Geographia Technica, Vol 21(2), Special Issue: Artificial Intelligence Applications in Geography, 2026, pp. 158-173
GEOSPATIAL DEEP LEARNING FOR BIOMASS AND CARBON STOCK ESTIMATION USING UAV-DERIVED CANOPY METRICS
Pimpilai WUNAPHAI 
, Teerawong LAOSUWAN , Satith SANGPRADID , Yannawut UTTARUK , Narueset PRASERTSRI , Thinnakon ANGKAHAD , Piyatida AWICHIN
ABSTRACT: Accurate estimation of biomass and carbon stock is important for forest monitoring and climate studies. This study developed a geospatial deep learning framework to estimate biomass and carbon stocks in rubber plantations by integrating UAV data with field measurements. The research was conducted in a rubber plantation area in northern Thailand, covering approximately 137 ha, with field data and UAV imagery collected during the 2024 growing season. UAV images were processed to generate Digital Surface Models (DSM), Digital Elevation Models (DEM), and Canopy Height Models (CHM). From these datasets, canopy structural variables such as canopy height, crown dimensions, and tree density were extracted and used as inputs for the model. Tree detection was performed using a YOLO-based deep learning approach, while field measurements were used for model training and validation. The integration of canopy structural information with deep learning improved biomass estimation at the plantation scale. The results showed strong agreement between predicted and field-measured values, with overall detection accuracy exceeding 92%. The model successfully captured spatial variation in biomass and carbon stocks across the plantation, and the estimated carbon values were consistent with plot-level observations. These findings indicate that combining UAV data, canopy structure information, and deep learning provides a practical and cost-effective approach for biomass and carbon estimation in plantation forests. The proposed workflow can also be applied to other plantation areas with similar environmental conditions.
Keywords: Geospatial deep learning; UAV remote sensing; canopy metrics; biomass estimation; carbon stock estimation; rubber plantation
Full article here