融合Sentinel-1/2数据和机器学习算法的冬小麦产量估算方法研究
Research on yield estimation method of winter wheat based on Sentinel-1/2 data and ma-chine learning algorithms
张永彬 1李想 1满卫东 2刘明月 3樊继好 4胡皓然 4宋利杰 1刘玮佳1
作者信息
- 1. 华北理工大学矿业工程学院,河北唐山 063210
- 2. 华北理工大学矿业工程学院,河北唐山 063210;唐山市资源与环境遥感重点实验室,河北唐山 063210;河北省矿区生态修复产业技术研究院,河北唐山 063210
- 3. 华北理工大学矿业工程学院,河北唐山 063210;唐山市资源与环境遥感重点实验室,河北唐山 063210;矿产资源绿色开发与生态修复协同创新中心,河北唐山 063210
- 4. 高分辨率对地观测系统河北唐山数据应用中心,河北唐山 063210
- 折叠
摘要
针对光学影像容易受到云雨天气影响,导致农作物产量估算精度低的问题,本研究融合冬小麦孕穗期Sentinel-2光谱信息和Sentinel-1后向散射系数,并采用极端梯度提升、随机森林和支持向量机3种机器学习回归方法建立唐山市冬小麦产量估算模型,选用最佳模型实现唐山市冬小麦产量反演.结果表明:基于植被指数和后向散射系数的极端梯度提升模型的估产效果最好,决定系数(R2)为0.654,均方根误差(RMSE)为0.499 t·hm-2,归一化均方根误差(nRMSE)为10.02％.24个遥感特征变量中,NDMI、NDVIre3和NDVIre2的重要性远高于后向散射系数.基于最佳估产模型反演唐山市冬小麦产量空间分布,冬小麦产量范围主要集中在7.00～8.00 t·hm-2,所占比例达到40.75％,冬小麦产量分布总体上与地面真实情况相近.本研究提出Sentinel-1/2数据和机器学习算法相融合的冬小麦产量估算方法,有效提高了机器学习方法反演冬小麦产量的准确性,并加强了模型的解释性,该方法具有一定可行性.
Abstract
Aiming at the problem that optical images are easily affected by cloud and rain weather,resulting in low accuracy of crop yield estimation,in this study,the Sentinel-1/2 spectral information and backscattering coefficient at winter wheat heading stage were combined,and three machine learning regression methods of extreme gradient boosting,random forest and support vector machine were used to establish the winter wheat yield estimation model in Tangshan,the best model was selected to realize the winter wheat yield inversion in Tangshan.The results show that:the extreme gradient boosting model based on vegetation index and backscattering coefficient had the best estimation effect,with the determination coefficient(R2)of 0.654,the root mean square error(RMSE)of 0.499 t·hm-2,and the normalized root mean square error(nRMSE)of 10.02％.Among the 24 remote sensing feature variables,the importance of NDMI,NDVIre3 and NDVIre2 was much higher than that of the backscattering coefficient.Inverse spatial distribution of winter wheat yield in Tangshan based on optimal yield estimation model,the yield range of win-ter wheat was mainly concentrated in 7.00-8.00 t·hm-2,accounting for 40.75％,the distribution of winter wheat yield was generally similar to the ground truth.This study proposed Sentinel-1/2 data and integration of machine learn-ing algorithms of winter wheat yield estimation method,effectively improve the inversion accuracy of winter wheat yield and machine learning method to strengthen the explanatory of the model,the method has certain feasibility.
关键词
遥感/产量/冬小麦/Sentinel-1/2数据/机器学习算法Key words
remote sensing/yield/winter wheat/Sentinel-1/2 data/machine learning algorithm引用本文复制引用
出版年
2024
NETL
NSTL
万方数据