光谱学与光谱分析, 2016, 36 (6): 1848, 网络出版: 2016-12-20
基于土壤植被光谱协同分析的土壤盐度推理模型构建研究
Soil Salinity Modelling Study with Salinity Inference Model Based on the Integration of Soil and Vegetation Spectrum in Arid Land
土壤盐分 植被指数 推理模型 线性混合像元分解模型 Soil salinity Vegetation index Inference model The linear spectral unmixing model
摘要
土壤组成较为复杂, 单纯利用土壤光谱信息探测土壤盐度, 反演精度不足以满足实际需求。 通过遥感获取的植被信息(植被类型和生长状况)可间接反映土壤盐分的空间分布特性, 弥补上述不足。 为此, 基于干旱区土壤盐度与植被之间的协同变化, 尝试结合土壤和植被光谱信息, 借助二维特征空间理论, 构建土壤盐度推理模型, 提高土壤盐度推理精度。 对于干旱区单个像元下土壤光谱的影响, 利用归一化植被指数(normalized difference vegetation index, NDVI) 难以准确反演干旱区稀疏植被参数。 因此, 首先利用线性混合像元分解模型(linear spectral unmixing model, LSUM)提取研究区地物组分, 构建植被组分指示因子(combined vegetation indicative factor, CVIF)方程, 并与土壤盐度指数(salinity index, SI)构建二维特征空间。 分析二维特征空间内散点走势与土壤盐分之间的关系, 建立土壤盐度推理模型(salinity inference model, SID)。 验证结果显示, CVIF提取的植被信息精度(R2>0.84, RMSE=3.92)高于应用较为广泛的NDVI(R2>0.66, RMSE=13.77)。 构建的SID模型与前人基于NDVI建立的联合光谱指数(combined spectral response index, CORSI)相比, 前者(R2>0.86, RMSE=6.86)推理精度优于后者(R2>0.71, RMSE=16.21)。 由此得出结论, 基于土壤和植被光谱信息双重判定的SID模型对土壤盐渍化的高精度遥感监测研究具有较好促进作用。
Abstract
Only using soil spectrum to model soil salinity is not enough to meet the actual demands because of the complicated soil context. As a remotely sensed indicator, the vegetation type and its growing condition can provide a spatial overview of salinity distribution. Based on the synergistic relationship between soil salinity and vegetation in arid land, this paper tries to combine the spectrum of soil and vegetation to quantitatively estimate the salt content with the help of the concept of two-dimensional feature space. After the analysis of scatter diagram, the soil salinity detecting model was constructed to improve reasoning precision. However, because the impact of soil reflectance on the quantification of vegetation parameters under the individual pixel, the Normalized Difference Vegetation Index (NDVI) was difficult to accurately obtain sparse vegetation cover in arid areas. Therefore, in order to avoid the limitations of NDVI, the Combined Vegetation Indicative Factor(CVIF)was created and supported by Linear Spectral Unmixing Model (LSUM). Then, the study constructed the feature space based on the CVIF and salinity index (SI) and analyzed the response relationship between soil salinity and the trend of scattered points. Finally, a new and operational model termed Salinity Inference Model (SID) was developed. The results showed that the CVIF (R2>0.84, RMSE=3.92) performed better than NDVI(R2>0.66, RMSE=13.77), which means the CVIF was more appropriate for analyzing variations in vegetation cover (particularly halophytes) than NDVI in the study area. The SID was then compared to the Combined Cpectral Response Index (COSRI)(NDVI-based) from field measurements with respect to the soil salt content. The results indicated that the SID values are highly correlated with soil salinity, in contrast to the performance of COSRI. Strong exponential relationships were observed between soil salinity and SID (R2>0.86, RMSE<6.86) compared to COSRI (R2=0.71, RMSE=16.21). These results suggested that the feature space related to biophysical properties combined with CVIF and SI can effectively provide information on soil salinity.
王飞, 丁建丽. 基于土壤植被光谱协同分析的土壤盐度推理模型构建研究[J]. 光谱学与光谱分析, 2016, 36(6): 1848. WANG Fei, DING Jian-li. Soil Salinity Modelling Study with Salinity Inference Model Based on the Integration of Soil and Vegetation Spectrum in Arid Land[J]. Spectroscopy and Spectral Analysis, 2016, 36(6): 1848.