BP神经网络组合模型在次洪量预测中的应用

[目的]探讨BP神经网络组合模型在次洪量预测中的应用,为黄土高原淤地坝群的安全度汛提供决策依据。[方法]构建基于多元线性回归模型(MLR)和去趋势互相关分析法(DCCA)的BP神经网络组合模型;选择均方差(MSE)、平均绝对误差(MAE)、平均绝对百分比误差(MAPE)以及确定性系数(DC)作为评价指标,与单一模型(多元线性回归模型、BP神经网络模型以及去趋势互相关分析法)进行比较。[结果]BP神经网络组合模型的4项指标MSE,MAE,MAPE和DC分别为2.144,5.453,0.074和0.988,均优于单一模型;模型预测效果从优到劣分别为BP神经网络组合模型、BP神经网络模型、多元线性回归模型和去趋势互相关分析法。[结论]BP神经网络组合模型较单一模型平稳性增强,提高了预测效果,可用于淤地坝群的次暴雨洪量预测。     

气象资料受限条件下BP神经网络优化模型模拟参考作物蒸散量:以京津冀地区为例

为确定气象数据缺乏地区参考作物蒸散量(ET₀)的最优简化估算模型,本文以京津冀地区作为研究区域,以传统BP神经网络模型为基础,基于粒子群算法(PSO)、遗传算法(GA)、思维进化算法(MEA)、麻雀算法(SSA)和人工鱼群算法(AF)5种优化算法,构建了PSO-BP、GA-BP、MEA-BP、SSA-BP、AF-BP共5种优化模型,并将计算结果与3种传统机器学习模型BP模型、随机森林模型(RF)、小波神经网络模型(WNN)和2种经验模型Hargreaves模型(HS)、Droogres-Allen模型(DA)进行对比,在仅输入温度数据的条件下,得出区域ET₀最优估算模型。结果表明:在不同区域,5种优化模型计算精度显著高于其余模型,其中,SSA-BP模型均表现出了较高的精度,RMSE、R~2、E_ns和MAE分别为0.297～0.402mm·d^-1、0.879～0.946、0.862～0.940、0.210～0.300mm·d^-1,模型GPI在研究区域内排名第1位;在相同气象数据条...     

基于神经网络优化算法的云南土壤有机质含量数字制图——以景洪市为例

  目的  利用自然成土作用变量,预测并制作栅格化的土壤有机质分布图,对发展热带数字化精细农业具有重要意义。  方法  使用2006年云南省景洪市测土配方样点数据,应用BP神经网络（BPNN）、基于强分类器算法的BP神经网络模型（BPNN-Ada）、基于粒子群算法优化的BP神经网络（PSO-BPNN）、基于遗传算法优化的BP神经网络（GA-BPNN）和多元线性回归（MLR）对土壤有机质的含量预测并进行空间化制图。  结果  ① 土壤样点X、Y坐标值能够有效提高算法精度且充分表现环境因子与土壤有机质空间分布上的协同关系。② 4种神经网络算法预测结果土壤有机质空间分布基本类似,均呈现南高北低的趋势。③ 研究区域内4种神经网络模型的在建模集拟合程度从高至低依此次为:BPNN-Ada > GA-BPNN > PSO-BPNN > BPNN,在建模集中PSO-BPNN和GA-BPNN与BPNN拟合精度一致,BPNN-Ada的拟合精度R2最高为0.98。在验证集的预测能力由高至低依次为:BPNN-Ada > GA-BPNN > PSO-BPNN > BPNN。BPNN-Ada有着最高的预测精度和算法稳定性:RMSE = 4.47、MAE = 3.3、MRE = 0.05、R2 = 0.976。  结论  在景洪地区进行土壤有机质神经网络建模时加入地理坐标能够有效提高模型精度,且基于学习规则的神经网络优化算法效果要优于优化初始权重和阈值的神经网络算法及传统的BPNN算法。     

