基于BP神经网络的土壤水力学参数预测

为了获取区域土壤水分和溶质运移模拟所需的土壤水力学参数,利用黄淮海平原曲周县的试验资料建立基于BP神经网络的土壤转换函数模型。本文采用土壤粒径分布、容重、有机质含量等土壤基本理化性质,来预测土壤饱和导水率Ks、饱和含水量sθ、残余含水量θr、以及van Genuchten公式参数α、n的对数形式ln(α)和ln(n),并与多元线性逐步回归方法进行比较。t检验结果表明,BP神经网络训练和预测得到的模拟值与实测值之间吻合很好,该方法具有较高的预测精度。通过对平均相对误差的比较,得出在粒径分布的基础上增加容重、有机质含量等输入项目,可以提高部分土壤水力学参数的预测精度,而有些参数的预测精度反而降低。以误差平方和为标准的比较结果表明,BP神经网络模型的预测效果总的来看要优于多元线性回归法。     

土壤硝态氮含量检测和分布图生成中的最优样本量

土壤参数的时空变异是实施精细农业时要考虑的重要因素,在土壤检测的栅格采样中有必要确定最优样本量。本研究的试验田是一块生长中的玉米地块,试验区的面积为4.2 m×4.2 m,该试验区被假定为土壤采样中的一个栅格,该栅格又被细分为49个0.6 m×0.6 m的子栅格。采样时,所分析的土壤参数为土壤硝态氮含量,从播种到收获共进行了7次采样。通过对土壤样本土壤硝态氮时空变异的分析,揭示了样本量和土壤硝态氮含量预测误差之间的相关关系。土壤硝态氮含量呈非正态分布,通过对玉米各个生长期获得的数据分析表明：含量水平的预测误差随深度的增加而增大;当从一个栅格只采集1个土壤样本时(样本量为1),预测误差基本在50％左右(显著水平：α≈0.10),而当从一个栅格采集5个土壤样本时(样本量为5),预测误差将降至25％左右。另一方面,当要求预测误差低于30％时,对于普通生长条件下的土壤需要从1个栅格至少采集3个样本,而对于追肥后的土壤则至少需采集15个样本。     

基于支持向量机的土壤水力学参数预测

为了分析支持向量机在土壤水力学参数预测方面的效果,应用支持向量机构建用于预测土壤水力学参数的土壤传递函数,以土壤粒径分布、容重、有机质含量等土壤理化性质为输入项,分别预测土壤饱和导水率、饱和含水率、残余含水率,以及van Genuchten公式参数的对数形式。结果表明预测值和实测值不存在显著性差异,用支持向量机预测土壤水力学参数是可行的。不同输入项处理的预测分析表明,输入项为粒径分布、粒径分布和容重、粒径分布和有机质含量3种情况的预测效果差异不明显,而输入项为粒径分布、容重和有机质含量时预测效果优于前3种情况。支持向量机在预测土壤水力学参数方面的效果要优于多元线性逐步回归模型,而与BP神经网络模型相比不具有明显好的预测效果。     

土壤有机质可见光-近红外光谱预测样本优化选择

土壤有机质可见光-近红外光谱预测中建模样本的优化选择对提高有机质模型估算精度具有重要作用。本文以湟水流域土壤有机质为例,采用基于土壤单一属性信息考虑的建模样本选择方法:浓度梯度法、Kennard-Stone(KS)方法,以及基于土壤多种信息考虑的建模样本选择方法:Rank-KS(RKS)法、土壤类型结合浓度梯度法以及土壤类型结合KS法。通过偏最小二乘回归建模,探索可见光–近红外光谱预测青海湟水流域有机质的最优样本集。结果表明:不同级别样本数的最佳建模样本选择方法不同,整体表现为基于土壤多种信息挑选的建模样本集的模型精度相比土壤单一信息均较高,特别是KS方法结合土壤类型后的建模样本集模型精度明显提高且在样本数较少时更为明显。土壤类型可以优化建模样本选择方法提高模型预测精度。在保证固定验证样本模型预测精度的情况下,土壤类型参与建模样本的选择可以有效减少建模样本数,进而降低了建模成本。     

