Aboveground mangrove biomass estimation in Beibu Gulf using machine learning and UAV remote sensing

Abstract

On the basis of canopy height variables, vegetation index, texture index, and laser point cloud index measured with unmanned aerial vehicle (UAV) low altitude remote sensing, we used eight machine learning (ML) models to estimate the aboveground biomass of different species of mangroves in Beibu Gulf and compared the accuracy of different ML models for these estimations. The main species of typical mangrove communities in Kangxiling were Aegiceras corniculata and Sonneratia apetala. The trunks of Sonneratia apetala were thicker, with an average height of 11.82 m, whereas Aegiceras corniculata trees were shorter, with an average height of 2.58 m. The XGBoost regressor (XGBR) model had the highest accuracy in estimating mangrove aboveground biomass (R2 = 0.8319, RMSE = 22.7638 Mg/ha), followed by the random forest regressor model (R2 = 0.7887, RMSE = 25.5193 Mg/ha). Support vector regression, decision tree regressor, and extra trees regressor had poor fitting effects. Mangrove texture index ranked first in importance for the model, followed by the mangrove laser point cloud height index, and the laser point cloud intensity index performed the worst in the model. Mangrove aboveground biomass in the study area is high in the north and low in the south, ranging from 38.23 to 171.80 Mg/ha, with an average value of 94.37 Mg/ha. Generally, the XGBR method can better estimate the aboveground biomass of mangroves based on the measured mangrove plot data and UAV low-altitude remote sensing data.