• 研究报告 •

### 亚热带常绿林型遥感识别及尺度效应

1. (1山东科技大学， 青岛 266590; 2中国科学院森林生态与管理重点实验室(沈阳应用生态研究所)， 沈阳 110016; 3中国科学院会同森林生态实验站， 沈阳 110016; 4中国科学院清原森林生态系统观测研究站， 沈阳 113300)
• 出版日期:2020-05-10 发布日期:2020-05-10

### Remote sensing-based identification of forest types and the scale effect in subtropical evergreen forests.

ZHANG Yue-nan1,2, FANG Lei2*, QIAO Ze-yu1,2, CHEN Long-chi2,3, ZHANG Wei-dong2,3, ZHENG Xiao2,4, JIANG Tao1

1. (1Shandong University of Science and Technology, Qingdao 266590, Shandong, China;2CAS Key Laboratory of Forest Ecology and Management, Institute of Applied Ecology, Chinese Academy of Sciences, Shenyang 110016, China; 3Huitong Experimental Station of Forest Ecology of Chinese Academy of Sciences, Shenyang 110016, China; 4Qingyuan Forest CERN, Chinese Academy of Sciences, Shenyang 113300, China).
• Online:2020-05-10 Published:2020-05-10

Abstract: Optical remote sensing (ORS) is the primary tool to obtain information on regionalscale vegetation cover. Few efforts have been made to identify forest types within subtropical evergreen forests using ORS. Scale effects have been reported in literature, yet the optimal spatial resolution to identify evergreen forest types is still unclear in practical application. In this study, we used four types of ORS imagery $Pléiades (2 m), RapidEye (5 m), and Landsat-8 (15 m and 30 m)$ to investigate whether the classification of three typical evergreen forest types $Chinese fir forest (CFF), Masson pine forest (MPF), and evergreen broadleaved forest (EBF)$ in subtropical landscapes would be influenced by scale effects. Moreover, we tested the optimal spatial resolution for forest classification. The Random Forest Model was combined with predictive features derived from spectral reflectance, image texture, and vegetation coverage to map landcover types at four spatial resolutions. The results showed that the overall accuracy (OA) of four land-cover maps had a U-curve tendency with increasing spatial resolution. The 2 m Pléiades image generated the highest classification accuracy (Kappa=0.70, OA=0.77) among four types of images. The accuracy of three evergreen forest types also had a similar U-curve tendency. The ranges of the rate of identification (RI) were RICFF=68%-87%, RIMPF=55%-84%, and RIEBF=29%-74%. The CFF and MPF generated lower classification errors in terms of omission error (OECFF=0.26-0.46; OEMPF=0.31-0.50) and commission error (CECFF=0.32-0.53; CEMPF=0.31-0.46)compared with the EBF (CEEBF=0.39-0.66; OEEBF=0.47-0.71). Our results showed that the identification of forest types in subtropical regions is clearly subject to scale effects. Despite this, Landsat-8 imagery at 30 m resolution can produce the highest mapping precision due to its broader spectrum sensors. We proposed that the practical mapping of forest types in subtropical areas should consider the inherent trade-off between spectral features and spatial resolution of remote sensors rather than blindly pursuing high spatial resolution.