Fig. 4
From: MetaFlux: Meta-learning global carbon fluxes from sparse spatiotemporal observations
Meta-learning datasets where base tasks are used for meta-training and target tasks for meta-update. The subscripts train and test are used to train and evaluate the model in each of the two meta-learning steps. Target tasks are selected by randomly sampling half of the FLUXNET2015 stations in the tropical and semi-arid regions, which are sparse, and one station from each plant functional type (PFT) including those in the cropland and boreal regions. By extension, the base tasks consist of the remaining stations.