土壤有机质含量对土壤水动力学参数的影响

通过测定并分析不同有机质含量的壤质土样的饱和导水率,水分特征曲线,水分扩散率及几个水分常数,阐明了土壤有机质含量与水动力学参数的关系,从动力学角度探讨了有机质影响水分运动的机理。     

土壤颗粒分布参数模型对黄土性土壤的适应性研究

土壤颗粒组成是土壤最基本的物理性质之一,其分布曲线可用来估算土壤的水力学性质,然而对于土壤颗粒分布曲线的完整表达需要借助于参数模型,对于不同类型的土壤,参数模型的拟合效果不尽相同.为了选择能够较好描述黄土性土壤颗粒分布状况的参数模型,该文采用了3个指标--相关系数(R),均方根误差(RMSE)和Akaike信息准则(AIC)值,对3类共10个参数模型(单参数模型2个,二参数模型6个,三参数模型2个)在黄土性土壤上的适应性进行了评价(共828个土壤颗粒分析资料).结果表明:简化的三参数Fredlund模型对黄土性土壤颗粒分布的拟合效果最好,且受质地影响较小,二参数Weibull模型次之,单参数的Jaky模型效果最差.三参数Fredlund模型是估算黄土性土壤颗粒组成的最适宜的模型.     

Green-Ampt模型参数简化及与土壤物理参数的关系

简化模型表达形式从而减少参数个数,对于Green-Ampt入渗模型的实际应用具有重要的现实意义。该文通过推导湿润锋处平均基质吸力与Philip模型中土壤吸湿率关系基础上提出了简化的Green-Ampt入渗模型,基于新疆222兵团两块壤质土壤田块上土壤水分入渗试验资料,分析了Green-Ampt简化入渗模型参数与土壤物理参数之间的关系,建立了模型参数与土壤物理参数之间的定量经验转换函数。结果表明,入渗参数A(组合参数)与土壤初始含水率呈对数负相关,相关系数为0.77,A与土壤紧实度和黏粒含量均呈指数负相关,相关系数分别为0.70和0.74。饱和导水率Ks与土壤紧实度和黏粒呈指数负相关,相关系数分别为0.74和0.73。A和Ks与土壤初始含水率、土壤紧实度和黏粒含量呈高度和中度多元线性相关,相关系数分别为0.9和0.79。研究表明Green-Ampt简化入渗模型能够在一定精度下分析土壤入渗过程。     

基于土壤粒度参数的荒漠草原地表粗粒化过程

草原退化特征已由植被变化为主演变为土壤退化的阶段,其植被覆盖变化可在一定程度上反映退化状况与过程,探明荒漠草原不同退化阶段(覆盖度)地表风蚀状况,对合理利用与保护草地资源意义重大。以希拉穆仁荒漠草原7个覆盖度(裸地,5%,20%,40%,60%,80%和100%)下的表层土壤为研究对象,利用激光衍射技术测量表层0—1cm土壤粒度组成,分析不同覆盖度下的平均粒径、标准偏差、偏度、峰态及分形维数等粒度参数变化情况及颗粒频率分布。结果表明:(1)土壤粒度组成均以沙粒和粉粒为主,黏粒含量较低;偏度可作为有效粒度参数指标。(2)随着盖度降低,平均粒径数值越大,分选性越好,频率曲线由近于对称—正偏—极负偏转变,峰态由尖锐变平缓,分形维数先增加后降低。(3)粒径分布基本呈双峰分布,盖度越小第二波峰滞后越明显,且分布均匀程度越低,易风蚀颗粒范围为144~869μm。     

基于光谱分析的土壤游离铁预测研究

土壤游离铁含量的高低可作为土壤系统分类中判断土壤类型的诊断指标,同时也对土壤风化程度具有指示作用,并在一定程度上反映了土壤的成土环境。本研究调查了安徽宣城的91个土壤剖面,共398个土壤样品,采集了样品在350~2500 nm波段的漫反射光谱数据,并对游离铁含量进行化学分析。光谱数据包括反射率(R)、反射率一阶导数(FDR)和吸收度(Log(1/R))3种形式。本文采用偏最小二乘回归算法(PLSR)和反向神经网络(BPNN)建模预测游离铁含量,并分析不同形式光谱数据的建模预测效果。结果表明:当存在游离铁20 g/kg的样本时,传统建模方法不能准确预测游离铁含量(R20.6,RPD1.5),相对R和Log(1/R)两种光谱数据,以FDR作为自变量建模预测游离铁含量的效果最差。     

