在农田尺度下, 土壤表层因理化性状、 粗糙度、 作物根系和秸秆残茬等引起的空间异质性较为明显, 样点间含水量差别相对较小, 这对基于近红外漫反射光谱技术的表层土壤含水量原位测定带来极大的挑战。 本研究分别利用基于单一波长(1 200, 1 400, 1 450, 1 820, 1 940, 2 000和2 250 nm)反射率构建的指数衰减模型、 基于归一化土壤湿度指数(NSMI)和相对吸收深度(RAD)构建的线性模型、 基于土壤含水量高斯模型(SMGM)所得的拐点宽度(σ)、 函数中心振幅(Rd)和高斯曲线面积(A)三个参数构建的线性或二次模型, 以及基于波长区间反射率构建的偏最小二乘模型(PLS), 对土壤体积含水量(VMC)进行定量分析。 结果表明: (1)在单一波长反射率构建的所有指数衰减模型中, 2 000 nm波长显示出最佳验证效果, RMSEp最低(2.463), RPD最大(1.06); (2)与RAD相比, NSMI的验证精度更令人满意, R2(0.312)和RPD(1.224)更高, RMSEp(2.133)更低; (3)在SMGM模型参数以及PLS模型拟合VMC的验证结果中, Rd具有最佳拟合精度, 其R2(0.253)和RPD(1.175)最高, RMSEp(2.222)最低; (4)总的来看, NSMI指标构建的线性模型是所有方法中精度最高的, 而且计算过程简单, 易于操作, 可作为表层土壤含水量原位测定的首选方法。

At field scale, surface soil had special characteristics of volumetric moisture content (VMC) with a relatively little difference and spatial heterogeneity induced by physical and chemical properties, roughness, straw residues, etc. It has been a great challenge for near infrared diffuse reflectance spectroscopy (NIR-DRS) measurement of surface soil moisture in situ. In this study, exonential decay models based on seven water-related wavelengths (1 200, 1 400, 1 450, 1 820, 1 940, 2 000 and 2 250 nm), linear models of normalized difference soil moisture index (NSMI) and relative absorption depth (RAD) based on wavelength combinations, linear or quadratic model of width of the inflection (σ), center amplitude of the function (Rd) and area under the Gaussian curve (A) from soil moisture Gaussian model (SMGM), and partial least square (PLS) regression models based on bands were used to quantify VMC. The results indicated that (1) of all the single wavelengths, 2 000 nm showed the best validation result, indicated by the lowest RMSEp (2.463) and the highest RPD value (1.060). (2) Comparing with RAD, the validation of NSMI was satisfactory with higher R2 (0.312), lower RMSEp (2.133) and higher RPD value (1.224). (3) In the validation results of SMGM parameters and PLS fitting, Rd was found to produce the best fitting quality identified by the highest R2 (0.253), the lowest RMSEp (2.222), and the highest RPD value (1.175). (4) Comprehensively, a linear model based on NSMI showed the highest validation accuracy of all the methods. What is more, its calculation process is simple and easy to operate, and therefore become the preferred method to quantify surface soil moisture content in situ.