BP神经网络和SVM模型对施加生物炭土壤水分预测的适用性

生物炭作为土壤改良剂对半干旱区土壤水分有良好的吸持作用,为确定施加生物炭对土壤水分预测模型适用性的影响,依托黄土高原半干旱区固原生态站开展了小区定位试验。向土壤中施加不同种类及比例的生物炭,定期监测土壤水分含量;考虑土壤含水量的非线性特征以及生物炭对土壤水分的影响,选取BP神经网络和SVM支持向量机两种模型,建立施加生物炭土壤水分预测模型。计算预测值,并与实测值对比,分析相对误差;利用RMSE、MRE、MAE和R2评估BP神经网络和SVM模型的精度。结果表明;BP神经网络预测值的平均相对误差为3.78%,最大误差为13.14%;SVM模型的平均相对误差为0.56%,最大误差为2.42%。SVM模型的RMSE、MRE、MAE值(分别为0.34~0.17,0.07,0.56~1.27)均小于BP神经网络的(分别为1.04~1.16,0.47~0.68,3.78~4.57),且决定系数R2值SVM模型(0.96~0.99)大于BP神经网络(0.56~0.64)。BP神经网络和SVM模型均能很好地预测施加生物炭的土壤水分,但SVM模型预测结果更加稳定,精度较高,更适于施加生物炭土壤水分的预测。该研究可为半干旱地区生物炭还田土壤水分的预测及管理提供理论依据。     

基于不同PSO-ELM模型的碾压黏土抗剪强度预测方法研究

碾压黏土的抗剪强度直接影响碾压土石坝的质量和使用寿命。为得出碾压黏土抗剪强度的最优预测模型,通过粒子群算法优化极限学习机模型(PSO-ELM),分别以Sine函数、radbas函数和hardlim函数3种激活函数为基础,构建PSO-ELM_sin,PSO-ELM_rad和PSO-ELM_hard3种模型,并将模型结果与ELM模型、广义回归神经网络模型(GRNN)、随机森林模型(RF)和BP神经网络模型进行了对比。结果表明:在黏聚力和内摩擦角的拟合结果中,PSO-ELM_sin模型精度最高,其拟合方程斜率分别为1.005,1.032; 在月值模拟中,PSO-ELM_sin模型与实测值的拟合度最高,相对误差仅在6.0%～9.3%; PSO-ELM_sin模型在黏聚力模拟中RMSE,RRMSE,MAE,E_ns和R²分别为0.776 kPa,1.80%,0.641 kPa,0.993和0.997,该模型在内摩擦角模拟中RMSE,RRMSE,MAE,E_ns和R²分别为1.635°,6.98%,1.616°,0.983和0.998,模型精度均排名第一。因此,PSO-ELM_sin模型在所有模型中精度最高,可作为碾压黏土抗剪强度预测的标准模型使用。     

葡萄大棚小气候预测模型研究

为精准把控并及时调节葡萄大棚棚内小气候,利用清徐县葡萄大棚农田小气候站观测数据及气象站、辐射站、土壤水分站资料,建立以棚外气温、相对湿度、风速、总辐射、土壤湿度为输入变量,棚内气温、相对湿度、土壤温度为输出变量的基于BP神经网络葡萄大棚小气候预测模型。为了对比分析BP神经网络的精确度和稳定性,同时建立多元线性回归模型。结果表明,基于BP神经网络建立的预测模型,其训练值和实测值之间的绝对误差分别为1.55 ℃、4.46%、0.77 ℃,标准误差分别为2.18 ℃、5.94%、1.00 ℃;预测值和实测值之间的绝对误差分别为1.37 ℃、2.84%、0.42 ℃,标准误差分别为1.96 ℃、4.60%、0.53 ℃。预测效果明显优于多元线性回归模型,预测精度满足棚内小气候要素预报要求。     

基于水云模型的Sentinel-1A双极化反演植被覆盖区土壤水分

根据开展的景县地区玉米农田主被动微波遥感协同反演土壤水分监测试验,提出利用FY-3 B WMRI被动微波数据计算微波极化差异指数(MPDI),建立了植被层含水量反演模型,去除植被含水量对于农田土壤水分反演的影响,然后结合植被层含水量反演模型和水云模型,以及Sentinel-1A主动微波数据和部分实测样点土壤体积含水量数据建立植被覆盖区农田土壤水分半经验反演模型,最后验证分析Sentinel-1A VV/VH不同极化条件下土壤水分反演模型的精度。结果表明:VV极化条件下土壤水分反演模型反演精度为R~2=0.7422,RMSE=0.0674 cm~3/cm~3,MAE=0.0305 cm~3/cm~3,MaxE为0.1196 cm~3/cm~3,MinE=0.0024 cm~3/cm~3;VH极化模型为R~2=0.1898,RMSE=0.0768 cm~3/cm~3,MAE=0.0474 cm~3/cm~3,MaxE=0.1933 cm~3/cm~3,MinE=0.0190 cm~3/cm~3。研究区VH极化模型反演值存在普遍偏低现象,VV极化模型土壤水分反演结果优于VH极化模型。VV极化方式具有更强的穿透性,受到植被层衰减作用影响较小,对土壤水分含量变化也更为敏感。建立的VV极化条件下土壤水分反演半经验模型能较好地表征研究区土壤水分空间分布情况。     