Utilizing Existing Sensor Technology to Predict Spring Wheat Grain Nitrogen Concentration

Optimum grain nitrogen (N) concentration and yield in spring wheat (Triticum aestivum L.) can be problematic without proper N fertilizer management. Sensor-based technologies have been used for application of fertilizers and also to predict yield in wheat, although little has been done in the prediction of grain N. Field studies were conducted in South Dakota in 2006 (Gettysburg, Bath, and Cresbard) and 2007 (Gettysburg, Aurora, Leola, and Artas). There were five N treatments (0, 56, 112, 168, and 224 kg N ha^?1) applied pre-plant with a second N application applied foliar at anthesis. Sensor readings were taken at growth stages Feekes 10, anthesis, and postfoliar application using the GreenSeeker Hand Held optical sensor. Grain samples were taken at maturity and analyzed for total N. Using similar information collected in 2003 and 2005, a critical normalized difference vegetation index (NDVI) value was determined using the Cate–Nelson procedure. The critical NDVI value needed to ensure optimum grain N was 0.70. In 2006 and 2007, the plots that received an application of N at anthesis had higher grain N than the plots not receiving N. There was also a significant response between applied N and grain yield. The results show that with further studies, the Greenseeker could be used to apply N to maximize yield and grain N in a precise and accurate manner.     

Influence of Soil Particle Size on the Measurement of Sodium by Near-Infrared Reflectance Spectroscopy

This study evaluates the effect of soil particle size (SPS) on the measurement of exchangeable sodium (Na) (EXC-Na) by near-infrared reflectance (NIR) spectroscopy. Three hundred thirty-two (n = 332) top soil samples (0–10 cm) were taken from different locations across Uruguay, analyzed by EXC-Na using emission spectrometry, and scanned in reflectance using a NIR spectrophotometer (1100–2500 nm). Partial least squares (PLS) and principal component regression (PCR) models between reference chemical data and NIR data were developed using cross validation (leaving one out). The coefficient of determination in calibration (R²) and the root mean square of the standard error of cross validation (RMSECV) for EXC-Na concentration were 0.44 (RMSECV: 0.12 mg kg^–1) for soil with small particle size (SPS-0.053) and 0.77 (RMSECV: 0.09 mg kg^–1) for soils with particle sizes greater than 0.212 mm (SPS-0.212), using the NIR region after second derivative as mathematical transformation. The R² and RMSECV for EXC-Na concentration using PCR were 0.54 (RMSECV: 0.07 mg kg^–1) and 0.80 (RMSECV: 0.03 mg kg^–1) for SPS-0.053 and SPS-0.212 samples, respectively.     

MAIZE GRAIN YIELD RESPONSE TO VARIABLE ROW NITROGEN FERTILIZATION

Crop yields are affected by the rate and method of nitrogen (N) fertilizer application. This study was conducted to determine the effects of applying variable N rates by row on maize grain yields. The effects of variable rates and row application were investigated at the R.L. Westerman Irrigation Research Facility near Stillwater, Oklahoma on a Port-Oscar silt loam (fine-silty, mixed, super active, thermic Cumulic Haplustolls) and at Hennessey, Oklahoma on a Bethany silt loam (fine, mixed, thermic Pachic Paleustolls). For 2005 that was characterized by high yields, significant yield differences occurred in non-fertilized rows adjacent to N (67, 100, 134 kg N ha^?1) fertilized rows, but not when adjacent to low N [34 and 67 (some cases) kg N ha^?1]. In 2006, which had a dry growing season, grain yields were significantly lower than those produced in 2005. With few exceptions, rows receiving N did not produce significantly higher yields in 2006. In 2007, trends were similar to those observed in 2005. Excluding 2006, all site-years showed a significant reduction in yield when N fertilizer was not applied to each row. In order to maximize corn grain yields, N fertilizer should be applied by row, while alternate row N application should be avoided.     

