Building a high-resolution site index map using boosted regression trees: the Norwegian case

Accurate estimation of site productivity is essential for forest projections and scenario modelling. We present evaluate models to predict index (SI) whether a productive (potential total stem volume production ≥ 1 m 3 ·ha −1 ·year ) in wall-to-wall high-resolution (16 × 16 m) SI map Norway. investigate remotely sensed data improve predictions. also study the advantages disadvantages using boosted regression trees (BRT), machine-learning algorithm, create high-accuracy maps. use climatic topographical data, soil parent material, land resource map, depth water, together with Sentinel-2 satellite images airborne laser scanning metrics, as predictor variables. observed at more than 10 000 National Forest Inventory (NFI) sample plots throughout Norway fit BRT validate 5822 independent temporary from NFI. benchmark our results against estimates monitoring inventories. find that has root mean squared error (RMSE) ranging 2.3 (hardwoods) 3.6 (spruce) when tested validation NFI plots. These RMSEs are similar or marginally better an evaluation operational management plans where normally stems manual photo interpretation.