青海三江源地区土壤水分常数转换函数的建立与比较

利用土壤理化性质数据建立转换函数是间接获得土壤水力参数的重要手段之一。基于测定的土壤理化性质和土壤水分常数数据,本文采用回归分析、BP神经网络和基于BP神经网络的Rosetta模型3种方式分别建立了青海三江源地区土壤饱和含水量、毛管持水量和田间持水量的转换函数,并对其预测精度进行了比较。结果表明:(1)回归分析方法总体预测效果比较理想,特别是田间持水量的平均误差(ME)和均方根误差(RMSE)都在3.397%以下,决定系数(R2)高达0.868;(2)BP神经网络方法的预测效果非常理想,各土壤水分常数平均误差和均方根误差都在4.685%以下,并且决定系数均在0.857以上;(3)Rosetta模型的预测效果相对较差,特别是饱和含水量和毛管持水量,平均误差(ME)和均方根误差(RMSE)相对较大,决定系数(R2)相对较小。3种方式中,BP神经网络方法所建立的毛管持水量和饱和含水量转换函数均为最佳,回归方法所建立的田间持水量的转换函数要好于BP神经网络方法和Rosetta模型,Rosetta模型对土壤水分常数的预测效果不如其他两种方式。研究可为青海三江源地区土壤水力特性参数研究以及区域尺度上土壤水分估算提供科学依据。     

基于AMSR-2蒙古高原土壤水分反演及 对气象因子响应分析

土壤水分是地表水文过程研究的一个重要参数,是众多环境因子综合作用的结果,科学判定土壤水分对环境因子的响应特性,对在蒙古高原地区开展干旱监测预警,调整农业生产结构,改善区域生态环境具有重要意义。本研究基于AMSR-2观测亮温、SPOT-NDVI数据,利用微波辐射传输模型及粗糙地表发射率Qp模型,构建适合蒙古高原的土壤水分反演方程,同时将模型应用于2013年蒙古高原植被生长期土壤水分反演。在此基础上,结合TRMM 3B43降雨量及气象站点气温数据,探讨了蒙古高原土壤水分对气象因子及植被的响应特性。结果表明:1)构建的蒙古高原表层土壤水分反演模型精度较高,土壤水分反演值与实测值的判定系数为0.680 6,均方根误差(RMSE)达0.031 6 cm3·cm-3,反演结果明显优于JAXA提供的AMSR-2土壤水分产品数据(RMSE=0.044 1 cm3·cm-3)。2)TRMM 3B43降雨数据与实测降雨量线性拟合,其判定系数为0.859 8,直线拟合斜率K=0.941 5,在数值上较站点实测值略微偏低,表明TRMM 3B43数据精度较高,在蒙古高原具有很好的适用性。3)蒙古高原植被生长期土壤水分、植被指数及降水量在空间格局上均表现出由北向南、由东北向西南逐渐减少的趋势。干旱区,土壤水分对气温变化最敏感,二者表现出显著正相关关系,其次为降水和植被;半干旱区,植被是影响土壤水分的关键因子,而气温与降水对土壤水分影响呈现出季节性变化;半湿润区3个因子对土壤水分的影响程度表现为植被降水气温。总之,利用土壤水分对气象因子和植被的响应特性,可以采取适当措施降低蒙古高原灾害发生风险,为区域生态环境建设提供科学依据。     

利用线性和非线性耦合方式建立温室温湿度预测模型

基于蔬菜种植试验温室内温度、相对湿度和光照强度的实测数据，根据ARIMA模型和RBF神经网络对线性和非线性问题的预测能力差异，构建ARIMA-RBF神经网络权重组合的温湿度预测模型，对温室内温度和湿度的动态变化进行预测，并比较各模型预测精度。结果表明：温室内温湿度分别具有更明显的线性和非线性变化特征，对应预测性能较好的单一模型分别为ARIMA模型和RBF模型。相较单一模型，ARIMA-RBF神经网络权重组合模型的预测精度更高、稳定性更好。最佳温度组合模型的MAE、MAPE和RMSE分别为1.04℃、2.95%和1.21℃；最佳湿度组合模型的MAE、MAPE和RMSE分别为0.35个百分点、0.36%和0.55个百分点。权重组合模型通过适当的加权策略充分发挥了单一模型对数据不同特征的处理能力，能较好地评估温室内温湿度状态，可为建立更具普适性的温室环境因子模型提供参考。     

