPI; potential ecological risk index, RI). Three nonlinear machine learning techniques, including cubist regression tree (Cubist), Gaussian process regression (GPR), and support vector machine (SVM), were compared with partial least squares regression (PLSR) to determine the most suitable model for predicting the soil heavy metals and pollution indices.

Results

The results showed that the nonlinear machine learning models performed significantly better than the PLSR model in most cases. Overall, the SVM model showed a higher prediction accuracy and a stronger generalization for Zn (R²_V?=?0.95, RMSE_V?=?6.75 mg kg^?1), Cu (R²_V?=?0.95, RMSE_V?=?8.04 mg kg^?1), Cr (R²_V?=?0.90, RMSE_V?=?6.57 mg kg^?1), Pb (R²_V?=?0.86, RMSE_V?=?4.14 mg kg^?1), NIPI (R²_V?=?0.93, RMSE_V?=?0.31), and RI (R²_V?=?0.90, RMSE_V 3.88). In addition, the research results proved that the high prediction accuracy of the three heavy metal elements Cu, Pb, and Zn and their significant positive correlations with the soil pollution indices were the reason for the accurate prediction of NIPI and RI.

Conclusion

Using VNIR to obtain soil pollution indices quickly and accurately is of great significance for the comprehensive evaluation, prevention, and control of soil heavy metal pollution.

     

Enhancing pedotransfer functions with environmental data for estimating bulk density and effective cation exchange capacity in a data‐sparse situation

Soil bulk density (BD) and effective cation exchange capacity (ECEC) are among the most important soil properties required for crop growth and environmental management. This study aimed to explore the combination of soil and environmental data in developing pedotransfer functions (PTFs) for BD and ECEC. Multiple linear regression (MLR) and random forest model (RFM) were employed in developing PTFs using three different data sets: soil data (PTF‐1), environmental data (PTF‐2) and the combination of soil and environmental data (PTF‐3). In developing the PTFs, three depth increments were also considered: all depth, topsoil (<0.40 m) and subsoil (>0.40 m). Results showed that PTF‐3 (R²; 0.29–0.69) outperformed both PTF‐1 (R²; 0.11–0.18) and PTF‐2 (R²; 0.22–0.59) in BD estimation. However, for ECEC estimation, PTF‐3 (R²; 0.61–0.86) performed comparably as PTF‐1 (R²; 0.58–0.76) with both PTFs out‐performing PTF‐2 (R²; 0.30–0.71). Also, grouping of data into different soil depth increments improves the estimation of BD with PTFs (especially PTF‐2 and PTF‐3) performing better at subsoils than topsoils. Generally, the most important predictors of BD are sand, silt, elevation, rainfall, temperature for estimation at topsoil while EVI, elevation, temperature and clay are the most important BD predictors in the subsoil. Also, clay, sand, pH, rainfall and SOC are the most important predictors of ECEC in the topsoil while pH, sand, clay, temperature and rainfall are the most important predictors of ECEC in the subsoil. Findings are important for overcoming the challenges of building national soil databases for large‐scale modelling in most data‐sparse countries, especially in the sub‐Saharan Africa (SSA).     

Comparison of Different Methods of Estimating Saffron Yield Based on Soil Properties in Golestan Province

ABSTRACT

Evaluation of the relationship between soil properties and saffron yield estimation may contribute to agricultural planning in finding suitable lands for the growth of this valuable product. This study aimed to investigate the performance of artificial neural network (ANN), multiple linear regression (MLR), and adaptive neuro-fuzzy inference system (ANFIS) in terms of saffron yield estimation in some lands of Golestan province, Iran. To this end, 100 areas under saffron cultivation were selected. For rapid and low-cost saffron yield estimation, six different models were designed based on soil properties as inputs using MLR, ANN, and ANFIS methods. According to the results, ANN showed the highest accuracy (R² = 0.58–0.89) in estimating saffron yield as compared to MLR (R² = 0.41–0.47) and ANFIS (R² = 0.41–0.69) models. A comparison of the results obtained from the six models defined in these three methods indicated that Model 4 (R² _Reg = 0.45, R² _ANFIS = 0.57, R² _ANN = 0.87), with the inputs, organic phosphorus, potassium, and calcium carbonate, was the best model in terms of accuracy and speed in estimating saffron yield phosphorus. The RI indexes for ANN in the model were 50% and 34% relative to MLR and ANFIS, respectively, demonstrating the higher accuracy of ANN in saffron yield estimation. The study results can be used to identify lands suitable for saffron cultivation in the study area using organic phosphorus and organic matter levels in the soil.     

