A meta-analysis of predictive accuracies and errors of biomass estimation models in Sub-Saharan Africa

Accurate biomass estimation is essential for forest monitoring, energy planning and carbon accounting in Sub-Saharan Africa (SSA), where destructive sampling is often impractical. Biomass estimation models (BEMs) offer scalable alternatives, but their predictive accuracy varies across forest types, species and data sources. This study conducted a systematic meta-analysis of 39 BEMs from 22 peer-reviewed studies conducted in SSA, evaluating their model performance using standardised metrics of coefficient of determination (R2) and root mean square error (RMSE). Data were sourced from Global Allometric Tree database, Scopus and Web of Science, following PRISMA guidelines. Both destructive and non-destructive models based on field and remote sensing (RS) data were included. Meta-analytic computations incorporated Fisher's Z-transformation and random-effects modelling to account for heterogeneity. Results indicate high predictive accuracy (mean R2 = 0.82), but substantial variation in error (mean RMSE = 108.9 Mg/ha, SD = 511.6), reflecting methodological and ecological diversity (I2 = 99.87 %). Locally calibrated allometric models achieved the highest accuracy, while RS-based models using optical data alone exhibited higher error rates. Hybrid models integrating LiDAR, radar and optical data demonstrated superior performance when combined with machine learning techniques. Key predictors such as diameter at breast height, tree height and wood density consistently improved model accuracy. Emerging evidence underscores the significance of trees outside forests in national carbon inventories. This study recommends adopting hybrid BEMs tailored to local ecological conditions and incorporating multi-sensor RS data. The findings inform biomass monitoring strategies for forest conservation, REDD+ MRV systems and sustainable energy planning in SSA.