Thanks for the clarification, that makes sense and it is interesting that you keep radar to simple extrapolation. Have you tested any ML based approaches for short term convective development, or do you find optical flow sufficient for your typical forecast horizons?