The adaptation of social-ecological systems such as managed forests depends largely on decisions taken by forest managers who must choose among a wide range of possible futures to spread risks. We used robust decision theory to guide management decisions on the translocation of tree populations to compensate for climate change.We calibrated machine learning correlational models using tree height data collected from five common garden tests in France where Abies alba provenances from 11 European countries are planted. Resulting models were used to simulate tree height in the planting sites under current and 2050 climates (regional concentration pathway scenarios (RCPs) 2.6, 4.5, 6.0 and 8.5). Our results suggest an overall increase in tree height by 2050, but with large variation among the predictions depending on the provenance and the RCPs. We applied maximin, maximax and minimax decision rules to address outcomes under five uncertain states of the world represented by the four RCPs and the present climate (baseline). The maximin rule indicated that for 2050, the best translocation option for maximising tree height would be the use of provenances from Northwest France into all target zones. The maximax and minimax regret rules pointed out the same result for all target zones except for the ‘Les Chauvets’ trial, where the East provenance was selected. Our results show that decision theory can help management by reducing the number of options if most decision rules converge. Interestingly, the commonly suggested recommendation of using multiple provenances to mitigate long-term maladaptation risks or from ‘pre-adapted’ populations from the south was not supported by our approach.