Development of ensemble pedotransfer functions for cation exchange capacity of soils of Qingdao in China

Soil cation exchange capacity (CEC), which is considered to be an indicator of buffering capacity, is an important soil attribute that influences soil fertility but is costly, time‐consuming and labour‐intensive to measure. Pedotransfer functions (PTFs) have routinely been used to predict soil CEC from easily measured soil properties, such as soil pH, texture and organic matter content. However, uncertainty in which one to select can be substantial as different PTFs do not necessarily produce the same result. In this study, a total of 100 soil samples were collected from surface horizons (0–20 cm) in different regions of Qingdao City, China. Three ensemble PTFs (ePTFs), including simple ensemble mean (SEM), individually bias‐removed ensemble mean (IBREM) and collective bias‐removed ensemble mean (CBREM), were developed to reduce the uncertainty in CEC prediction based on 12 published regression‐based PTFs. In addition, a local PTF (LPTF) for CEC was also developed using multiple stepwise regression and basic soil properties. The performances of the three ePTFs were compared with those of the published PTFs and LPTF. Results show that the differences between the performances of the published PTFs were substantial. When the systematic bias of each published PTF was removed separately, the prediction capability of the PTFs was increased. The performance of LPTF was significantly better than that of SEM, but slightly worse than IBREM. It is noted that CBREM had higher accuracy than all of the other methods. Overall, CBREM is a promising approach for estimating soil CEC in the study area.     

长短期记忆神经网络在多时次土壤水分动态预测中的应用

基于长沙站2016—2019年10 cm深度土壤水分自动观测小时数据集,利用长短期记忆神经网络(LSTM)模型结合随机采样学习方法,开展了土壤水分多时次预测,结果表明:LSTM模型对6、12、24、48h后的土壤体积含水量预测均方根误差(RMSE)分别为0.22%、0.28%、0.38%、0.54%,决定系数(R2)分别为0.99、0.99、0.98、0.96,除6 h预测步长外,准确率均优于自回归整合滑动平均(ARIMA)模型,且误差稳定、无异常值出现,预测准确率远优于相关研究。该结果证实了基于LSTM模型精准预测土壤水分动态的可行性,为精准灌溉和干旱预警提供了计算机技术及手段支撑,为政府及科研部门水资源管理政策的制定提供了数据支持。     

Pedo-transfer functions for estimating the hydraulic properties of paddy soils in subtropical central China

Pedo-transfer functions (PTFs) have been widely used to estimate soil hydraulic properties in the simulation of catchment eco-hydrological processes. However, the accuracy of existing PTFs is usually inadequate for use. To develop PTFs for local use, soil columns were collected from a double rice-cropped agricultural catchment in subtropical central China. The PTFs for saturated soil hydraulic conductivity (K_s) and parameters (θ_s, α, and n) of the van Genuchten model for the soil water retention curve (SWRC) were obtained based on soil’s basic properties, and compared with models developed by Li et al. in 2007 and Wösten et al. in 1999, respectively. Our results indicated that K_s in the range of 0.04–1087 cm d^?1 and θ_s in the range of 0.34–0.51 cm³ cm^?3 were both well estimated with the R²_adj of 0.72 and 0.87, respectively, but α (0.04–0.65 cm^?1) and n (1.05–1.21) were relatively poorly predicted with the respective R²_adj of 0.38 and 0.55, despite the use of more input parameters. Our local derived PTFs outperformed the other two existing models. However, if the local PTFs for paddy soils are not available, the Wösten et al. 1999 model can be proposed as a useful alternative. Therefore, this study can improve our understanding of the development and application of PTFs for predicting paddy soil hydraulic properties in China.