TY - JOUR
T1 - Developing a sub-meter phenological spectral feature for mapping poplars and willows in urban environment
AU - Li, Xiangcai
AU - Tian, Jinyan
AU - Li, Xiaojuan
AU - Wang, Le
AU - Gong, Huili
AU - Shi, Chen
AU - Nie, Sheng
AU - Zhu, Lin
AU - Chen, Beibei
AU - Pan, Yun
AU - He, Jijun
AU - Ni, Rongguang
AU - Diao, Chunyuan
N1 - Funding Information:
We sincerely thank the anonymous reviewers for their insightful comments and suggestions. We also sincerely thank Dr. Junfei Xie and Piesat Information Technology Co., Ltd. for the data support. This work is supported by the National Natural Science Foundation of China (No. 42171330) and the Beijing Outstanding Young Scientist Program (No. BJJWZYJH01201910028032).
Funding Information:
We sincerely thank the anonymous reviewers for their insightful comments and suggestions. We also sincerely thank Dr. Junfei Xie and Piesat Information Technology Co. Ltd. for the data support. This work is supported by the National Natural Science Foundation of China (No. 42171330) and the Beijing Outstanding Young Scientist Program (No. BJJWZYJH01201910028032).
Publisher Copyright:
© 2022
PY - 2022/11
Y1 - 2022/11
N2 - Poplar and willow catkins (PWCs) have caused severe impacts on human health and environmental quality, and accurate poplars and willows (PaWs) mapping with remote sensing is essential to monitor and manage the PWCs. However, two major issues have constrained the urban tree species (e.g. PaWs) identification: (1) the urban tree landscapes are highly fragmented and susceptible to the existence of mixed pixels in the remote sensing imagery; (2) the tree species in urban environment are diverse with high spectral similarity. To this end, this study developed a sub-meter phenological spectral feature (Spsf) with multi-scale and multi-temporal remote sensing imagery for monitoring PaWs at the tree species level. Spsf includes three steps: (1) exploring three key phenological periods of PaWs (leafless period, greenleaf period, and senescence period); (2) selecting one or three spectral indexes to characterize each phenological period; (3) stacking the spectral vegetation indexes from Sentinel-2 SR imagery and freely available sub-meter (0.8 m) Google Earth imagery together. Subsequently, Spsf was taken as the input data to train the deep learning DeepLabv3 + model for predicting the PaWs distribution. The Beijing Plain was chosen as the study area, where the distribution of PaWs was extensive and fragmented. Compared with the field survey reference data, the derived PaWs map achieved the overall accuracy higher than 92 % and the Kappa coefficient of 0.83. The Spsf integrated rich spatial information from sub-meter imagery and phenological spectral information from Sentinel-2 imagery, which may alleviate the impacts of mixed pixels and enhance the spectral separability between PaWs and other tree species effectively. The proposed Spsf-based method provides a new paradigm for sub-meter tree species mapping with multi-source free remote sensing data. The PaWs map can serve as reference data for the relevant departments to monitor and manage the PWCs.
AB - Poplar and willow catkins (PWCs) have caused severe impacts on human health and environmental quality, and accurate poplars and willows (PaWs) mapping with remote sensing is essential to monitor and manage the PWCs. However, two major issues have constrained the urban tree species (e.g. PaWs) identification: (1) the urban tree landscapes are highly fragmented and susceptible to the existence of mixed pixels in the remote sensing imagery; (2) the tree species in urban environment are diverse with high spectral similarity. To this end, this study developed a sub-meter phenological spectral feature (Spsf) with multi-scale and multi-temporal remote sensing imagery for monitoring PaWs at the tree species level. Spsf includes three steps: (1) exploring three key phenological periods of PaWs (leafless period, greenleaf period, and senescence period); (2) selecting one or three spectral indexes to characterize each phenological period; (3) stacking the spectral vegetation indexes from Sentinel-2 SR imagery and freely available sub-meter (0.8 m) Google Earth imagery together. Subsequently, Spsf was taken as the input data to train the deep learning DeepLabv3 + model for predicting the PaWs distribution. The Beijing Plain was chosen as the study area, where the distribution of PaWs was extensive and fragmented. Compared with the field survey reference data, the derived PaWs map achieved the overall accuracy higher than 92 % and the Kappa coefficient of 0.83. The Spsf integrated rich spatial information from sub-meter imagery and phenological spectral information from Sentinel-2 imagery, which may alleviate the impacts of mixed pixels and enhance the spectral separability between PaWs and other tree species effectively. The proposed Spsf-based method provides a new paradigm for sub-meter tree species mapping with multi-source free remote sensing data. The PaWs map can serve as reference data for the relevant departments to monitor and manage the PWCs.
KW - Deep learning
KW - Multi-scale
KW - Phenology
KW - Sub-meter
KW - Tree species classification
KW - Urban
UR - http://www.scopus.com/inward/record.url?scp=85138089341&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85138089341&partnerID=8YFLogxK
U2 - 10.1016/j.isprsjprs.2022.09.002
DO - 10.1016/j.isprsjprs.2022.09.002
M3 - Article
AN - SCOPUS:85138089341
SN - 0924-2716
VL - 193
SP - 77
EP - 89
JO - ISPRS Journal of Photogrammetry and Remote Sensing
JF - ISPRS Journal of Photogrammetry and Remote Sensing
ER -