Automation of takeoff data for aviation services using self-supervised LSTM approaches with time-series prediction

• Published in: Modeling earth systems and environment 2024-08, Vol.10 (4), p.5409-5425
• Main Authors: Shankar, Anand, Sarthi, Pradhan Parth, Singh, Deepak Kumar, Kumar, Mantosh, Kumar, Pankaj
• Format: Article
• Language: English
• Subjects: Aircraft configurations; Aircraft landing; Airports; Algorithms; Artificial neural networks; Automation; Aviation; Chemistry and Earth Sciences; Computer Science; Correlation coefficient; Correlation coefficients; Dew point; Earth and Environmental Science; Earth Sciences; Earth System Sciences; Economic models; Economics; Ecosystems; Environment; Flight; Flight planning; Fog; Landing behavior; Math. Appl. in Environmental Science; Mathematical Applications in the Physical Sciences; Mean sea level; Meteorological parameters; Multivariate analysis; Neural networks; Numerical weather forecasting; Original Article; Parameters; Physics; Predictions; Runways; Sea level pressure; Statistics for Engineering; Takeoff; Temperature; Temperature requirements; Time series; Weather; Weather forecasting; Weight; Wind; Wind direction; Wind speed
• ISSN: 2363-6203; 2363-6211
• DOI: 10.1007/s40808-024-02070-8

Landing and takeoff are the most crucial phases of any flight; in particular, the takeoff configuration of an aircraft is a delicate balance between the regulated takeoff weight, runway length available, and prevailing meteorological conditions to execute a safe takeoff. Apart from the fixed parameters like runway length, four variable meteorological parameters, i.e., wind, temperature, pressure, and visibility, govern the total permissible takeoff weight and thus influence the economic or commercial viability of any flight. Therefore, an accurate assessment of these parameters, which are collectively called takeoff data, is an important responsibility of any meteorological office associated with flight planning and operation at any airport. It is suggested in this research paper that instead of forecasters making decisions based on numerical weather prediction (NWP) models, multivariate self-supervised LSTM-based models should be used to make accurate predictions of temperature and pressure (mean sea level pressure or MSLP) parameters of takeoff data. As the proposed prediction algorithms are based on deep neural networks (LSTM), they are relatively easy to develop and require less time and resources. Based on this approach, a Nowcast of temperature and pressure for the next one to six hours could be generated using the time series multivariate dataset for the temperature and pressure of the airports (representative station: Patna Airport) with input features including date and time, temperature, atmospheric pressure, humidity, dew point temperature, wind direction, wind speed, cloud amount, and present and past weather (like fog, dense fog, drizzle, etc.). The input features were selected based on the best Pearson correlation coefficients. The predicted temperatures and pressure (MSLP) are compared with the observed temperatures, and the models' efficacy and accuracy are assessed based on the different performance measures. The ultimate objective of this approach is to develop and deploy robust and dynamic self-supervised LSTM models to automate the crucial takeoff data that is so crucial for the safe and economic operation of any flight.