基于光谱特征空间的农田植被区土壤湿度遥感监测

土壤湿度遥感动态监测在农业生产中具有重要作用。近年来,多种基于光谱特征空间的土壤水分监测指数被陆续提出,并得到广泛关注和应用,但当前多数监测指数未考虑混合像元的影响。该文针对垂直干旱指数(perpendicular drought index,PDI)在农田植被覆盖区监测精度降低问题,分析了植被覆盖下的PDI误差分布规律,引入垂直植被指数(perpendicular vegetation index,PVI)作为植被覆盖表征量,在PVI-PDI二维空间对PDI模型进行调整,提出了适于植被覆盖的植被调整垂直干旱指数(vegetation adjusted perpendicular drought index,VAPDI),并利用内蒙古明安镇研究区实测土壤湿度数据,对PDI与VAPDI进行了比较和验证。结果表明,在裸土、麦茬、土豆、豇豆4种植被覆盖类型中,PDI与土壤实测含水率的决定系数分别为0.630、0.504、0.571、0.543,VAPDI与土壤实测含水率的决定系数分别为0.599、0.523、0.602、0.585。VAPDI在植被区的误差略小于PDI,一定程度上克服了植被覆盖对监测精度的影响。通过PDI和VAPDI空间分布图的比较也说明,VAPDI对土壤湿度的差别有更好的区分能力,在中尺度土壤表层水分遥感反演方面具有一定的优势。该研究可为农田土壤湿度遥感监测方法选择及监测误差分析提供参考依据。

Soil moisture monitoring of vegetative area in farmland by remote sensing based on spectral feature space

Abstract: Dynamic monitoring of soil moisture by remote sensing can play a significant role in agricultural production, weather research, and ecological environment evaluation. In recent years, continuous attention has been focused on the thought that soil moisture information can be extracted from multi-dimensional spectral feature space. A variety of soil moisture monitoring indices have been put forward in succession based on the spectral feature space, and some of them have gotten widespread attention and application such as PDI (perpendicular drought index), an simple, effective, and feasible index which has been developed based on red and near infrared spectral feature space and applied in many areas. However, the majority of these indices are devised without considering the influence of mixed pixels, which makes the effect of soil moisture estimation worse in vegetation-covered areas than in bare areas. Aiming at the problem of accuracy degradation for PDI monitoring in vegetation-covered areas, this study thoroughly analyzed the distribution of vegetation-soil pixels in Nir-Red spectral feature space, and figured out the distribution characteristics of PDI monitoring error under vegetation coverage. By taking PVI (perpendicular vegetable index) as vegetation coverage characterization, we tried to improve the PDI model in the PVI-PDI two-dimensional feature space and developed a new index (vegetation adjusted perpendicular drought index, VAPDI) theoretically suitable for vegetation coverage. Then, we used the measured soil moisture data in the study area in the Mingan Town, Bayinnaoer city of the Inner Mongolia autonomous region, China to compare and validate the PDI and VAPDI. Results showed that the coefficient of determination between the PDI and measured soil moisture were 0.630, 0.504, 0.571 and 0.543 respectively in 4 vegetation cover types (bare soil, wheat stubble, potato, and cowpea), and the coefficient of determination between the VAPDI and measured data were 0.599, 0.523, 0.602, and 0.585, respectively. The accuracy of the VAPDI was higher than the PDI in vegetation areas, and the advantage of VAPDI on monitoring accuracy increased gradually with the increase of vegetation coverage degree. Therefore, the overall effect of the VAPDI was better than that of the PDI in farmland vegetation areas. The comparison of space distribution of the PDI and VAPDI also indicated that the VAPDI does a better job in distinguishing the differences of soil moisture. The VAPDI index with clear physical meaning is simple and practical, which implies a good application prospect in the aspect of the mesoscale soil moisture inversion by remote sensing. The research can provide valuable information for method selection and error analysis in farmland soil moisture monitoring by remote sensing.