基于1D-CNN的土壤全氮近红外光谱预测模型

摘要：【目的】传统的基于近红外光谱数据预测土壤全氮的方法需要对原始光谱数据做复杂的预处理，筛选出与土壤全氮含量相关性高的敏感波长之后进行模型的回归拟合。本文提出一种一维卷积神经网络（1D-CNN）模型，可以在对数据进行简单预处理甚至无处理的情况下达到非常理想的结果，实现用近红外光谱技术对土壤全氮含量的预测。【方法】于江苏无锡采集410个土壤样品，利用半微量开氏法(NY/T 53-1987)测定土壤的全氮含量，并利用NIR Quest 512光谱仪，在室内环境下对每份土壤样品做光谱检测，并用均值中心化(CT)、标准正态变换(SNV)、趋势校正(DT)对光谱进行预处理，运用偏最小二乘回归（PLS）、BP神经网络、1D-CNN方法建立土壤全氮含量的回归预测模型。每种模型在采用不同预处理方法的数据集上做十折交叉验证，记录预测模型的决定系数(R2)和均方根误差(RMSE)的平均值，并对比三种预处理方法对模型精度的影响。【结果】证明了本文提出的1D-CNN模型基于土壤近红外光谱数据预测土壤全氮含量的可靠性。使用原始数据与经均值中心化、标准正态变换、趋势校正预处理的数据训练得到的1D-CNN模型的决定系数分别为0.907、0.931、0.922、0.964，构建的PLS回归模型决定系数为0.856、0.863、0.861、0.880，训练的BP神经网络的决定系数为0.874、0.907、0.901、0.911。【结论】本文提出的1D-CNN模型在原始数据和经预处理的光谱数据上的表现都优于PLS和BP神经网络，且可以证明，对光谱数据进行预处理能够有效提高1D-CNN模型的性能，尤其是趋势校正对模型的提升效果最明显。研究表明，1D-CNN能更好地提取光谱特征并建立其与含氮量的映射关系，有效地避免过拟合，在未经过预处理的光谱数据上依然能够达到一定的精度。     

BP神经网络算法在河西绿洲玉米生产碳排放评估中的应用及算法有效性研究

针对作物生产碳排放预测较为困难的实际问题,提出基于BP神经网络算法的玉米生产碳排放预测模型。选择地处河西走廊石羊河下游的民勤绿洲246家农户,面对面调查玉米种植户农场内生产投入数据,将玉米生产投入数据作为神经网络输入层;查阅和梳理国内外相似区域玉米生产环节碳排放系数,运用碳足迹生命周期法计算得到的碳排放值作为神经网络输出层;基于BP人工神经网络算法,运用试凑法确定网络隐含层节点个数,建立河西绿洲玉米生产碳排放预测模型,选择多元线性回归模型、多元非线性回归模型,对该模型有效性进行评估。研究结果表明,3层且各层节点数9、10、1的神经网络结构能够准确预测河西绿洲玉米生产碳排放,其碳排放预测值为0.763 kg(CO_2-eq)·kg~(-1)(DM);9-10-1结构的神经网络预测模型的相关系数(R~2=0.984 7)高于多元线性和非线性回归模型,该神经网络结构模型的均方根误差(RMSE=0.069 1)、平均绝对误差(MAE=0.051 3)均低于其他模型,BP神经网络算法预测性能明显优于其他预测模型。该研究为准确预测农业生产碳排放提供了新思路和可操作方法。     

Pedotransfer function to predict density of forest soils in Switzerland

Soil density is an important soil property, but respective measurements are usually scarce. With data from 559 mineral soil horizons (134 sites) we developed a linear regression pedotransfer function (PTF) for the density of forest soils (sieved to ≤ 2 mm). The field estimate of density was the most important covariate. RMSE of 0.205 Mg m^?3 and R² of 0.67, calculated on independent data (131 horizons), were better than the statistics obtained by published, recalibrated PTF (RMSE 0.271–0.324 Mg m^?3; R² 0.28–0.42).     

