Allometric models for predicting aboveground biomass of Combretum-Terminalia woodlands in Amhara, Northwest Ethiopia

Abstract

Dry woodland ecosystems play an important role in the global carbon balance and these need accurate biomass predictions that can be obtained from tree allometry. However, information on allometric models is limited for Combretum-Terminalia woodlands of the Amhara Region in Northwestern Ethiopia. The objective of the study was to develop an allometric model for the aboveground biomass (AGBind) of the tree species in the study areas. A total of 149 trees were considered to predict the AGBind (stems, branches and leaves) of the selected species of which 101 samples were collected from nine tree species in Metema district, and 48 sample trees were obtained from four species in Quara district. A new model was developed using log-transformed data with linear regression, and their performance was assessed with the mean absolute prediction error (MAPE%) and the mean prediction error (MPE%). The results of the study indicated that Diameter at breast height (DBH) was the best predictor variable of biomass components for all species. For site-specific data, combined DBH, tree height and wood density best explained the stem and AGBind with the MAPE ranged 18.6–19.2% compared with the DBH-alone model (MAPE ranging from 29.8 to 30.2%). A positive validation was observed on all model forms, but M3 and M4 provided highly precise predictions of AGBind for independent data. However, a large bias was observed in the branch and leaf biomass model for site-specific data with the MAPE ranged 66.6–72.8%. A little improvement was observed in the model performance of plant functional type, while general models developed from the compiled data performed well. For a given model form, a comparable performance was found between the site-specific, tree functional type and general model. Generally, the general model with DBH and wood density together followed by the one which includes three predictors was a better AGBind predictor over the site and tree functional type. The species and site-specific models are thus accurate for site-based prediction whereas the general model contributes to the reliable prediction of AGBind in the studied Combretum-Terminalia woodlands. The comparison with the published models showed that most models produced large errors with MPE ranging from 16.10 to 36.60%, but the best model of tropical dry forests performed much better with MPE ranging from -2.24 to 0.34% and these models can probably be used for Combretum-Terminalia woodland.