基于Vis-NIR光谱的不同水分状态下土壤有机质预测

Prediction of soil organic matter under different soil moisture conditions using Vis-NIR spectroscopy

以河南省封丘县的86个土壤样本为研究对象,测定9种不同含水量状态下的土壤光谱反射率,运用偏最小二乘回归（PLSR）建立不同含水量状态下的土壤有机质预测模型;运用0-50 g·kg^-1,200-250 g·kg^-1,400-450 g·kg^-1水分状态下3组模型进行交互预测,以研究含水量状态差异较大情况下,土壤含水量对有机质预测精度的影响。结果显示,建模样本与预测样本处于同一含水量状态下,土壤含水量对土壤有机质含量预测精度的影响不显著;当验证集样本与建模集样本间水分状态差异较大时,预测结果会出现较大误差。当土壤处于同一湿度状态时,可直接应用Vis-NIR光谱预测湿土的有机质;当土壤样本间水分差异较大时,可依据含水量状态建立分组模型,以提高Vis-NIR光谱预测有机质的精度。

Reflection spectra of 86 soil samples collected from Fengqiu County were obtained in the laboratory under nine moisture conditions, and the prediction models for each moisture conditions were built using Partial least-squares regression. Models under the soil moisture content 0 - 50g·kg^-1, 200 -250g·kg^-1 and 400 - 450g·kg^-1 were then applied to the other validation sets with different moisture conditions to explore the effect of different soil moisture on the prediction accuracy. The results show that the prediction models for each moisture condition perform well with acceptable prediction accuracy when applied it to the validation set under the same moisture conditions with the calibration set. However, the accuracy decreased dramatically with the increase of difference in soil moisture conditions between the built data set and prediction data set. Spectroscopy could be directly used to predict soil organic matter on moist samples with the similar soil moisture condition. Moist samples should be classified by soil moisture content, and then predict soil organic matter with separate prediction models under the same soil moisture condition.