The red-edge bands and their derived vegetation indices play a crucial role in monitoring vegetation health.The Gaofen-6(GF-6)and Sentinel-2A satellites are equipped with two and three red-edge bands,respectively,thus...The red-edge bands and their derived vegetation indices play a crucial role in monitoring vegetation health.The Gaofen-6(GF-6)and Sentinel-2A satellites are equipped with two and three red-edge bands,respectively,thus making them invaluable for monit-oring forest health.To compare the performance of these two satellites’red-edge bands in monitoring forest health,this study selected forests in Liuyang City,Hunan Province and Tonggu County,Jiangxi Province and Hanzhong City,Shaanxi Province in China as study areas and used three commonly used red-edge indices and the Random Forest(RF)algorithm for the comparison.The three selected red-edge indices were the Normalized Difference Red-Edge Index 1(NDRE1),the Missouri emergency resource information system Ter-restrial Chlorophyll Index(MTCI),and the Inverted Red-Edge Chlorophyll Index(IRECI).Through training of sample regions,this study determined the spectral differences among three forest health levels and established classification criteria for these levels.The res-ults showed that GF-6 imagery provided higher accuracy in distinguishing forest health levels than Sentinel-2A,with an average accur-acy of 90.22%versus 76.55%.This difference is attributed to variations in the wavelengths used to construct the red-edge indices between GF-6 and Sentinel-2A.In the RF algorithm,this study employed three distinct band combinations for classification:all bands including red-edge bands,excluding red-edge bands,and only red-edge bands.The results indicated that GF-6 outperformed Sentinel-2A when using the first and second band combinations,yet slightly underperforming with the third.This outcome was closely associ-ated with the importance of each band’s contribution to classification accuracy reveled by the Gini importance score,their sensitivity in detecting forest health conditions,and the total number of bands employed in the classification process.Overall,the NDRE1 derived from GF-6 achieved the highest average accuracy(90.22%).This study provides a scientific basis for selecting appropriate remote sens-ing data and techniques for forest health monitoring,which is of significant importance for the future ecological protection of forests.展开更多
基金Under the auspices of National Natural Science Foundation of China(No.31971639)National Natural Science Foundation of Fujian Province,China(No.2023J01225)。
文摘The red-edge bands and their derived vegetation indices play a crucial role in monitoring vegetation health.The Gaofen-6(GF-6)and Sentinel-2A satellites are equipped with two and three red-edge bands,respectively,thus making them invaluable for monit-oring forest health.To compare the performance of these two satellites’red-edge bands in monitoring forest health,this study selected forests in Liuyang City,Hunan Province and Tonggu County,Jiangxi Province and Hanzhong City,Shaanxi Province in China as study areas and used three commonly used red-edge indices and the Random Forest(RF)algorithm for the comparison.The three selected red-edge indices were the Normalized Difference Red-Edge Index 1(NDRE1),the Missouri emergency resource information system Ter-restrial Chlorophyll Index(MTCI),and the Inverted Red-Edge Chlorophyll Index(IRECI).Through training of sample regions,this study determined the spectral differences among three forest health levels and established classification criteria for these levels.The res-ults showed that GF-6 imagery provided higher accuracy in distinguishing forest health levels than Sentinel-2A,with an average accur-acy of 90.22%versus 76.55%.This difference is attributed to variations in the wavelengths used to construct the red-edge indices between GF-6 and Sentinel-2A.In the RF algorithm,this study employed three distinct band combinations for classification:all bands including red-edge bands,excluding red-edge bands,and only red-edge bands.The results indicated that GF-6 outperformed Sentinel-2A when using the first and second band combinations,yet slightly underperforming with the third.This outcome was closely associ-ated with the importance of each band’s contribution to classification accuracy reveled by the Gini importance score,their sensitivity in detecting forest health conditions,and the total number of bands employed in the classification process.Overall,the NDRE1 derived from GF-6 achieved the highest average accuracy(90.22%).This study provides a scientific basis for selecting appropriate remote sens-ing data and techniques for forest health monitoring,which is of significant importance for the future ecological protection of forests.
