Earth's radiative feedback parameter measures how efficiently the planet sheds excess heat. When the surface warms, more radiation escapes to space, cooling the surface back down. The feedback parameter λ quantifies this response: more negative means more stable. At -3 W/m²/K, a one-degree warming triggers a 3 W/m² increase in outgoing radiation. The planet pushes back hard.
Van Loon et al. (arXiv:2603.12515) train a neural network on climate model output and apply it to observed temperature patterns through 2025. They find that λ reached its most stabilizing value around the mid-1990s, approximately -3 W/m²/K, and has since weakened to approximately -2 W/m²/K. Earth's thermostat is losing its grip.
The mechanism: specific surface warming patterns modulate the radiative response. Warming in the subtropical Northeast Pacific — verified through targeted model experiments — weakens the feedback by altering low cloud cover and lapse rate profiles. The location of warming matters as much as the amount. A degree of warming in one ocean basin can be twice as destabilizing as the same degree somewhere else.
The standard climate sensitivity debate focuses on equilibrium — how much warming per doubling of CO₂. But λ fluctuates on decadal timescales as ocean temperature patterns shift. The same concentration of greenhouse gases produces different effective sensitivities at different times, not because the physics changed but because the spatial pattern of warming changed. The feedback responds to the shape of the warming, not just the magnitude.
A climate that is losing its stabilizing feedback while accumulating more greenhouse gases is not just warming faster — it is becoming progressively less able to resist warming. The destabilization is itself destabilizing.