摘要
Abstract
To respond to the"carbon peaking and carbon neutrality"policy guidelines proposed in the national climate change strategy,and enhance the quality and stability of the ecosystem and the carbon sequestration capacity of forests in Inner Mongolia,we conducts a field investigation to classify eight types of forest vegetation cover,and combines Sentinel optical and radar image data and geographical environmental variables to construct a model for inversion of carbon density of forest in Xing'an League,Inner Mongolia Autonomous Region.We compares the applicability of three machine learning methods:Random Forest(RF),Extreme Gradient Boosting(XGBoost),and Long Short-Term Memory Network(LSTM),to estimate the carbon storage of the forest ecosystem in Xing'an League,and to clarify its reserve,spatial distribution pattern,and carbon sequestration potential.1)The field survey results show that among the second-level forest land categories,broad-leaved tree species have the highest carbon density,with an average of 64.00tC·hm-2,significantly higher than that of other tree species.Among the third-level forest land categories,willow trees have the highest carbon density.2)The model results reveal that Extremely Randomized Trees(ERT)can effectively screen important feature variables,improving model accuracy and operational efficiency.The RF model has the highest validation accuracy with R2=0.787,and its performance is more stable than the other two models.Using RF to invert the organic carbon density of forest land in Xing'an League from 2021 to 2023,the average organic carbon density of forest land slightly decreased from 110.95tC·hm-2 to 109.24tC·hm-2,which may be related to the fact that some of the new grown forest land in 2023 is still young.3)The carbon storage of the forest ecosystem in Xing'an League is on the rise,with a growth rate of 7.68％.The forest carbon storage in 2023 is approximately 0.851 ×108tC,indicating that the forest ecosystem has a strong carbon sequestration function in the past three years.The conclusions of this paper can provide a scientific basis for monitoring the changes in forest carbon storage in Xing'an League and the Daxing'anling area.关键词
哨兵影像/林地/机器学习/碳储量Key words
Sentinel images/forest land/machine learning/carbon storage分类
农业科技
