引入时相信息的耕地土壤有机质遥感反演模型

土壤有机质(soil organic matter,SOM)是土壤质量评价的重要指标。监测SOM含量及其空间分布对土壤利用与保护、土壤有机碳库估算等具有重要意义。该文以松嫩平原典型区为研究区,采集4种主要土壤类型样本共147个,获取裸土期多时相MODIS地表反射率8 d合成产品,以单期、多期影像所构建光谱指数作为输入量,构建包含含水量变化与有机质含量信息的多光谱指数,建立SOM线性回归遥感反演模型,揭示SOM空间分布规律。结果表明:由于土壤含水量空间差异随时间变化,基于单期影像构建的模型主要输入量发生规律性改变,其中年积日137 d裸土条件最好,反演模型最优;比值光谱指数R61与SOM显著相关,而和含水量相关性极小,适于作为反演模型输入量;基于多期影像构建的模型引入时相信息后,精度与稳定性较单期影像模型显著提高,其中基于年积日137、105 d两期影像光谱指数所建立的多元线性模型最优;松嫩平原SOM呈现由东北向西南递减趋势。

Remote sensing inversion model of soil organic matter in farmland by introducing temporal information

Abstract: Soil organic matter (SOM) is an important index to evaluate soil quality. The monitoring of SOM content and its spatial distribution is of great significance to soil utilization, soil conservation, estimation of soil organic carbon pool, and so on. It is difficult to estimate the reserves of soil organic carbon pool by using traditional methods. Early studies showed that there was a significant negative correlation between SOM and soil spectral reflectance, and there was a quantitative relationship between SOM and soil organic carbon, so we can estimate the SOM content by remote sensing and it will provide help to estimate the soil organic carbon pool. In this paper, 4 main soil types (black soil, chernozem soil, meadow soil and aeolian sand) were collected as sampling points in the typical area of Songnen Plain to construct prediction model for estimating the SOM content in the study area. The number of the soil sampling points was 147. Half of the sampling points (74 samples) were used to serve as calibration set and other sampling points (73 samples) were used to serve as validation set. The remote sensing inversion model was established to reveal the spatial distribution of SOM in the study area. To improve the accuracy and stability of inversion model and find the optimal model, spectral indices based on single MODIS image or multi MODIS images during the bare soil period, which could contain temporal information of the variation of soil moisture content and SOM content, were introduced into the multiple linear regression model. The results showed that the 8-day data of surface reflectance (MOD09A1 and MYD09A1) could be used to estimate the SOM content of different types of soils in Songnen Pain, Northeast China. The primary input variable of the model changed regularly, because the bare soil condition changed with the variation of date. Furthermore, because the condition of bare soil was the optimal, the prediction model based on the MODIS image on the 137th day in the year was the best among the models based on single MODIS image. The ratio of spectral index R61 (the ratio of Band 6 to Band 1) is significantly related to SOM, but it has little correlation with soil moisture, so it can eliminate the influence of soil moisture to a certain degree. R61 is suitable as a primary input variable for the inversion model to estimate SOM by using remote sensing method. The accuracy and stability of models based on multi MODIS images were generally better than the models based on single MODIS image. The model based on multi MODIS images on the 137th and 105th day in the year is the best among all models, and its R2 is 0.68, and its RMSE (root mean square error) is 0.84 for calibration set and 0.84 for validation set. The SOM content in Songnen Plain showed a decreasing trend from northeast to southwest. This study provides a rapid and nondestructive method to estimate SOM content in a large scale. Moreover, the results of remote sensing inversion give supports for soil degradation assessment, land use, and estimation of soil carbon pool.