紫色土光谱特征及其碱解氮的预测研究

文章分析探索了应用可见近红外光谱技术快速、高效、便捷测定土壤营养参数的可能性。采集蓬莱镇组紫色土样本，比对分析了不同肥力水平、土壤厚度和土壤粒径条件下采集土壤光谱对可见近红外光谱特征的影响，筛选出不同厚度、粒径土壤条件下的碱解氮含量预测模型。研究结果表明，土壤样本厚度为30mm时具有最大的光谱反射率，建立的氮含量预测模型效果最佳，校正集和验证集的相关系数分别为0.84和0.83，均方根误差分别为1.79和1.87。土样粒径在0.25-0.85mm时氮含量的预测效果最佳，校正集和验证集的相关系数均超过0.8，且均方根误差较小；但当土样粒径<0.25mm时，氮含量预测模型效果明显下降。采用20目（<0.85mm）过筛、30mm厚度土壤样本采集可见近红外光谱和偏最小二乘法（PLS）模型预测，可以实现对蓬莱镇组紫色土碱解氮含量的较好光谱预测。

Spectral Characterization of Purple Soil and Its Prediction of Available Nitrogen Content

This study aimed at discovering the possibility to quickly and efficiently predict the soil properties using visible near infrared (VNIR) spectroscopy. The purple soil samples were systematically collected from Chongqing. Different nutrition levels, thickness and particle sizes of purple soil samples were analysed to figure out the best thickness and particle size of purple soil sample in estimating its available Nitrogen (N) content based on partial least squares (PLS) method. The results show that, when the samples are 30 mm thick, the reflectance are the largest and have the best model in predicting the available N content with the correlation coefficients of calibration and validation sets being 0.84 and 0.83 respectively and root mean square error(RMSE) 1.79 and 1.87. When the particle size is between 0.25 and 0.85 mm, the prediction model is the best with the correlation coefficients of calibration and validation sets being over 0.8 and a small RMSE. But when the particle size is under 0.25 mm, the effect of prediction model decreases obviously. In short, we can achieve fast prediction of available N of Penglaizhen group purple soil with NIRS and PLS method when the sample is 30 mm thick and particle size is between 0.25 and 0.85 mm.