Beneath every city lies a question that engineers and planners have long struggled to answer: where, exactly, will the ground give way? A team of researchers has brought that answer into sharper focus, developing a machine learning framework that predicts land subsidence susceptibility with 98.5% accuracy by rethinking not just how models are trained, but what they are trained on. In doing so, they have addressed one of the quieter crises of the built world — the slow, invisible sinking of ground beneath infrastructure, driven by depleting aquifers and expanding urban weight — and offered a mo