基于土壤光谱特征的宁夏银北地区盐碱地盐分预测研究

Prediction of Soil Salt Content Based on Spectral Characteristics of Soil in Northern Yinchuan City,Ningxia Hui Autonomous Region

对宁夏回族自治区银北地区盐碱地野外土壤表层光谱反射率和土壤全盐及盐分进行定量分析,筛选出各土壤盐分指标的敏感波段,然后采用全回归和逐步回归的方法建立各盐分的预测模型。结果表明:表层土壤高光谱反射率r,及其平滑去嗓处理后的值R,lg(R)与全盐含量呈极显著正相关关系,1/R,lg(1/R)与全盐呈极显著负相关关系,(R)′和〔lg(1/R)〕′在特定单波段处表现较佳;土壤表层光谱反射率与CO2-3的相关性最强,其次是SO2-4;土壤光谱反射率与Na+的相关性在各种变换方法下均较强,其次为Mg2+,与Ca2+的相关性最弱。基于R的逐步回归方程为全盐含量预测的最佳模型;基于土壤光谱反射率拟合土壤CO2-3的准确度略高于对土壤HCO-3;敏感波段估测土壤SO2-4含量的决定系数明显高于其他阴离子;采用〔lg(1/R)〕′逐步回归得到的方程拟合土壤Na+,K+和Mg2+含量相对于其他变换方式效果更理想。预测模型中对土壤全盐和Na+的模型精度较高,预测能力强;光谱对土壤SO2-4和Mg2+的预测能力也较强;对土壤Cl-和Ca2+的预测稳定性、预测能力和精度都较差。

The field soil surface spectral reflectance, total soil salt and other salt parameters in northern Yin chuan City, Ningxia Hui Autonomous Region were quantitatively analyzed. The field reflectance data were transformed to several spectral indices to extract sensitive wavelengths of salt parameters in surface soil. Quantitative inversion models of soil salt parameters were constructed by total regression and stepwise multi ple linear regression analysis. Results showed that there were significantly positive correlations between the total salt content in surface soil and its original spectral reflectance（r）, transformation of smoothing reflec tance（R） and logarithmic reflectancelg（R）. There were significant negative correlations between the total salt content and the reciprocal of reflectance （l/R） and logarithmic reciprocal of reflectancelg （l/R） . The first derivate differential（R＇） and the first derivate differential of logarithmic reciprocal of reflectance[lg（1/ R） P] had better effect in some specific single wavelengths. The correlation between the spectral reflectance of surface soil and CO was the strongest in all anions and SO, next; spectral reflectance and Na＋ content had strongest relationship by the four kinds of transformation method; the next was Mg2＋ ; and the correla tion with Ca2＋ was weakest. The R based regression equation was the optimal model for prediction of the to tal salt content. The accuracy of CO content predicted was slightly better than HCOy. The determinative coefficient for SO predicted based on the sensitive wavelengths was significantly higher than other anions.The stepwise regression by using[lg （l/R）]＇ gave better ef compared with other transformations. Fitting degrees of pre higher in all models and t dictive ability of spectral formance on stability, fo fect in fitting Na＋, diction model on the he two models had higher accuracy and strong predictive K＋ and Mg2＋ contents, as soil total salt and Na＋ ability. Moreover, the reflectance for SO42 and Mg2＋ were stronger than other ions. There were poor recast ability and the precision of the prediction models about C1 and Ca2＋ were pre per