Predicting Weather Conditions Utilizing Artificial Neural Networks for C-17 Mission Planning

Weather forecasting remains an important field of research given the wide impact it has across many industries, including the Department of Defense (DoD). The ability to have an accurate forecast particularly affects the area of air mobility and transportation, where it is an important aspect of mission planning. With the DoD’s 2019 fiscal year budget, $24 billion is allocated just for fuel consumption. This highlights an impetus for providing consistent, reliable, and fuel-efficient flight plans in order to reduce fuel consumption, and ultimately save taxpayer dollars.

This paper examined the applicability of using temporal convolution nets, a form of artificial neural networks, to forecast wind speeds in the upper atmosphere. Weather data was obtained from the National Oceanic and Atmospheric Administration. The data was split by the defined pressure layers, then treated as a video frame. The network was then trained and tested to predict the next frame in the video, or in this case the weather conditions in that pressure layer one-time step in the future. This allows the ability to construct a model forecasting future weather conditions across a single pressure layer, around the entire globe. This can potentially reduce the computation time of current models, and increase accuracy. These results are then used to predict weather conditions at varying altitudes within a region encompassing the two travel locations. A shortest path algorithm is then employed to find the most fuel-efficient route between the two points.