Predicting wood assortment volumes from stand attributes, topography, and airborne laser scanning data using accurately positioned harvester data as a reference

Abstract Accurate information on sawn- and pulpwood volumes is critical for efficient forest management and pre-harvest planning. Stand-level data on forest attributes are routinely maintained and form the basis for management decisions, while digital terrain models (DTMs) provide topographic information that can complement this information. Airborne laser scanning (ALS) provides more detailed and structural data for estimating forest attributes, but its acquisition and processing are resource intensive. The main goal of this study was to quantify the individual and combined contributions of stand, topographic, and ALS data to estimate wood assortment volumes in stands scheduled for harvest. We used reference data obtained from cut-to-length harvesters equipped with real-time kinematic (RTK) positioning and crane-mounted sensors that enabled sub-meter accuracy in positioning of harvested trees. The data comprised 402 176 trees harvested in 125 operations in South Norway. Using multivariate random forest models, we assessed the contribution of each data source individually and in combination and tested a classical calibration of predictions. Variable importance analysis showed that ALS variables were most influential in the modelling by a considerable margin. Accuracies of total, sawnwood, pulpwood volume, and sawnwood proportion estimates ranged from 15–40 per cent, 22–55 per cent, 24–43 per cent and 16–30 per cent relative root mean square error (RMSE), respectively, depending on the data source combination and the use of classical calibration. Using stand and topographic data only marginally improved accuracies compared to using ALS data alone. Classical calibration of predictions did not improve the accuracies in most cases, although it tended to reduce systematic under- and overestimation at the extremes of the prediction range. The results highlight the utility of ALS data for predicting wood assortment volumes and emphasize the importance of prioritizing its acquisition in operational planning.