基于Landsat8的清河水库总氮、总磷反演研究

发布时间：2018-09-07 13:56

【摘要】：本文研究目的是建立反演清河水库总氮、总磷含量的最适模型。以清河水库为研究区域,总氮总磷的含量为研究对象,分别建立了单波段遥感模型、波段组合遥感模型、最小二乘支持向量机模型,研究分析三种模型预测值与实测值的相对误差,得到最适用于清河水库总氮、总磷含量反演的模型,即最小二乘支持向量机模型。研究内容包括,实测数据的实验室分析;利用spss软件对波段值与实测值进行Pearson相关性分析并得到敏感波段及波段组合;利用敏感波段及波段组合分别建立单波段遥感模型、波段组合遥感模型、最小二乘支持向量机模型,最后对三种模型的适应性进行分析研究。本文研究重点是最小二乘支持向量机模型的建立及对三种模型精准度的分析对比。研究结果如下:(1)所研究数据共80组,分别在2015年6月、7月、9月、10月采集,6月份总氮含量平均值最高,为1.73mg/L,9月份总氮含量平均值最小,为1.11mg/L;6月份总磷含量平均值最高,为0.07mg/L,其他三个月总磷含量的平均值相同,均为0.04mg/L。(2)Landsat8 OLI数据中与总磷相关性最好的单波段,夏季6、7月份与秋季9、10月份均为B4波段,与总氮相关性最好的单波段,两个时段均为B5波段;波段组合中,夏季6、7月份与总磷相关性最好的为B4/B3,与总氮相关性最好的为B3/B4,秋季9、10月份与总磷相关性最好的为B4/B5,与总氮相关性最好的为B3+B4。(3)以最佳单波段或波段组合值为自变量,清河水库总氮、总磷含量实测值为因变量分别建立的单波段模型、波段组合模型、最小二乘支持向量机模型中,单波段模型预测总氮最高误差值为71.49%,平均值为27.21%;预测总磷最高误差值为92.25%,平均值为36.12%。波段组合模型预测总氮最高误差值为42.73%,平均值为20.19%;预测总磷相对误差最高值为67.43%,平均值为24.23%。最小二乘支持向量机模型预测总氮最高误差值为17.73%,平均值为7.10%;预测总磷的最高误差为17.91%,平均值为6.70%。因此,最小二乘向量机模型的反演结果要优于波段组合模型,波段组合模型要优于单波段模型。单波段模型与波段组合模型不能用于清河水库总氮、总磷的反演研究中,反演精度过低,达不到反演精度要求。最小二乘支持向量机模型可以用于清河水库总氮、总磷的反演,为今后对清河水库水质监测提供了有效的监测方法与理论依据。
[Abstract]:The purpose of this study is to establish an optimal model for inversion of total nitrogen and total phosphorus contents in Qinghe Reservoir. Taking the Qinghe Reservoir as the study area and the total nitrogen and phosphorus content as the object of study, the single band remote sensing model, the band combination remote sensing model and the least square support vector machine model are established respectively. Based on the analysis of the relative errors between the predicted and measured values of the three models, the least square support vector machine (LS-SVM) model is obtained, which is most suitable for the inversion of total nitrogen and total phosphorus in Qinghe Reservoir. The contents of the research include laboratory analysis of measured data, Pearson correlation analysis of band value and measured value with spss software, and establishment of single band remote sensing model using sensitive band and band combination. The band combined remote sensing model, the least square support vector machine model, and the adaptability of the three models are analyzed and studied. This paper focuses on the establishment of the least squares support vector machine model and the analysis and comparison of the accuracy of the three models. The results are as follows: (1) the data of 80 groups were collected in June, July, September and October 2015 respectively. The average of total nitrogen content in June was the highest, which was 1.73 mg / L in September, the lowest in September, and the highest in June in 1.11 mg 路L ~ (-1) 路L ~ (-1). For 0.07 mg / L, the average value of total phosphorus in the other three months was the same, which was 0.04 mg / L. (2) the single band with the best correlation with total phosphorus in Landsat8 OLI data was B4 in summer and September and October in autumn, and the single band with the best correlation with total nitrogen. Both periods are B5 bands. The best correlation between total phosphorus and total phosphorus in summer is B4 / B3, the best correlation with total nitrogen is B3 / B4, the best correlation with TP in September and October in autumn is B4 / B5, and the best correlation with total nitrogen is B3B4. (3) the best single band or band combination value is the independent variable. In the single band model, band combination model and least square support vector machine model, the measured values of total nitrogen and total phosphorus in Qinghe Reservoir are dependent variables. The maximum error of total nitrogen predicted by single band model is 71.49, the average value is 27.21, and the maximum error of total phosphorus prediction is 92.25, with an average of 36.12. The maximum error of total nitrogen predicted by the band combination model is 42.73 with an average of 20.19, and the highest relative error of total phosphorus is 67.43 with an average of 24.23. The maximum error of total nitrogen predicted by the least square support vector machine model is 17.733.The average value is 7.10 and the maximum error of predicting total phosphorus is 17.91 and the average value is 6.70. Therefore, the inversion result of the least square vector machine model is better than that of the band combination model, and the band combination model is superior to the single band model. The single band model and the combined band model can not be used in the inversion of total nitrogen and total phosphorus in Qinghe Reservoir. The inversion accuracy is too low to meet the requirements of inversion accuracy. The least square support vector machine model can be applied to the inversion of total nitrogen and total phosphorus in Qinghe Reservoir, which provides an effective monitoring method and theoretical basis for the monitoring of water quality in Qinghe Reservoir in the future.
【学位授予单位】：沈阳农业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：X524

【相似文献】