Machine learning prediction of tropical forest above- ground biomass estimation

Nur Ilyani Mohd Zulkiflee, . and Nurul Ain Mohd Zaki, . and Tajul Rosli Razak, . and Hamdan Omar, . and Shajoeril Tajudin, . and Rohayu Haron Narashid, . and Mohd Nazip Suratman, . and Zulkiflee Abd Latif, . (2023) Machine learning prediction of tropical forest above- ground biomass estimation. Journal of Sustainability Science and Management (Malaysia), 18 (12). pp. 95-110. ISSN 2672-7226

Forests play a significant role as forest sources and have been commonly used to measure carbon stocks within the international carbon cycle and biomass of the forest. Land biomass is an essential element in determining the carbon and carbon balance capabilities of forest ecosystems. This study aimed to estimate forest biomass carbon stocks from the field, Airborne LiDAR, and WorldView-3 data using an Artificial Neural Network and Random Forest. In total, 245 observations and five variables including independent variables, the total height of the tree measured in field (hF), diameter at breast height (DBH), height extracted from Lidar (hL), crown projection area (CPA) and dependent variables (CS) at which based on the data used, multiple regression has been carried out to estimate the forest carbon stocks. ANN has been tested with different hidden layers by trying and error and for Random Forest, two parameters which are the number of randomly picked variables for each node of the tree (Mtry) and the number of trees to grow (Ntree), which was 500 have been used in this study. The best model obtained from both methods was used to generate the carbon stocks map prediction. This study result shows that Model 5 of the ANN algorithm obtains (RMSE = 92.248 Mg haˉ¹ and R² = 0.916). From this study, RF can be concluded as the best model that can be used for the estimation of biomass and carbon stocks as for this study as Model 3 of RF shows the lowest error compared to ANN (RMSE = 49.417 Mg haˉ¹ and R² = 0.976) and the effectiveness of R as the best model for biomass estimation has been proven from the previous research.

Full text available from: Official URL: https://jssm.umt.edu.my/wp-content/uploads/2024/03...

Abstract

Forests play a significant role as forest sources and have been commonly used to measure carbon stocks within the international carbon cycle and biomass of the forest. Land biomass is an essential element in determining the carbon and carbon balance capabilities of forest ecosystems. This study aimed to estimate forest biomass carbon stocks from the field, Airborne LiDAR, and WorldView-3 data using an Artificial Neural Network and Random Forest. In total, 245 observations and five variables including independent variables, the total height of the tree measured in field (hF), diameter at breast height (DBH), height extracted from Lidar (hL), crown projection area (CPA) and dependent variables (CS) at which based on the data used, multiple regression has been carried out to estimate the forest carbon stocks. ANN has been tested with different hidden layers by trying and error and for Random Forest, two parameters which are the number of randomly picked variables for each node of the tree (Mtry) and the number of trees to grow (Ntree), which was 500 have been used in this study. The best model obtained from both methods was used to generate the carbon stocks map prediction. This study result shows that Model 5 of the ANN algorithm obtains (RMSE = 92.248 Mg haˉ¹ and R² = 0.916). From this study, RF can be concluded as the best model that can be used for the estimation of biomass and carbon stocks as for this study as Model 3 of RF shows the lowest error compared to ANN (RMSE = 49.417 Mg haˉ¹ and R² = 0.976) and the effectiveness of R as the best model for biomass estimation has been proven from the previous research.

[error in script]

Item Type:	Article
AGROVOC Term:	above ground tree biomass
AGROVOC Term:	tropical forests
AGROVOC Term:	carbon stock assessments
AGROVOC Term:	LIDAR
AGROVOC Term:	machine learning
AGROVOC Term:	image processing
AGROVOC Term:	forecasting
Geographical Term:	Malaysia
Uncontrolled Keywords:	artificial neural network , random forest
Depositing User:	Mr. Khoirul Asrimi Md Nor
Date Deposited:	28 Oct 2025 15:23
Last Modified:	28 Oct 2025 15:23
URI:	http://webagris.upm.edu.my/id/eprint/2176

Actions (login required)

View Item