Abstract Biomass of trees may be predicted either directly applying allometric models or indirectly from volume and biomass expansion factors (BEFs). For the Dry Afromontane forests, the second largest biomass pool in Ethiopia, such methods are not devised and properly documented. The main objective of this study was to explore different aboveground tree biomass prediction options based on destructively sampled tree biomass data. We explored the direct method by means of 1) new mixed-species general biomass models developed in the present study, and 2) some previously developed models including the pan-tropical models, and the indirect method by means of 3) volume and BEFs. From two sites in south-central Ethiopia, based on information from systematic sample plot inventories, 63 trees from 30 different species that contributed about 87% to the total forest basal area, were destructively sampled. Weighted nonlinear regression was applied to fit new models and their performance was assessed using root mean squared error (RMSE, %), mean prediction error (MPE, %) and pseudo-R2 based on leave-one-out-cross-validation. Previously developed models and the indirect method were also evaluated by means of RMSE and MPE. The new general total biomass models performed well with pseudo-R2 ranging between 0.87 and 0.96 and are presented along with covariance matrices for the parameter estimates enabling error propagation in biomass estimation. Most previously developed models resulted in significant MPEs up to 78%, while the best pan-tropical model performed much better with an MPE of about 7%. The indirect method also showed poor performance with MPEs ranging between 5% and 30%. Generally, the new models are accurate and flexible, thus, preferred over all previously developed models and the indirect method for application. However, their application to Dry Afromontane forests outside the study sites should be made only after thoroughly evaluating growing conditions and species composition. The results are step forward to enhance decisions made towards sustainable forest management including the REDD+ implementation for Dry Afromontane forests in Ethiopia.

中文翻译：

---

埃塞俄比亚中南部干旱非洲山地森林地上树木生物量预测方案

摘要 树木的生物量可以直接应用异速生长模型或间接通过体积和生物量扩展因子 (BEF) 进行预测。对于埃塞俄比亚第二大生物量库——干燥的非洲山地森林，此类方法没有设计和正确记录。本研究的主要目的是基于破坏性采样的树木生物量数据探索不同的地上树木生物量预测选项。我们通过 1) 在本研究中开发的新的混合物种一般生物量模型，和 2) 一些以前开发的模型，包括泛热带模型，以及通过 3) 体积和 BEF 的间接方法来探索直接方法。从埃塞俄比亚中南部的两个地点，根据来自系统样地清单的信息，来自 30 个不同物种的 63 棵树木被破坏性地采样，这些树占森林总面积的 87%。应用加权非线性回归来拟合新模型，并使用均方根误差 (RMSE, %)、平均预测误差 (MPE, %) 和基于留一法交叉验证的伪 R2 评估其性能。先前开发的模型和间接方法也通过 RMSE 和 MPE 进行了评估。新的通用总生物量模型在伪 R2 范围在 0.87 和 0.96 之间表现良好，并与参数估计的协方差矩阵一起呈现，使生物量估计中的误差传播成为可能。大多数以前开发的模型产生了高达 78% 的显着 MPE，而最好的泛热带模型表现要好得多，MPE 约为 7%。间接方法也表现出较差的性能，MPE 在 5% 到 30% 之间。一般而言，新模型准确且灵活，因此优于所有先前开发的模型和应用的间接方法。然而，只有在彻底评估生长条件和物种组成后，才能将它们应用于研究地点以外的干燥非洲山地森林。结果是进一步加强对可持续森林管理的决策，包括对埃塞俄比亚干旱的非洲山地森林实施 REDD+。只有在彻底评估了生长条件和物种组成后，才能将它们应用于研究地点以外的干燥非洲山地森林。结果是进一步加强对可持续森林管理的决策，包括对埃塞俄比亚干旱的非洲山地森林实施 REDD+。只有在彻底评估了生长条件和物种组成后，才能将它们应用于研究地点以外的干燥非洲山地森林。结果是进一步加强为可持续森林管理做出的决策，包括对埃塞俄比亚干旱的非洲山地森林实施 REDD+。