Table 2 Summary of uncertainties and possible biases: factors that might explain why gross primary production (GPP) would be biased low, and/or soil respiration (R_S) too high.

1) R_S data are less diverse than those of GPP, with almost all Rs_lit ultimately deriving from a large but single global database¹⁸.

2) Tropical and subtropical forests are greatly under-sampled⁵².

3) Jian et al.³⁹ showed that uneven distribution of R_S sites may cause overestimation of global R_S by ~6 Pg C yr⁻¹.

4) In situ Rs measurements may not be representative of Rs at ecosystem-scale^{53, 54}.

5) Rs cannot be measured directly at the ecosystem scale or using remote sensing, and we must upscale in situ measurements^{14, 15, 23, 55}.

6) Models do not have a clear mechanistic representation of Rs (as compared with GPP)^{14, 15, 23, 55}.

1) Satellite data algorithms and thus products have significant uncertainties (e.g., LAI and PAR conversion efficiency, ε)^{7, 56,57,58,59,60,61,62}.

2) Remote sensing may not fully account for understory production²⁴ or belowground C allocation²⁵.

3) GPP is probably underestimated in the tropics^{9, 27}, as well as in managed and fertilized croplands²⁸.

4) There are totally more than 900 flux tower sites worldwide (https://fluxnet.org/sites/site-summary/), but they are not evenly distributed, with some ecosystem types (e.g., tropic forests) less represented⁶³ (Supplementary Fig. 5).

5) Lack of R_root: R_A ratio data for low photosynthesis productivity region (Supplementary Fig. 7d)