基于可见-近红外波段反射率估算表层土壤含水量

发布时间：2018-11-04 18:45

【摘要】：土壤含水量(θ)是影响植物生长和作物产量的主要因素之一。因此,有效和无损地监测土壤表层含水量对农业灌溉和作物生长具有重要意义。本研究在室内采用积分球法和光谱仪法获取四种土壤(砂土、壤土、黏壤土和砂质壤土)的含水量和光谱反射率数据,用图像法获取了这四种土壤的图像信息。其主要结果如下：第一,用光谱特征参数：绿边反射率总和(决定系数R2=0.24和均方根误差RMSE= 0.09 m3 m-3)和780-970 nm吸收深度(R2=0.31和RMSE=0.11 m3 m-3)、900-970 nm最大反射率与900-970 nm反射率总和(R2=0.92和RMSE=0.05 m3 m-3)、900-970 nm最大反射率(R2=0.86和RMSE=0.03 m3 m-3)与900-970nm积分(R2=0.85和RMSE=0.03 m3 m-3)及900-970nm最大反射率和900-970 nm反射率总和(R2=0.87和RMSE=0.02 m3 m-3)分别估算砂土、壤土、黏壤土和砂质壤土的土壤含水量估算精度最好。用900-970nm最大反射率(R2=0.48和RMSE=0.05 m3 m-3)与900-970nm反射率总和(R2=0.47和RMSE=0.05 m3 m-3)估算不同土壤质地的含水量相关性最高。采用人工神经网络(ANN)能够更好地提高四种土壤质地的土壤含水量的测量精度(R2=0.95和RMSE为0.03 m3m-3)。第二,用积分球法研究了不同土壤质地的土壤,在不同土壤容重(ρ)下和土壤含水量的土壤反射率的变化。结果表明,在不同土壤容重和土壤质地(砂土、壤土、黏壤土和砂质壤土),土壤反射率随着含水量的增加而降低,可用土壤反射率估算θ,其估算精度如下：砂土(R2=0.79和RMSE为0.05 m3m-3)、壤土(R2=0.91和RMSE为0.04 m3 m-3)、黏壤土(R2=0.88和RMSE为0.04 m3 m-3)和砂质壤土(Rw2=0.86和RMSE为0.04 m3 m-3)；估算四种土壤的含水量精度为R2=0.68和RMSE为0.07 m3m-3。第三,提取和分析了不同容重下土壤含水量(θ)的图像信息,包括亮度(Value,用V表示)、饱和度(Saturation,用S表示)和色调(Hue,用H表示),建立亮度与饱和度的估算模型。研究结果表明：1)对砂土,当其土壤容重为1.50 g cm-3时,估算模型相关性最好(R2=0.82和RMSE=0.05 m3m-3);对壤土而言,当其土壤容重为1.40 g cm-3时,估算模型相关性最好(R2=0.93和RMSE=0.04 m3m-3);对黏壤土而言,当其土壤容重为1.60 g cm-3时,相关性最好(R2=0.93和RMSE=0.03 m3 m-3);对砂质壤土,当其土壤容重为1.60 g cm-3时,相关性最好(R2=0.87和RMSE=0.01 m3 m-3); 2)考虑土壤容重的变化,黏壤土含水量估算模型θ=1.89-0.72×V-1.50×S(R2=0.72和6 m3m-3),壤土的R2=0.70和RMSE=0.07 m3 m-3,砂土的R2=0.70和RMSE=0.07 m3 m-3,砂质壤土的R2=0.60和RMSE=0.07 m3m-3。结合人工神经网络(ANN)能够更好地提高四种土壤质地的土壤含水量的测量精度,砂土估算模型的精度为R2=0.67和RMSE为0.07 m3m-3,壤土的则是R2=0.79和RMSE为0.07 m3m-3；黏壤土为R2=0.82并RMSE为0.07 m3m-3；砂质壤土的是R2=0.88和RMSE为0.05m3m-3。综上所述,本研究中采用了积分球法、光谱仪法和图像法获取了不同土壤容重下四种土壤的表层含水量的反射率和土壤颜色信息,基于土壤反射率和图像信息建立二者与θ之间的回归模型,估算θ;分析和比较了这三种方法估算θ模型的精度。结果表明,在不同的ρ和土壤质地(砂土、壤土、黏壤土和砂质壤土)估算θ的精度大小顺序如下：积分球法光谱仪法图像法。采用人工神经网络则能够提高土壤含水量的测量精度。
[Abstract]:Soil moisture content (soil moisture) is one of the main factors affecting plant growth and crop yield. Therefore, the effective and non-destructive monitoring of soil surface water content is of great significance to agricultural irrigation and crop growth. The water content and spectral reflectance data of four kinds of soil (sandy soil, loam, clay loam and sandy loam) were obtained by means of XRF and spectrometer, and the image information of these four kinds of soil was obtained by image method. The main results are as follows: First, spectral characteristic parameters: the sum of the reflectance of green side (coefficient R2 = 0.9m3 m-3) and the absorption depth of 780-970 nm (R2 = 0.9m3 m-3) and 780-970 nm absorption depth (R2 = 0. 31 and RMSE = 0. 11 m3 m-3), the maximum reflectance of 900-970 nm and the reflectivity of 900-970 nm (R2 = 0.992 and RMSE = 0. 05 m3 m-3). The maximum reflectivity of 900-970 nm (R2 = 0.886 and RMSE = 0. 03 m3 m-3) and 900-9600 nm integral (R2 = 0. 85 and RMSE = 0. 03 m3 m-3), and the total reflectance of 900-9600 nm and the reflectance of 900-970 nm (R2 = 0.987 and RMSE = 0.02m3 m-3) respectively estimate the soil moisture content estimation accuracy of sandy loam, loam, clay loam and sandy loam. The correlation of water content of different soil texture was estimated by the total reflectance (R2 = 0. 48 and RMSE = 0. 