Event-based vision sensors offer high dynamic range, low latency, and sparse data representation, making them promising for autonomous lunar navigation and landing under extreme illumination and motion conditions. Their asynchronous operation also aligns naturally with Spiking Neural Networks (SNNs), enabling low-power, real-time perception. However, progress in event-based perception for space applications is limited by the lack of realistic datasets with accurate motion ground truth and dedicated architectures for event-based data. This work presents two complementary contributions: an event-based lunar vision dataset and a UNet-like SNN for dense optical flow estimation. The dataset includes Low Lunar Orbit and representative landing sequences, providing high-resolution rendered images, event streams, depth maps, and dense optical-flow ground truth. Images are generated using ABRAM, a physics-based radiometrically validated renderer, while event streams are produced with v2e to emulate realistic event-camera behaviour. Depth and optical flow are analytically derived from the simulated geometry and camera motion. The proposed SNN estimates dense optical flow directly from event streams using a multiscale encoder–decoder architecture. Together, the dataset and model provide a framework for benchmarking event-based perception algorithms for autonomous lunar navigation.