基于高光谱特征的三峡库区紫色土有机质含量预测研究——以重庆市北碚区白鹤林为例

Prediction of Organic Matter Content in Purple Soil of Three Gorges Reservoir Area Based on Hyperspectral Characteristics:Taking Baihelin,Beibei District,Chongqing City as an Example

【目的】为快速有效测定土壤有机质含量,提高三峡库区农作物产量,实现农业可持续发展。【方法】以三峡库区广泛分布的紫色土为研究对象,对它的有机质含量和原始光谱反射率(R)进行测定,基于R进行C(R)、log_(10)(1/R)、R′、R″、[log_(10)(1/R)]″等5种光谱反射率形式变换,构建紫色土有机质含量的MLSR、PLSR和BPNN高光谱反演模型。【结果】1)紫色土有机质含量范围为7.68~31.49g·kg^(-1),变异系数为31.65%,属中等变异性质,总体上处于缺乏水平;2)有机质含量与R呈负相关,与log_(10)(1/R)则呈正相关关系,且不同光谱变换形式下的最佳显著性波段主要集中在534~889nm、1450~1976nm和2281~2328nm;除log_(10)(1/R)外,R的另外4种变换形式与有机质含量的相关性较R有显著提高,最大相关系数达0.676。3)对比MLSR、PLSR、BPNN等3种反演模型,C(R)处理的PLSR模型是预测紫色土有机质含量的最佳模型,建模集和验证集的决定系数分别为0.671和0.532,RMSE分别为2.99和4.03g·kg^(-1)。【结论】PLSR-C(R)模型可以较好地预测三峡库区紫色土有机质含量,为三峡库区紫色土肥力管理和农业速测紫色土有机质含量提供了新的参考。

[Purposes]It aims to explore how to rapidly predict soil organic matter(SOM) in purple soil to increase crop yields and achieve sustainable agricultural development in the Three Gorges Reservoir Area. [Methods]Taking purple soil which widely distributed in the Three Gorges Reservoir area as research area, the SOM content and original reflectance(R) of purple soil were measured. Then based on R to perform mathematical transformation(continuum-removal C(R)), reciprocal logarithm(log_(10)(1/R)), first-order differential(R′), second-order differential(R″), reciprocal-logarithmic second-order differential([log_(10)(1/R)]″)), and the prediction models of multiple linear stepwise regression(MLSR), partial least squares regression(PLSR) and back propagation neural network(BPNN) were established. [Findings]1)The distribution range of purple soil organic matter was 7.68 g·kg^(-1)~ 31.49 g·kg^(-1)and the coefficient of variation was 31.65%, which was of moderate variation and was generally at a deficient level. 2) The SOM was negatively associated with the R, but was positively associated with the log_(10)(1/R). After different mathematical transformation, the best significant bands mainly concentrated in 534~889 nm, 1 450~1 976 nm and 2 281~2 328 nm. Except for log_(10)(1/R), the correlation between SOM and other transformations was strengthened, the maximum correlation coefficient is 0.676. 3) PLSR model processed by C(R) is more suitable for predicting the SOM content of purple soil than the models of MLSR and BPNN(Modeling set: R;= 0.671, RMSE = 2.99;Validation set: R;= 0.532, RMSE = 4.03). [Conclusions]PLSR-C(R) model could better predict the purple soil organic matter content in Three Gorges Reservoir Area, that would provide a new reference for fertility management of purple soil in the Three Gorges Reservoir area and rapid measurement of SOM content in agriculture.