05 m3 m-3) and the total reflectance (R2 = 0. 47 and RMSE = 0. 05 m3 m-3). Artificial neural network (ANN) was used to improve soil moisture content (R2 = 0.995 and RMSE 0. 03 m3m-3). Secondly, soil with different soil texture was studied by TUNEL method, and the soil reflectance changes under different soil bulk density (soil volume) and soil moisture content were studied. The results show that the soil reflectance decreases with the increase of water content in different soil bulk density and soil texture (sandy soil, loam, clay loam and sandy loam). The estimation accuracy of soil reflectance is as follows: sandy soil (R2 = 0.979 and RMSE is 0. 05 m3m-3). loam (R2 = 0.991 and RMSE 0. 04 m3 m-3), clay loam (R2 = 0.988 and RMSE 0. 04 m3 m-3) and sandy loam (Rw2 = 0.886 and RMSE 0. 04 m3 m-3); The water content accuracy of the four soils was estimated to be R2 = 0.968 and RMSE 0. 07m3m-3. Thirdly, the image information of soil moisture content (LAI) under different bulk density is extracted and analyzed, including the brightness (Value, V representation), saturation (SSPS) and hue (Hue, H is denoted by H), and the estimation model of brightness and saturation is established. The results show that: 1) For sandy soil, when the bulk density of soil is 1. 50 g cm-3, the estimated model correlation is best (R2 = 0.882 and RMSE = 0. 05 m3m-3). For loam, when the soil bulk density is 1. 40 g cm-3, the estimated model correlation is best (R2 = 0.993 and RMSE = 0. 04 m3m-3); for clay loam, When the bulk density of soil is 1. 60 g cm-3, the correlation is best (R2 = 0.993 and RMSE = 0. 03 m3 m-3); for sandy loam, when its soil bulk density is 1. 60 g cm-3, the correlation is best (R2 = 0.987 and RMSE = 0. 01 m3 m-3); 2) Considering the change of soil bulk density, the estimation model of moisture content of clay loam is 1.89-0.72mv-1.50m-S (R2 = 0. 72 and 6 m3m-3). R2 = 0. 70 and RMSE = 0.07m3 m-3 in loam, R2 = 0.970 and RMSE = 0.07m3 m-3 in sandy loam, R2 = 0.960 and RMSE = 0.07m3m-3 in sandy loam. Artificial neural network (ANN) can improve soil moisture content of four kinds of soil texture better. The accuracy of sand estimation model is R2 = 0.967 and RMSE is 0.07m3m-3. The loam is R2 = 0.979 and RMSE is 0.07m3m-3; clay loam is R2 = 0.882 and RMSE is 0.07m3m-3; sandy loam is R2 = 0.888 and RMSE is 0. 05m3m-3. To sum up, the reflectance and soil color information of the surface water content of four kinds of soil under different soil bulk density were obtained by the method of TUNEL, spectrometer and image method, and the regression model between them and the soil was established on the basis of soil reflectance and image information. This paper analyzes and compares the three methods to estimate the accuracy of the model. The results show that the accuracy and the order of the soil texture (sandy soil, loam, clay loam and sandy loam) in different soil texture and soil texture (sandy soil, loam, clay loam and sandy loam) are as follows: The artificial neural network can improve the measurement precision of soil moisture content.
【学位授予单位】：中国农业大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：S152.7

【相似文献】