Extended Data Fig. 10: The uncertainty of resilience trend due to the choice of the discounting parameter controlling the information decay rate.

Spatial patterns of resilience trend, same as Fig. 1 (default discounting parameter δ0.98), but with alternative discounting parameters of δ0.97 (a) and δ0.99 (b), respectively. Model performances are measured by log-likelihoods and compared between the alternative parameters and the default, shown as differenced maps and density plots, the black dashed line shows the average log-likelihood difference (δ0.98-δ0.97 in c; δ0.98-δ0.99 in d). The model performs the best using δ0.98, according to the average log-likelihood, which is used as the default in Fig. 1. All three discounting parameters result in consistent trend directions in southern boreal forests, supporting the robustness of reduced resilience in boreal forests. The model suggests relatively lower confidence in the increasing resilience in the Arctic tundra given δ0.99 yields reverse trends from δ0.98 and δ0.97, in despite of a similar model likelihood as that using δ0.98 (d).