Ignoring previous water conditions underestimates global terrestrial ecosystem productivity in severely arid vegetation regions

In the context of global climate change, changes in water conditions significantly affect the productivity of terrestrial ecosystems. However, the impact of previous water conditions has not received sufficient attention. Using machine learning and a large amount of global ground observation data, we evaluated the impact of previous water conditions at various time scales on global productivity estimation through paired experiments and residual comparison methods. By incorporating run theory, we investigated the uncertainty introduced by neglecting previous water conditions in estimating ecosystem productivity across various levels of vegetation drought frequency, intensity, duration, and severity. We found that 36.8 % of global net primary productivity (NPP) is predominantly influenced by water conditions over the previous 24 months, a percentage higher than those observed at 6, 12, and 18-month time. Considering previous water conditions, the estimated global productivity was 66.72 Pg C yr⁻¹ (468.98 g C m⁻² yr⁻¹). Neglecting previous water conditions may lead to an underestimation of global productivity by up to 1.27 Pg C(p < 0.05). Regions experiencing severe, prolonged, and intense drought may have significant NPP estimation biases without considering previous water conditions. This study provides an important step toward understanding the impact of previous water conditions and vegetation drought disasters on the carbon sink function of global terrestrial ecosystems.