Cloud shadows and floodwater look nearly identical to a satellite. Both absorb near-infrared. Both have ragged, blurry edges that wash out ground texture. The standard water indices flag both as "water," because the underlying physics is the same. And the models doing the flagging are trained on disaster data that punishes a missed flood harder than a false alarm — so they're built to over-trigger. Now make a $2M parametric payout hinge on telling them apart from one image. It fails both ways. In Nagaland, India, a parametric scheme never triggered despite confirmed flooding — the threshold was set too high. Real flood, no payout, lawsuits. Flip it: a cloud shadow triggers a payout for a flood that never happened — drained reserves, an open door to fraud. And "just use radar, it sees through clouds" isn't the fix. Backscatter drops over smooth water — but also over mountain terrain shadows. NASA's near-real-time flood product carries so many false positives in its 1-day composite that NASA won't publish it. The honest answer: no single sensor solves this. What works is temporal radar-and-optical fusion — stacking images over time so a real flood separates from a shadow that sat there yesterday and stays tomorrow, with a pixel-level confidence trail that survives an actuarial and legal audit. If you've priced or paid a parametric flood trigger: how do you handle the basis risk between what the satellite sees and the ground? #ParametricInsurance #FloodRisk #RemoteSensing