<![CDATA[The forestry sector plays a major role in Tanzania’s timber economy, yet lumber recovery estimates in sawmills still rely on manual, log-based measurements that limit accuracy and operational planning. This study developed predictive models to estimate lumber recovery of Pinus patula using forest inventory variables: diameter at breast height (DBH), tree height, and taper. Data were obtained from 80 trees, yielding 254 logs and 2,364 boards, processed at the Laimet 120 and Slidetec Tommi Laine (STL) circular sawmills. Regression models (logarithmic, log-linear, polynomial, and power) were fitted and evaluated using Akaike information criterion, coefficient of determination, root mean square error, mean absolute error, coefficient significance, and K-fold cross-validation. Model performance showed that all equations explained more than 73% of the variation in lumber recovery, with polynomial models providing the highest accuracy, lowest error values, and most stable cross-validated estimates. Predictor importance differed by sawmill: DBH and height were most influential for Laimet 120, while taper improved predictions for STL due to greater variation in stem form. These results demonstrate that forest inventory data can be used to reliably estimate lumber recovery. The developed equations provide sawmills and forest managers with a practical tool for planning log allocation, enhancing efficiency, and minimizing processing waste. Keywords: cross-validation, regression models, sawmill efficiency, tree characteristics, wood processing]]>
Copyrights © 2026
