A geo-prediction model for estimating global terrestrial net ecosystem productivity based on the third law of geography

Terrestrial Net Ecosystem Productivity (NEP), representing the net carbon uptake in terrestrial ecosystems, is a key indicator for assessing carbon sequestration capacity. Data-driven approaches have been widely used for NEP estimation but remain limited by the uneven spatial coverage of global flux networks. To address this issue, we developed a framework based on the Third Law of Geography, translating this traditional geographical principle into a practical model for NEP geo-prediction. In this framework, multidimensional environmental variables were selected to quantify similarities. These variable-wise similarities were aggregated into an overall similarity score, which served as weights to estimate NEP as the weighted average of the 100 most similar samples. Prediction uncertainty was defined as one minus the mean similarity of selected sites, providing a spatially interpretable indicator of confidence and reflecting limitations in flux network coverage. Cross-validation yielded reliable estimates (R 2 = 0. 31–0. 54; RMSE = 26. 24–64. 54 gC m −2 month −1) across five land types. The NEP prediction model based on the Third Law of Geography (GEO3NEP) was also employed to produce a global NEP dataset for 2001–2024. This NEP dataset captures expected spatial and temporal patterns. Spatially, it aligns well with existing NEP products and mitigates tropical overestimation; temporally, it maintains consistent trends even in regions with sparse flux coverage. These results suggest that the GEO3NEP model is an effective way in mitigating the negative impacts of the limited coverage of flux networks in NEP estimation. • It is the first attempt to predict global NEP using the Third Law of Geography. • A global NEP Geo-prediction model is developed based on the Third Law of Geography (GEO3NEP). • The GEO3NEP model enhances interpretability of NEP estimates with quantified uncertainty. • The novel global monthly GEO3NEP dataset mitigates NEP prediction bias well.