Atmospheric Visibility and Cloud Ceiling Predictions With Hybrid IIS-LSTM Integrated Model: Case Studies for Fiji’s Aviation Industry

Abstract

Atmospheric visibility and cloud ceiling forecasts are essential for the safety and efficiency of flight operations and the aviation industry. Routine hourly aviation meteorological observations are recorded at every airport. However, forecasts of these two meteorological parameters using artificial intelligence techniques are limited. This research utilizes data from two study sites in Fiji, Nadi, and Nausori International Airport, and proposes a hybrid Iterative Input Selection – Long Short-Term Memory (IIS-LSTM) integrated model to forecast the consecutive hour’s visibility and ceiling parameters. The IIS algorithm acts as a feature selector from the global predictor matrix of predictor variables with its significant lagged inputs and the significant lagged inputs of the target variable, while the LSTM algorithm acts as the learning model and makes forecasts. The performance of the proposed hybrid IIS-LSTM model is evaluated using seven statistical score metrics and compared with four competing benchmark models. The evaluated results illustrate the superiority of the proposed hybrid IIS-LSTM integrated model and its advanced capability to generate accurate atmospheric visibility and cloud ceiling forecasts for the next consecutive hour compared to the benchmark models. The most important features selected were the second lagged input of visibility and first lagged input of rainfall to improve visibility forecasts while the first and the fifth lagged inputs of the total low cloud cover were paramount for accurate cloud ceiling forecasts. Considering the geography of the study sites, the overall efficacy of the IIS method is strongly advocated to screen most suitable model predictors and the subsequent integration of this input selection method with the LSTM predictive algorithm to attain enhanced performance of the hybrid IIS-LSTM forecast model. This objective model is therefore proposed to be an efficient and cost-effective predictive tool for atmospheric visibility and cloud ceiling forecasts, especially its applications in the aviation industry for aeronautical purposes.

Keywords

• Visibility forecast
• ceiling forecast
• deep learning
• machine learning
• iterative input selection
• long short-term memory