Work

Global patterns and future dynamics of belowground net primary productivity

Year: 2026

Type: article

Abstract: Belowground net primary productivity (BNPP) represents a major pathway of plant-derived carbon input to soils and can serve as an indicator of belowground ecosystem functioning. Quantifying its broad spatial patterns, environmental associations, and potential future changes is therefore important for evaluating terrestrial carbon-cycle responses to climate change. However, existing model-based assessments of BNPP remain limited, particularly with respect to biome-specific environmental associations and scenario-based future sensitivity. We integrated 725 global BNPP observations with climate, soil, and topographic covariates using the Light Gradient Boosting Machine (LightGBM) to generate spatially continuous estimates of BNPP in the upper 2 m of soil for broad-scale analysis. We then identified environmental predictors using SHapley Additive exPlanations (SHAP) and projected late-century BNPP changes under CMIP6 scenarios using a Bayesian-weighted multi-model ensemble. We found that the global mean BNPP was 1.58 Mg·ha −1 ·yr −1 , with the highest values in tropical forests and warm temperate forests and the lowest values in dry tundra and tundra ecosystems. Globally, climate variables accounted for the largest share of model attribution, followed by soil variables and topographic properties, but this hierarchy varied markedly across biomes. Soil physical properties were strongly associated with BNPP in humid forests, whereas climatic predictors were more important in drylands and temperature-related predictors were more important in cold ecosystems. Under late-century climate scenarios, global BNPP was projected to increase overall by approximately 0.13 Mg·ha −1 ·yr −1 , with dry tundra and tundra showing particularly high relative sensitivity and projection variability. These results suggest that BNPP provides a useful indicator of biome-dependent belowground carbon input responses to environmental gradients and future climate forcing. However, the estimates should be interpreted as broad-scale, covariate-based projections rather than precise local predictions, particularly in regions affected by sparse observations, managed land use, or non-analog future conditions.

Source: Ecological Indicators

DOI: 10.1016/j.ecolind.2026.115436