Surrogate-Assisted Differential Evolution Light Gradient Boosting Machine for Mangrove Aboveground Biomass inversion

Abstract

Mangrove Aboveground Biomass (AGB) inversion holds considerable importance in safeguarding and rehabilitating blue carbon ecosystems, as well as their ability to adapt to climate change. In recent years, machine learning models based on heuristic algorithms for solving mangrove AGB inversion problems have gained wildly attention. However, these hybrid models are facing challenges, including expensive computation costs and low convergence speeds. Thus, an efficient Surrogate-Assisted Differential Evolution Light Gradient Boosting Machine algorithm (SADE-LGBM) is proposed to estimate mangrove AGB in the Maowei Sea, Beibu Gulf of China. This algorithm mainly contains: (i) Introducing a radial basis function surrogate model to predict the virtual fitness and identify promising solutions. (ii) An updating population strategy is proposed to update promising solutions to population efficiently. (iii) DE algorithm and LGBM model are combined to address hyperparameter optimization and feature selection simultaneously. To evaluate the performance of SADE-LGBM, we used a dataset consisting of 227 quadrat data collected from field surveys and compared SADE-LGBM with fourteen other algorithms. The experimental results illustrate that SADE-LGBM achieves the best metrics of ${\text{R}}^2$ = 0.8351, RMSE $=$ 220.4698, with the predicted range of mangrove AGB being 3.2765–207.5331 Mg/ha. The SADE-LGBM algorithm demonstrates its potential as a reliable algorithm for estimating large-scale mangrove AGB.