When we first overlaid our satellite-derived estimates of water stress on those of the latest CMIP6 land simulations, there was a reasonable agreement across most of the globe. One region, however, kept glowing red: the Amazon basin. Model grids there suggested the forest spends almost a third of each year under water stress, while solar-induced fluorescence, eddy-covariance towers and GRACE data insisted it is rarely water limited.
To further test this mismatch, we treated every 2.5° pixel as its own experiment and asked: in an average year, how many months does soil moisture fall below the local plant water-stress threshold? By forcing nine land models with the same observed meteorology, we isolated land-surface characteristics—such as root depth and stomatal control—from rainfall uncertainty.
Globally, the models overshoot water-stress frequency by 14%. That gap increases to 26% in the wet tropics, and is even higher in the Amazon.
Why this is not good news
Water limitation is just one piece of a crowded stress mosaic that includes rising vapour-pressure deficit, record heatwaves, fire, deforestation and shifting rainfall seasonality. Models can exaggerate today’s frequency of water limitation, yet still underestimate tomorrow’s die-back if its trees remain too insensitive to warming or if other hazards compound in ways the model misses. Our study calls for a better representation of plant-water interactions through the soil-plant-atmosphere continuum.
Moreover, evapotranspiration from Amazon forests recycles approximately a third of regional rainfall. If models shut that pump down too early, they can skew rainfall and atmospheric circulation patterns and distort simulated land carbon uptake globally. Getting evapotranspiration right therefore matters far beyond South America; it influences how confidently we can project climate on every continent.
Looking ahead
Rooting depth and plant available water available matter. Field studies show that many Amazonian trees draw water from many metres below the surface, while our basin-scale benchmark that blends fluorescence, gravimetry and measurments from flux towers provides a comprehensive test for how Earth system models simulate water stress.
We hope our analysis becomes a living test that model developers revisit—much as field ecologists revisit permanent plots—until the virtual Amazon is as deeply rooted as the forest whose fate it seeks to foresee.