Watershed water sampling is challenged by discretely distributed hydrological points and complex terrain, demanding path planning that satisfies multi-dimensional constraints (task priority, tributary sampling sequence, UAV load and endurance) while minimizing path length and response time. This problem is essentially formulated as a Dynamic Vehicle Routing Problem (D-VRP). Unlike conventional path planning methods that often ignore scenario-specific hydrological rules and UAVs’ physical limitations, this study first defines a D-VRP-based optimization model that integrates the aforementioned constraints for hydrological sampling scenarios. A tailored heuristic algorithm, namely the adaptive large neighborhood search (ALNS), is then developed for model solving, with the advantages of fast convergence for large-scale sampling tasks and flexible adjustment for emergency sampling demands. The proposed solution provides a systematic and constraint-aware approach for UAV path planning, which enhances the efficiency and flexibility of field sampling operations in practical hydrological monitoring, and offers a scalable framework for similar environmental monitoring missions involving unmanned aerial vehicles in complex terrains.