Table 3 Robustness of meta-trained ensemble when inferring extreme Reco observations across climate zones at the normalized-target threshold, t > 1.0 (all units in gC m−2 d−1).

From: MetaFlux: Meta-learning global carbon fluxes from sparse spatiotemporal observations

Climate zones

Mean extreme Reco

Baseline RMSE

Meta-trained RMSE

Semi-arid

8.80

1.37 ± 0.10

1.42 ± 0.06

Continental

8.70

2.29 ± 0.06

2.27 ± 0.05

Temperate

8.51

2.13 ± 0.08

2.06 ± 0.06

Tropics

8.71

2.65 ± 0.06

2.35 ± 0.04

  1. The numbers in bold represent models with lower RMSE between the baseline and meta-learning cases across different climate zones.
Back to article page