Short-rotation plantations (SRPs), including eucalypt woodlots, now constitute an integral component of the afforestation strategies for many tropical nations under the Paris Agreement. However, wildfires and undocumented clear-cutting harvests are common in these plantations, often occurring prior to reliable measurement and recording of stem volume variables, especially the diameter at breast height (DBH referred here as D), a key parameter for forest biomass and carbon scaling. In such scenarios, D can be reconstructed from stump diameter (DS) using allometric equations. However, locally derived equations to relate DS to D (DS–D models) are scarce in East Africa. As a result, forest inventories in the region mostly rely on generic equations whose broad specifications do not adequately capture site-specific disturbances such as fire and rotational clear-cutting. This study developed DS–D models using data from 115 destructively and non-destructively sampled Eucalyptus camaldulensis trees across first-rotation, post-fire, and coppice-regenerated stands in the Kenyan savanna. Generalized additive models, analysis of covariates, and linear mixed-effects models were employed using under-bark (DSuB) and over-bark (DSoB) stump diameters serving as primary predictors. Model validations were through conventional analytical techniques and machine learning-based cross-validation procedures. The results showed negative shift in intercept coefficients in the post-fire plots, with effect sizes and confidence intervals indicating consistent 1–1.5 cm reductions in D relative to undisturbed stands, suggesting variation in bark or wood allocation as a response to a previous fire episode. Models based on DSuB demonstrated greater predictive stability than those based on DSoB across different disturbance conditions. Locally calibrated models improved prediction power by up to 22% with residual error reduction lower by about 30% in fire and clear-cutting disturbed plots, emphasizing the importance of disturbance-sensitive, site-specific models for enhancing biomass and carbon accounting in forest ecosystems.
Austin et al. (Tue,) studied this question.