Allometric modelling and biomass estimation in South American Savannas : advances from the Brazilian Cerrado

Abstract

Tropical savannas cover almost one-fifth of the Earth’s land surface, surpassing tropical forests in area, and play a critical role in biodiversity conservation, ecosystem services, and global carbon cycling. Despite its ecological and climatic importance, accurately estimating biomass remains challenging due to the biome's structural heterogeneity and the limited applicability of existing pantropical allometric models, which have primarily been developed for tropical forests. In this thesis, we developed new allometric models tailored to Brazilian savanna vegetation using destructive sampling data from 915 stems (123 species ) across eight Cerrado regions and applied them to 50 permanent plots comprising (>37,000 stems). To harmonise measurement standards, we first constructed a conversion model between diameter at 0.30 m height (D0.3) and diameter at breast height (D1.3), and subsequently tested six mixed-effects models incorporating tree height (H) and wood density (ρ). The best predictive performance was obtained with model mB2a (ln(B) ~ ln(D1.3) + ln(D1.3)2 + ln(ρ)), which accounted for spatial variability by including region as a random effect. We also developed a height–diameter (H–D1.3) model to estimate missing height data. Although models including height offered slightly higher explanatory power, mB2a yielded superior cross-validation results (RMSE = 1.49 kg, MAE = 1.34 kg), emerging as the most reliable predictor. Beyond the Cerrado, structural and floristic convergence suggests these models may be transferable to other South American savannas, such as the Llanos and Amazonian patches. By providing biome-specific allometric tools, our study improves the reliability of carbon stock assessments, with direct implications for conservation planning, REDD+ initiatives, and climate change mitigation across South America’s dry tropics.