“…The top-down EF-based hypotheses implemented here yielded NBE errors that matched or outperformed those from traditional PFTs at a sizable fraction of pixels (55%; Figure 4d), suggesting that the introduction of more realistic trait variability in large-scale TBMs can help to improve predictions of its future behavior, as previously hypothesized (Matheny et al, 2017;Scheiter et al, 2013;van Bodegom et al, 2014;Xu & Trugman, 2021). Overall, our findings support the growing paradigm shift away from the representation of static PFTs and towards the incorporation of realistic trait variability into large-scale TBMs (Berzaghi et al, 2020;Bloom et al, 2016;Jung & Hararuk, 2022;Liu et al, 2022;van Bodegom et al, 2014). EF-based hypotheses represent one promising and flexible approach for doing so, although they are not a panacea-PFT-based assumptions are still superior at nearly half of vegetated pixels in our analysis (45%; Figure 4d).…”