Amplified future drying of tropical land constrained by physical theory

Near-surface relative humidity is a fundamental metric of land climate, modulating temperature, precipitation, evapotranspiration, wildfires, and heat stress in a warming world. The response of relative humidity over land to climate change is likely to be substantial but remains uncertain, with climate models projecting a wide range of future changes. Previous work highlighted the inability of models to accurately reproduce observed decreases in land relative humidity over the historical period, inviting scrutiny of the veracity of their future projections. Here, to constrain model projections of land relative humidity, we derive a simple physical theory -- grounded in an ocean‐influence perspective on boundary-layer moisture over land -- that links fractional changes in tropical land relative humidity to the land–ocean warming contrast. According to this theory, due to stronger warming over land relative to oceans in a changing climate, increases in the water-holding capacity of air over land outpace increases in the supply of moisture from oceanic regions, implying a decrease in land relative humidity. The ocean‐influence theory explains much of the spread across models in historical land relative humidity trends as well as the trend in reanalysis data. Combining the theory with observational estimates of the radiatively-forced land-ocean warming contrast, we obtain constrained estimates of future tropical land relative humidity change (–6.4 %/K assuming a climate-invariant warming contrast; -4.4 %/K when changes in warming contrast across climate states are accounted for). Both constrained estimates suggest substantially stronger tropical drying compared to the unconstrained multi-model median change (-1.5 %/K). This emergent constraint emphasises the systematic underestimation of future drying by models and reveals its physical basis, with important implications for projecting land-climate impacts and sharpening directions for future research to minimise this key model bias.