RAPID DIAGNOSIS OF NUTRIENT ELEMENTS IN FINGERED CITRON LEAF USING NEAR INFRARED REFLECTANCE SPECTROSCOPY

In order to provide references for leaf nutrition diagnosis of fingered citron, the technique of near infrared reflectance spectroscopy (NIRS) was introduced to analyze nitrogen (N), phosphorus (P), potassium (K), iron (Fe), manganese (Mn), zinc (Zn), and copper (Cu) in the dry-leaf samples of fingered citron. The best calibration model for N was developed with high RSQ_CAL (0.90), SD/SECV (2.73) and low SEC (1.06 mg g^?1), good calibration models were obtained for P, K, Fe and Mn, and no significant correlations were found between the spectra and the individual amounts of Zn and Cu. When tested using a validation set (n = 38), N was well predicted with low values of SEP (1.21 mg g^?1) and high RPD (2.5). The values of SEP and RPD were also acceptable for the external validation of P, Fe and Mn. Near-infrared spectroscopy analysis technique shows potential of diagnosing minerals in fingered citron, particularly for N, P, Fe and Mn.     

低空遥感技术及其在精准农业中的应用

以中国农业科学院土壤肥料研究所所进行的低空遥感为例,系统描述了低空遥感的技术体系、硬件设备的工作原理和影像处理过程。介绍了低空遥感技术在精准农业中的应用情况,如地块边界数字化、地块面积量算、作物种类识别、作物长势分析等。同时分析了低空遥感技术的应用前景。     

基于VNIR和机器学习算法的原状土剖面Cu含量预测

快速测量土壤剖面重金属含量是评估土壤重金属污染状况并选择相应修复技术的关键。为了探讨可见光-近红外光谱法（Visible and Near-Infrared Reflectance Spectroscopy,VNIR）预测原状土壤剖面重金属含量的潜力,以江西省两个典型工矿厂周边农田土壤为研究对象,共采集了19个深度约100 cm的完整土壤剖面样品,分别测定土壤剖面样品的VNIR数据及其Cu含量。采用偏最小二乘回归法（Partial Least Squares Regression,PLSR）、Cubist混合线性回归决策树（Cubist Regression Tree,Cubist）、高斯过程回归（Gaussian Process Regression,GPR）和支持向量机（Support Vector Machine Regression,SVM）方法研究不同光谱预处理方法对土壤Cu含量预测精度的影响。结果显示,Cubist、GPR和SVM这三种机器学习算法的预测精度普遍高于PLSR,其中一阶导数（First-Order Derivative,FD）预处理的SVM模型预测精度最高（R2=0.95,均方根误差为7.94 mg/kg,相对分析误差为4.34）。这表明利用VNIR和机器学习可以对原状土壤剖面Cu含量进行有效预测,为快速监测Cu及其他重金属含量的相关研究提供参考。     

Comparing local vs. global visible and near-infrared (VisNIR) diffuse reflectance spectroscopy (DRS) calibrations for the prediction of soil clay, organic C and inorganic C

Local, field-scale, VisNIR-DRS soil calibrations generally yield the most accurate predictions but require a substantial number of local calibration samples at every application site. Global to regional calibrations are more economically efficient, but don't provide sufficient accuracy for many applications. In this study, we quantified the value of augmenting a large global spectral library with relatively few local calibration samples for VisNIR-DRS predictions of soil clay content (clay), organic carbon content (SOC), and inorganic carbon content (IC). VisNIR models were constructed with boosted regression trees employing global, local + global, and local spectral data, using local samples from two low-relief, sedimentary bedrock controlled, semiarid grassland sites, and one granitic, montane, subalpine forest site, in Montana, USA. The local + global calibration yielded the most accurate SOC predictions for all three sites [Standard Error of Prediction (SEP) = 3.8, 6.7, and 26.2 g kg^− 1]. This was similarly true for clay (SEP = 95.3 and 102.5 g kg^− 1) and IC (SEP = 5.5 and 6.0 g kg^− 1) predictions at the two semiarid grassland sites. A purely local calibration produced the best validation results for soil clay content at the subalpine forest site (SEP = 49.2 g kg^− 1), which also had the largest number of local calibration samples (N = 210). Using only samples from calcareous soils in the global spectral library combined with local samples produced the best SOC and IC results at the more arid of the two semiarid sites. Global samples alone never achieved more accurate predictions than the best local + global calibrations. For the temperate soils used in this study, the augmentation of a large global spectral library with relatively few local samples generally improved the prediction of soil clay, SOC, and IC relative to global or local samples alone.