Fig. 8
From: MetaFlux: Meta-learning global carbon fluxes from sparse spatiotemporal observations
Comparison between the performance of meta-trained and baseline models at an increasing level of extreme (a) GPP and (b) Reco daily observations. The y-axis indicates the average RMSE when the ensemble predicts carbon fluxes with z-value greater than the threshold t. Overall, we find that the meta-trained ensemble has lower RMSE across increasing extreme threshold than its baseline counterpart.
