集成时间序列Sentinel-1/2数据的江汉平原油菜早期制图

Early mapping of canola in Jianghan Plain by integrating time series Sentinel-1/2 data

关键词:  江汉平原  农业遥感  早期识别  油菜

Key words:  Jianghan Plain  agricultural remote sensing  early identification  canola

摘要:

【目的】 江汉平原是湖北省主要的农作物生产基地，油菜早期阶段的准确定位对于作物生长监测和作物产量预测至关重要。【方法】 文章借助谷歌地球引擎（Google Earth Engine，GEE）云平台，通过结合物候信息的随机森林模型（Random Forest，RF），以10 d为间隔，基于2022年10月1日至2023年4月30日的遥感影像，分别组成不同的生育区间组合，观察对比10 d间隔的各生育区间组合的精度变化情况，寻找油菜的最早可识别时间（Earliest Identifiable Timing，EIT），并制作江汉平原油菜早期识别图。【结果】 （1）雷达影像的VV、VH极化特征加入更有利于油菜的识别；与仅使用光学遥感数据的研究相比，利用雷达数据的极化特征增加了数据的多样性和丰富性。（2）基于Sentinel-1/2数据利用RF分类器获得的油菜早期识别最佳时段具体是2022年12月1—11日，总体精度达到0.88，F1得分为0.88，即可在油菜收获前5个月进行识别提取。（3）该文提出的方法可实现油菜大范围早期提取和快速制图。【结论】 基于Sentinel-1与Sentinel-2集成的时间序列数据最早可在油菜收获前5个月进行识别，综合利用Sentinel-1和Sentinel-2数据在获取最早可识别时段和早期作物制图方面具有良好的效果，该研究可为该地区油菜生产管理、农业种植结构调整和粮油安全保障提供数据支撑和科学服务。

Abstract:

Purpose Jianghan Plain is the main crop production base in Hubei Province，and the accurate positioning of canola in the early stage is very important for crop growth monitoring and crop yield prediction.Method In this paper，with the help of Google Earth Engine（GEE）cloud platform，through the Random Forest（RF）model combined with phenological information，based on the remote sensing images from 1 October 2022 to 30 April 2023，the 10-day interval was used to form different fertility interval combinations，and the accuracy changes of each fertility interval combination with 10-day interval were observed and compared to find the Earliest Identifiable Timing（EIT）of canola. The early identification map of canola in Jianghan Plain was made.Result （1）The addition of VV and VH polarization characteristics of radar images was more conducive to the identification of canola；compared with the research using only optical remote sensing data，the polarization characteristics of radar data increased the diversity and richness of data. （2）Based on Sentinel-1/2 data，the best period for early identification of canola obtained by RF classifier was from 1 to 11 December 2022，with an overall accuracy of 0. 88 and an F1 score of 0. 88，which can be identified and extracted five months before canola harvest. （3）The method proposed in this paper could realize large-scale early extraction and rapid mapping of canola.Conclusion The earliest identification time of canola based on the time series data integrated by Sentinel-1 and Sentinel-2 is five months before harvest. The comprehensive utilization of Sentinel-1 and Sentinel-2 data has a good effect in obtaining the earliest identification time and early crop mapping. This study can provide data support and scientific services for canola production management，agricultural planting structure adjustment and grain and oil security in the region.