摘要
土壤有机碳含量是准确评价土壤肥力的重要依据,也是影响植被生长的重要因素。本研究以南疆阿瓦提、温宿、和田、新和的农田土壤为研究对象,利用所采集的261个土样的室内有机碳含量和光谱测定数据,经多项式(savizkg golag, SG)平滑、归一化、线性基线校正、标准正态变量变化(standard normal variation, SNV)、多项式平滑+归一化预处理后,分别采用偏最小二乘法(partial least squares regression, PLSR)、主成分回归法(principal component regression, PCR)估算了全波段土壤有机碳含量。研究结果表明:采用不同数据预处理和建模方法,所建模型精度有明显差异,在所选取的5种预处理方法中,经SG平滑+归一化处理后的PLSR模型精度最高,其R2为0.84,RMSE为3.47 g·kg-1,RPD为2.63;经SNV处理后的PCR模型精度也较高,其R2为0.79,RMSE为3.90 g·kg-1,RPD为2.30。根据评价标准,这两种模型都具有较好的预测能力,相比较而言,基于SG平滑+归一化处理后的PLSR模型较基于SNV处理后的PCR模型结果更好,其R2更大,RMSE更小,RPD也更高。由此可见,针对相同的数据预处理方法,PLSR模型较PCR模型更适用于估算土壤有机碳的含量,模型结果更优。因此,采用PLSR模型具有较优的预测效果,可推荐为农田土壤有机碳含量的估算。
Soil organic carbon content is an important basis for accurate evaluation of soil fertility, and it is also an important factor affecting vegetation growth. This study takes farmland soils in Awati, Wensu, Hotan, and Xinhe in southern Xinjiang as the research object. The indoor organic carbon content and spectral measurement data of 261 soil samples collected were used for polynomial savizkg golag(SG)smoothing, normalization, and linearization. After baseline correction, standard normal variation(SNV), smoothing + normalization pretreatment, partial least squares regression(PLSR)and principal component regression(PCR)were used to estimate the soil organic carbon content in the whole wave band.The research results showed that the accuracy of the model built by using different data preprocessing and modeling methods is significantly different. Among the five preprocessing methods selected, SG smoothing +normalized PLSR model had the highest accuracy, with an R2 of 0.84, an RMSE of 3.47 g·kg-1, and an RPD of 2.63;the PCR model after the SNV had a higher accuracy, and its R2 was 0.79, RMSE was 3.90 g·kg-1, and RPD was 2.30. According to the evaluation criteria, these two models both had better predictive ability. In comparison, the former model had the better results, with larger R2, smaller RMSE, and higher RPD. It can be seen that for the same data preprocessing method, the PLSR model is more suitable for estimating the content of soil organic carbon than the PCR model, and the model results are better. Therefore, the PLSR model has a better prediction effect and can be recommended as the estimation of the organic carbon content of farmland soil.
作者
张丽
郝梦洁
鲁新新
郭又波
阿迪力·亚森
蒋青松
Zhang Li;Hao Mengjie;Lu Xinxin;Guo Youbo;Yasen·Adili;Jiang Qingsong(College of Information Engineering,Tarim University,Alar,Xinjiang 843300)
出处
《塔里木大学学报》
2020年第4期49-58,共10页
Journal of Tarim University
基金
中央高校专项基金中农-塔大联合项目“基于近地传感-模型耦合的南疆农田土壤盐分监测与三维制图”(ZNLH201904)
国家自然科学基金项目“盐分对南疆土壤有机质高光谱特征与定量反演的影响及方法研究”(41361048)。
关键词
光谱数据
土壤有机碳
南疆农田
估算模型
spectral data
soil organic carbon
farmland in southern Xinjiang
estimation model
