Atmospheric visibility and cloud cover forecasting with novel artificial intelligence methods for Fiji's aviation sector

Atmospheric visibility and cloud cover forecasting with novel artificial intelligence methods for Fiji's aviation sector

Visibility and ceiling are two important meteorological parameters affecting the operation of aircraft, especially during the critical phases of take-off and landing at airports. Apart from being a key factor in safe and efficient flight operations, accurate forecasts of these two meteorological parameters also contribute to improving the economics of air transportation. This Master of Research (MRES) study aims to develop a new forecasting model based on the latest artificial intelligence methods to predict atmospheric visibility and cloud cover (or ‘ceiling’). This study will address existing gaps in the area by advancing the practical application of deep learning with the following three objectives. Firstly, the study adopts the Iterative Input Selection (IIS) feature selection technique to deduce the optimum features for the proposed model from a global pool of features. Secondly, it aims to design and implement the proposed hybrid IIS-LSTM integrated model for a 1-hour forecast horizon and further compare the outcomes with four alternative AI models. Thirdly, the performance of the hybrid IIS-LSTM model is compared with the alternative models using performance evaluation metrics and graphical analysis. The study also elaborates on the suitability of the objective model for practical visibility and ceiling forecasts and discusses limitations to provide recommendations for future research. The objectives are achieved by using aeronautical meteorological data from two international airports in Fiji from 2012 to 2021. The proposed hybrid IIS-LSTM integrated model combines the feature selection characteristics of the IIS algorithm and the effective time series forecasting LSTM model. The model achieved the desired outcomes of the research by isolating key features for each study site. These optimum features maximised the efficiency of the forecasting component of the algorithm by reducing dimensionality and increasing generalisability of the model. The performance of this model showed its reliability in making accurate forecasts and was consistent for both study sites and for both target variables. It achieved the highest agreement metrics (Willmott’s Index) against the comparison model and the lowest error metrics (RMSE). The model’s performance against these benchmark models demonstrated its superiority over these models and further endorses it as a reliable practical tool. Therefore, the research outcomes present the proposed model as a useful practical tool for future implementation in the aviation industry and could enable a better understanding of the visibility and ceiling parameter predictions for this study region in the future.

File reproduced in accordance with the copyright policy of the publisher/author/creator.