Aboveground biomass estimation of black locust planted forests with aspect variable using machine learning regression algorithms

Abstract

An accurate estimation of forest aboveground biomass (AGB) is important for carbon accounting and afforestation policy making, and the aspect factors that affect forest stand growth are important to the accuracy of AGB estimation. In this study, aspect was incorporated as a variable into three machine learning algorithms (MLAs) (support vector machine (SVM), artificial neural network (ANN) and random forest (RF)), to estimate the AGB of black locust (Robinia pseudoacacia) planted forests in 96 field plots with four different aspects (sunny slope, semi-sunny slope, semi-shady slope and shady slope). The results showed that in the models incorporating aspect variables, the increase in accuracy varied from 36.72% to 41.23% for 29 validation plots based on R2. The A-RF model (RF with aspect variable), which had the highest R2 (0.8519) and lowest RMSE and rRMSE (12.552 Mg/ha and 0.175) was considered optimal for AGB estimation of black locust planted forests. The overestimation of sunny and shady slopes, and the underestimation of semi-sunny and semi-shady slopes are reduced by incorporating the aspect variable. Areas with lower AGB values mainly occur on sunny and shady slopes, and areas with higher AGB values mainly occur on semi-sunny and semi-shady slopes. Overall, our study demonstrates that the introduction of the aspect variable provided the model with a basis for the effects of different growth conditions of black locust planted forests on different aspects, which can improve the accuracy of AGB estimation.

Keywords

• Aboveground biomass
• Aspect
• Machine learning algorithms
• Black locust planted forests