[1]李爱民,王海隆,许有成.优化随机森林算法的城市湖泊DOC浓度遥感反演[J].郑州大学学报(工学版),2022,43(6):97-103.
 LI A M,WANG H L,XU Y C.Remote Sensing Retrieval of Urban Lake DOC Concentration Based on Optimized Random Forest Algorithm[J].Journal of Zhengzhou University (Engineering Science),2022,43(6):97-103.
点击复制

优化随机森林算法的城市湖泊DOC浓度遥感反演()
分享到:

《郑州大学学报(工学版)》[ISSN:1671-6833/CN:41-1339/T]

卷:
43
期数:
2022年6期
页码:
97-103
栏目:
出版日期:
2022-09-02

文章信息/Info

Title:
Remote Sensing Retrieval of Urban Lake DOC Concentration Based on Optimized Random Forest Algorithm
作者:
李爱民 王海隆 许有成
Author(s):
LI A MWANG H LXU Y C
文献标志码:
A
摘要:
在城市湖泊的可溶性有机碳( DOC)含量的遥感监测问题中,传统回归模型难以描述非线性关系而不能满足精度的要求,因此,将贝叶斯优化算法引入到随机森林模型的参数优化中,提出一种贝叶斯优化随机森林模型( BO-RF)的城市湖泊 DOC 质量浓度反演方法。 以郑州市天德湖水域为例,基于高时空分辨率的 Planet 卫星影像数据和实测的 DOC 水质数据,开展城市湖泊 DOC 质量浓度的遥感反演方法研究。 PEARSON 相关性分析结果表明:反演 DOC 质量浓度的 Planet 卫星影像波段最佳波段 组合为B2 / B4。 利用传统回归方法得到的波段比值模型决定系数 R2 = 0. 466,均方根误差 RMSE = 0. 515 mg / L,无法满足精度要求。 利用支持向量机和 BP 神经网络建模精度有所提升,拟合度 R2 分别为 0. 772 和0. 806,均方根误差 RMSE 分别为 0. 328 mg / L 和 0. 302 mg / L。 引入贝叶 斯优化算法对随机森林模型进行优化得到BO-RF 模型,其拟合度 R2 = 0. 865,均方根误差 RMSE = 0. 253 mg / L。 优化后的模型 BO-RF拟合度较好,模型精度显著提高。 贝叶斯优化后的随机森林 BO-RF 算法更适合反演天德湖水体 DOC 质量浓度,为城市湖泊水质的遥感监测提供参考。
Abstract:
In remote sensing monitoring of dissolved organic carbon (DOC) content in urban lakes, the tradi- tional regression model is difficult to describe the nonlinear relationship and can not meet the accuracy require- ments. In this study, Bayesian optimization algorithm was introduced into the parameter optimization of random forest model, and a DOC concentration inversion method of urban lakes based on Bayesian optimization random forest model (BO-RF) was proposed. Taking the water area of Tiande Lake in Zhengzhou city as an example, the remote sensing inversion method of DOC concentration in urban lakes was studied based on high spatial- temporal resolution Planet satellite image data and measured DOC water quality data. Through PEARSON cor- relation analysis, the results showed that the best band combination of Planet satellite image band for retrieving DOC concentration was B2/B4. The determination coefficient R of the band ratio model obtained by the tradi- tional regression method was 0. 466, and the root mean square error RMSE was 0. 515 mg/L, which could not meet the accuracy requirements. The modeling accuracy was improved by using support vector machine and BP neural network, the fitting R2 was 0. 772 and 0. 806, respectively, and the root mean square error RMSE was 0. 328 mg/L and 0. 302 mg/L, respectively. Bayesian optimization algorithm was introduced to optimize the random forest model to obtain the BO-RF model, and its fitting degree R was 0. 865 and root mean square error RMSE was 0. 253 mg/L. The BO-RF fit of the optimized model was good, and the accuracy of the model was significantly improved. The Bayesian optimized random forest BO-RF algorithm was more suitable for re- trieving DOC concentration in Tiande lake, which could provide a reference for remote sensing monitoring of urban lake water quality.
更新日期/Last Update: 2022-10-03