顾及土壤湿度的土壤有机质高光谱预测模型传递研究

高光谱遥感技术作为当前遥感发展的前沿科技，通过电磁波与地物的相互作用，可以定量反演地物的物理化学性质。土壤有机质是重要的土壤养分信息参数，利用高光谱遥感技术快速获取其含量信息可以为精准农业的发展提供必要的数据支撑。然而，由于受到外部参数差异的干扰，导致建模精度降低的同时，还会造成已有模型传递性的“失效”。为了消除湿度差异的干扰，进一步拓展已有模型的适用空间，以江汉平原滨湖地区为例，通过对95个土壤样本进行加湿处理，在实验室自然风干的条件下，量测得到13套不同湿度等级土壤样本的可见—近红外反射光谱数据，建立了各湿度等级下土壤有机质的光谱反演模型，研究水分差异对建模精度的影响;在此基础上，运用Direct Standardization(DS)算法对湿土光谱进行校正，进而探讨该算法在提高模型传递性能方面的潜力。结果表明：基于风干土光谱建立的模型预测精度最高，未经校正的湿土光谱无法通过该模型进行土壤有机质含量预测，预测偏差在-8.34～3.32 g·kg-1，RPD在0.64～2.04;经过DS算法校正后的湿土光谱可以通过该模型进行土壤有机质含量预测，预测偏差降低至0，RPD值提高至7.01。研究表明DS算法能有效降低湿度差异对光谱反演土壤有机质的影响，使土壤有机质光谱反演模型适用于不同水分含量的土壤样本。

Transferability of Hyperspectral Model for Estimating Soil Organic Matter Concerned with Soil Moisture

Hyperspectral remote sensing, known as the state-of-the-art technology in the field of remote sensing, can be used to retrieve physical and chemical properties of surface objects based on the interactions between electromagnetic waves and the objects. Soil organic matter (SOM) is one of the most important parameters used in the assessment of soil fertility. Quick estimation of SOM with hyperspectral remote sensing technique can provide essential soil data to support the development of precision agriculture. The presence of external parameters, however, may affect the modeling precision, and further handicap the transferability of existing model. With the aim to study the effects of soil moisture on the Vis/NIR estimation of soil organic matter, and the capacity of direct standardization(DS)algorithm in the calibration transfer, 95 soil samples collected in the Jianghan plain were rewetted and air-dried. Reflectance of these samplesat 13 moisture levels was measured. Results show that the model calibrated using air-dried samples has the highest prediction accuracy. This model, however, was not suitable for SOM prediction of the rewetted samples. Prediction bias and RPD improved from -8.34～3.32 g·kg-1 and 0.64～2.04 to 0 and 7.01, when DS algorithm was applied to the spectra of the rewetted samples. DS algorithm has been proven to be effective in removing the effects of soil moisture on the Vis/NIR estimation of SOM, ensuring a transferrable model for SOM prediction with soil samples at different moisture levels.