Causality-guided explainable machine learning shows groundwater accounts for 48% to 101% of aridity's effect on forest photosynthesis across the contiguous United States

Groundwater's relative importance in regulating the spatial pattern of photosynthesis across the contiguous United States is comparable to that of climatic aridity, with groundwater accounting for 48% to 101% of aridity's effect in forests. The contribution of groundwater to spatial variation in mean annual photosynthesis had remained poorly quantified relative to well-known factors such as climatic aridity; this study provides relative-importance ranges across vegetation types. Based on satellite solar-induced fluorescence observations combined with model estimates of water table depth and aridity, analyzed with causality-guided explainable machine learning; evidence comes from the analytical framework and the ranges reported in the abstract.

Groundwater importance is not limited to forests: relative to aridity it remains substantial in savannahs and shrublands (30% to 58%), grasslands (22% to 42%), and croplands (15% to 32%). Extends assessment of the groundwater-photosynthesis relationship from a single vegetation type to multiple types, providing a gradient of relative importance across ecosystems. Per-vegetation-type results from the same causality-guided explainable machine learning framework, with percentage ranges reported in the abstract.

The authors interpret this pattern as water tables regulating water available in the rooting zone, with an effect comparable in magnitude to climatic aridity. Links the statistical attribution result to a rooting-zone water-supply mechanism, adding a groundwater dimension to modeling of land-atmosphere interactions. A mechanistic interpretation proposed by the authors based on observations and model attribution, expressed in the abstract as reflecting 'the role of water tables in regulating the water available in the rooting zone.'

The result is scoped to the contiguous United States and to an analysis setting that uses solar-induced fluorescence as a proxy for photosynthesis and model estimates of water table depth and aridity; it is directly relevant to researchers and model developers working on land-atmosphere interactions, vegetation water stress, and groundwater-ecology coupling, and the causality-guided explainable machine learning framework may inform similar attribution in other regions or vegetation types.

The abstract does not specify the spatiotemporal resolution or time period of the solar-induced fluorescence, water table depth, and aridity data, nor the concrete implementation and robustness checks of the causality-guided explainable machine learning. The relative-importance range for groundwater is wide (48% to 101% in forests), with the upper bound exceeding aridity's effect, so the origin and uncertainty of this range warrant confirmation in the full text. In addition, the conclusions are based on the contiguous United States and should be extrapolated cautiously to regions with different climate and hydrological conditions.