Research highlights: Using 10-year tree height data obtained after planting from the range-wide provenance trials of Abies sachalinensis, we constructed multivariate random forests (MRF), a machine learning algorithm, with climatic variables. The constructed MRF enabled prediction of the optimum seed source to achieve good performance in terms of height growth at every planting site on a fine scale. Background and objectives: Because forest tree species are adapted to the local environment, local seeds are empirically considered as the best sources for planting. However, in some cases, local seed sources show lower performance in height growth than that showed by non-local seed sources. Tree improvement programs aim to identify seed sources for obtaining high-quality timber products by performing provenance trials. Materials and methods: Range-wide provenance trials for one of the most important silvicultural species, Abies sachalinensis, were established in 1980 at nine transplanting experimental sites. We constructed an MRF to estimate the responses of tree height at 10 years after planting at eight climatic variables at 1 km × 1 km resolution. The model was applied for prediction of tree height throughout Hokkaido Island. Results: Our model showed that four environmental variables were major factors affecting height growth—winter solar radiation, warmth index, maximum snow depth, and spring solar radiation. A tree height prediction map revealed that local seeds showed the best performance except in the southernmost region and several parts of northern regions. Moreover, the map of optimum seed provenance suggested that deployment of distant seed sources can outperform local sources in the southernmost and northern regions. Conclusions: We predicted that local seeds showed optimum growth, whereas non-local seeds had the potential to outperform local seeds in some regions. Several deployment options were proposed to improve tree growth.