Flash floods are one of the most common natural disasters worldwide, causing thousands of casualties every year. The emergence of Unmanned Aerial Vehicles (UAVs) gives the possibility to monitor these events over large geographical areas. In this talk, we focus on the problem of trajectory planning for a swarm of unmanned aerial vehicles sensing flooding conditions. We first use deep-learning to efficiently approximate the flood evolution over time, given external inputs. We then formulate the problem of maximizing the information gained over a finite time horizon, and solve it to determine optimal UAV trajectories. Simulation results show that this approach greatly reduces uncertainty, while being tractable in near real-time on a regular desktop computer.