主成分分析和长短时记忆神经网络预测水产养殖水体溶解氧

为了提高水产养殖溶解氧预测的精度,提出了基于主成分分析(principal component analysis,PCA)和长短时记忆神经网络(long short-term memory,LSTM)的水产养殖溶解氧预测模型。首先通过主成分分析提取水产养殖溶解氧的关键影响因子,消除了原始变量之间的相关性,降低了模型输入向量维度;然后,在Tensorflow深度学习框架的基础上建立LSTM神经网络的水产养殖溶解氧预测模型;最后,利用该模型对浙江省淡水水产养殖研究所综合实验基地某池塘溶解氧进行验证。试验结果表明:该模型与BP神经网络等其他浅层模型相比,模型评价指标平均绝对误差、均方根误差和平均绝对误差分别为0.274、0.089和0.147,均优于传统的预测方法;该模型具有良好的预测性能和泛化能力,能够满足水产养殖溶解氧精确预测的实际需要,可以为水产养殖水质精准调控提供参考。     

东北农牧交错带耕地土壤有机质遥感反演研究

农牧交错带是农耕区与草原牧区的过渡带，土壤有机质(SOM)的精确估算与变化监测对碳库估算与农业生产具有重要研究意义。以东北典型农牧交错带为研究区，Landsat 8 OLI影像和ALOS 12.5m DEM为数据源，基于波段反射率、反射率对数、亮度指数与相关地形因子，分别利用多元线性逐步回归(MLSR)模型、随机森林(RF)模型和BP神经网络(BPNN)模型，构建农牧交错带SOM多光谱反演模型。结果表明：(1)根据重要性排序，选择Landsat8OLI第4波段的对数、第5波段、第6波段和亮度指数作为输入量，RF和BPNN模型的精度优于MLSR模型。(2)引入高程(E)与坡向变率(SOA)后，3种模型的预测精度提高，BPNN模型精度提高最多，R2提高了0.22,RMSE降低了0.40 g/kg。3种模型最优反演精度由高到低为：BPNN模型(R2=0.82,RMSE=1.4 g/kg)>RF模型(R2=0.71,RMSE=1.9 g/kg)>MLSR模型(R2=0.66,RMSE=8.8 g/kg)。研究结果可为农牧交错带SOM时空变化研究提供方法支撑。     

Spatial modeling of soil salinity using multiple linear regression,ordinary kriging and artificial neural network methods

Salinity as an important property of soil plays a major role in reducing the fertility in the world. Accurate information about the spatial change of soil salinity is essential for sustainable soil management and utilization in agriculture lands. For this purpose, 150 soil samples were collected from Dashte-e-Tabriz Iran and tested and soil salinity was estimated by land surface parameters including elevation, aspect, length of slope, wetness index, slope and normalized difference vegetation index as basic parameters. In order to model and predict the salinity, ordinary kriging (OK), artificial neural networks (ANN) and multiple linear regressions (MLR) were used. Accuracy of models was evaluated by the coefficient of determination (R²), root mean square error (RMSE) and mean absolute error (MAE). Based on Pearson correlation, elevation, normalized difference vegetation and wetness indices were selected for soil salinity spatial modeling from six land surface parameters. The results showed that the ANN had the lowest RMSE and highest R2. The values of R2, RMSE and MAE were 0.36, 25.89 and 17.06 for regression and 0.56, 17.70 and 13.05 for OK and 0.69, 16.06 and 11.60 for ANN, respectively, which indicated more accuracy of ANN in comparison with MLR and OK.     

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.     

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     

Neural network models to predict soil water retention

Soil water retention curves are needed to describe the availability of soil water to plants and to model movement of water through unsaturated soils. Measuring these characteristics is time-consuming, labour-intensive and therefore expensive. This study was conducted to develop and evaluate functions based on neural networks to predict soil water retention characteristics. Dutch and Scottish data sets were available; they contained data on 178 and 165 soil horizons, respectively. A series of three neural networks (A, B and C) was developed. Neural network A had the following input variables: topsoil, bulk density, organic matter, clay, silt and sand content. In addition neural network B had matric potential as input, and network C included soil structural data expressed as the upper and lower boundary of the ped-size class. Neural network A had three output variables: the volumetric water content at matric potentials of 0, –100 and –15 000 hPa. Both models B and C had volumetric water content, at the matric potential given as input, as output variable. The networks were tested against independent data that were extracted from the original sets of soil profiles. Accuracy of the predictions was quantified by the root of the mean squared difference (RMSE) between the measured and the predicted water contents, and the coefficient of determination (R²). For network A the RMSE varied for the three estimated water contents from 0.0264 to 0.0476 cm³ cm^–3, and R² varied from 0.80 to 0.93 for the individual model outputs. Networks B and C had an RMSE of 0.0435 and 0.0426 cm³ cm^–3, respectively. For both networks, R² was 0.89. The neural networks performed somewhat better than previous regression functions, but the improvements were not significant.     

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.