| Authors |
Hui Ma, Bingfang Wu, Fuyou Tian, Hang Zhao, Yifan Li, Mengxiao Li, Miao Zhang, Hantian Wu, Hongwei Zeng, Liang Zhu |
| Intro |
As a dominant component of the Three-North Shelter Forest Program (TNSFP) in the Three-North Region of
China (TNR), farmland shelterbelts (FSs or windbreaks) play crucial roles in reducing wind erosion and soil
degradation. However, large-scale FS products are currently unavailable, posing challenges for their monitoring
and optimization. To address this issue, we developed a deep learning-based framework to extract FSs from
Sentinel-2 data and analyze their spatial patterns in the TNR. This study produced the first 10-m resolution FS
map and the first vectorized FS network for the TNR in 2022. Our results indicated that the total length of FSs
was 915.09 ± 21.45 thousand km, while the total area was 1575.18 ± 58.75 thousand ha. Both FS length and
area were highest in Heilongjiang, Xinjiang, and Inner Mongolia, whereas the lowest values were observed in
Beijing, Tianjin, and Qinghai. The average FS to farmland ratio, connectivity, and FS deviation angle were 2.83 %
(± 0.11 %), 0.34, and 26◦, respectively, highlighting potential for structural optimization. FSs were more
concentrated, with well-structured spatial patterns in Northeast China and Xinjiang, while further construction
efforts were needed in the Hetao Plain, Guanzhong Plain, and Hexi Corridor. The final FS map demonstrated
robust performance across multiple evaluation metrics. Specifically, the pixel-based user’s accuracy was 79.20 %
and the producer’s accuracy was 79.39 %. The average object-based F1-score was 84.13 % and the average
length-weighted F1-score was 89.62 %. Overall, this study provides fundamental data and valuable insights to
support agricultural infrastructure assessment and optimization. |
English
中文