Random Forest Outperforms SVM and SGB in Estimating Shorea robusta Biomass Using Sentinel‐2 Indices in the Terai Region of Nepal
Open Access
- 1 January 2026
- journal article
- Published by Wiley in Journal of Sensors
- Vol. 2026 (1) , 3315456
Abstract
Accurate estimation of above‐ground tree biomass (AGTB) is vital for sustainable forest management and climate change mitigation. This study evaluates the performance of three machine learning algorithms (MLAs): support vector machine (SVM), random forest (RF), and stochastic gradient boosting (SGB) for predicting AGTB in Shorea robusta (Sal) forests of Nepal’s Terai region. Using field data from 198 sample plots and 11 Sentinel‐2 derived vegetation indices, models were calibrated and validated. RF outperformed SVM and SGB, achieving the lowest root mean square error (RMSE = 68.36 t ha −1 ), highest correlation coefficient ( r = 0.61), and coefficient of determination ( R 2 = 0.38). Predicted AGTB ranged from 33.93 to 567.03 t ha −1 , reflecting the forest’s heterogeneity. Variable importance analysis revealed normalized difference vegetation index (NDVI) as critical for SVM/SGB, while Modified Chlorophyll Absorption Ratio Index (MCARI) was key for RF. This study establishes RF as the most robust MLA for AGTB mapping in Sal forests, providing a scalable framework for enhancing Nepal’s forest carbon monitoring and supporting Reducing Emissions from Deforestation and Forest Degradation (REDD+) initiatives. Although the predictive accuracy was moderate, this study demonstrates the challenges of biomass estimation in Nepal’s highly heterogeneous Sal forests and provides a replicable framework under data‐limited conditions.Keywords
This publication has 40 references indexed in Scilit:
- Aboveground biomass estimation of black locust planted forests with aspect variable using machine learning regression algorithmsEcological Indicators, 2021
- Innovative trend analysis of spatio-temporal variations of rainfall in India during 1901–2019Theoretical and Applied Climatology, 2021
- Estimation of forest aboveground biomass using combination of Landsat 8 and Sentinel-1A data with random forest regression algorithm in Himalayan FoothillsTropical Ecology, 2021
- Forest aboveground biomass estimation using machine learning regression algorithm in Yok Don National Park, VietnamEcological Informatics, 2019
- Aboveground biomass estimation using multi-sensor data synergy and machine learning algorithms in a dense tropical forestApplied Geography, 2018
- Importance of tree basic density in biomass estimation and associated uncertainties: a case of three mangrove species in TanzaniaAnnals of Forest Science, 2016
- A review of radar remote sensing for biomass estimationInternational Journal of Environmental Science and Technology, 2015
- A universal airborne LiDAR approach for tropical forest carbon mappingOecologia, 2011
- A modified soil adjusted vegetation indexRemote Sensing of Environment, 1994
- Red and photographic infrared linear combinations for monitoring vegetationRemote Sensing of Environment, 1979