Allometric tree volume models for Pinus roxberghii and Cedrus deodara in Karnali Province, Nepal

Effective forest management is based on information, such as current growing stock and its future state, which is provided by various forestry models, including tree volume models. Tree volume models are essential tools for estimating growing stock, carbon accounting, timber valuation, identifying forest stands to harvest, growth and yield modeling, and forest ecosystem analysis. We developed tree volume models for two important tree species (Pinus roxberghii Sarg. and Cedrus deodara Roxb.) in Nepal using data from many individuals (mature trees and juveniles) representing wide variations of tree volume allometry. We used diameter at breast height, total height, and crown width of the individuals as predictors in tree volume models for Pinus roxberghii while only the former two predictors in the models for Cedrus deodara. All the mathematical functions described more than 91% and 97% tree volume variations for Cedrus deodara and Pinus roxberghii, respectively. Models for Pinus roxberghii revealed almost similar effect of crown width as tree height on tree volume allometry. Models for this species can be applicable for predicting tree volume with and without bark; however, models for Cedrus deodara can only predict the over-bark tree volume. Testing Pinus roxberghii model against external independent data confirmed a high accuracy of the model. The proposed tree volume models for the species of interest are biologically plausible and statistically robust, and therefore, can be applied to estimate the growing stock precisely, carbon accounting, timber valuation, forest growth and yield modeling. However, users need to apply the model with caution for the forest conditions not covered by modeling data for Cedrus deodara. The prediction accuracy of the models can be further improved through recalibration with additional data collected from wider distributions of the species of interest across the Karnali province and beyond, and application of more robust modeling methods, such as mixed-effects modeling and machine leaning.