Estimating timber assortment reduction and sawlog proportions with the application of harvester measurements and open big geodata

Abstract

This study aimed to estimate timber assortment reduction and sawlog proportions by integrating in situ tree quality information and remote sensing-based forest management inventory estimates in clear-cutting areas. Using harvester data from operational forestry, we analysed sawlog recoveries from 683 stands collected throughout eastern Finland. The k-Nearest Neighbour (k-NN) method was used to estimate both timber assortments and species-specific sawlog proportions. Remote-sensing based forest attribute maps, satellite image composites and additional geodata were used as predictors. In addition, a Random Forest (RF) model was employed to predict total sawlog proportions. Absolute sawlog volumes were derived by multiplying the model-derived sawlog proportions by volumes in the Metsään.fi forest data repository. Our results showed that the root mean square error (RMSE) values associated with sawlog volumes of Norway spruce (Picea abies (L.) Karst.) were between 40.8–41.0 %, and were between 54.7–59.5 % for Scots pine (Pinus sylvestris L.). The RMSE value associated with total sawlog volume varied from 25.3 % to 27.6 %, i.e. the accuracy was considerably better when species was ignored. Our spruce-dominated dataset yielded more precise results for spruce compared to the other species. The RF model showed better performance in predicting total sawlog proportions than the k-NN approach. Integration of harvester measurements with forest databases and other sources of big geodata can provide substantially improved estimates of sawlog recoveries compared to the current state-of-the-art approach.