北部湾茅尾海无瓣海桑红树林地上生物量反演——基于XGBoost机器学习算法

无瓣海桑是广西从自治区外引进的外来红树林树种，采用定量化算法精确估算无瓣海桑地上生物量对红树林生态修复以及海洋蓝碳监测提供经验和方法。论文以广西茅尾海自然保护区无瓣海桑红树林为研究对象，以野外实测无瓣海桑红树林地上生物量数据和Sentinel-1/2卫星提取的后向散射数据、波段数据、植被指数数据和纹理指数数据为数据源，通过分析各遥感因子与实测红树林地上生物量之间的重要性关系，采用极端梯度提升(XGBoost)机器学习算法对比了不同的变量组合对模型精度的影响，最后基于优选的变量组合反演了无瓣海桑红树林的地上生物量。结果表明：(1)研究区无瓣海桑红树林实测树高范围为1.55—13.58m,平均值为8.37m,胸径范围为0.7—41cm,平均值为15.62cm;(2)通过XGBoost算法优选的21个特征变量组合模型拟合效果较好，其模型在测试阶段R²=0.7237,RMSE=21.70Mg/hm²。XGBoost算法反演研究区无瓣海桑地上生物量介于19.14—138.46Mg/hm²之间，平均值为51.92Mg/hm^...

Aboveground biomass of Sonneratia apetala mangroves in Mawei Sea of Beibu Gulf based on XGBoost machine learning algorithm

Sonneratia apetala is an exotic mangrove species introduced from outside Guangxi Zhuagn Autonomous Region. Quantitative algorithm is used to accurately estimate the aboveground biomass (AGB) of Sonneratia apetala, which provides experience and methods for mangrove ecological restoration and marine blue carbon monitoring. This paper takes the Sonneratia apetalous mangrove of Mawei Sea Nature Reserve in Guangxi as the research object,and takes the field measurd aboveground biomass data of Apetalous mangrove and the backscatter data, band data, vegetation index data and texture index data extracted by Sentinel-1/2 satellite as the data sources. EXtreme Gradient Boosting (XGBoost) machine learning algorithm was used to compare the effects of different variable combinations on the model accuracy by analyzing the importance relationship between each remote sensing variable and the measured AGB of Sonneratia apetala mangrove. Finally, the AGB of Sonneratia apetala mangrove was retrieved based on the optimal combination of variables. The results showed that:(1) The measured height of Sonneratia apetala mangrove in the study area ranged from 1.55m to 13.58m, with an average of 8.37m, and the diameter at breast height(DBH) ranged from 0.7 cm to 41 cm, with an average of 15.62 cm. (2) The fitting effect of the 21 feature variables combination model optimized by XGBoost algorithm was better, and its model R²=0.7237 and RMSE=21.70Mg/hm² in the testing phase. The AGB of Sonneratia apetala mangrove in the study area ranged from 19.14Mg/hm² to 138.46Mg/hm², with an average of 51.92Mg/hm². (3) Cross polarization(VH) backscattering coefficient derived from Sentinel-1 data contributed the most to AGB of Sonneratia apetala mangrove. (4) The high-value areas of the aboveground biomass of Sonneratia apetala mangrove are mainly distributed in the north, northwest and southwest regions to the west, and the low-value areas are mainly distributed in the east and southeast regions to the east. The inversion results were consistent with the actual survey results. In conclusion, XGBoost machine learning algorithm shows good application ability in AGB of Sonneratia apetala mangrove.