Problem Statement: Extreme temperatures affect flight schedules and the maintenance of aircraft. One of the major concerns for major airlines around the world is how temperature changes affect the functioning of aircraft ultimately resulting in loss of revenue and increasing costs. As airliners have a maximum or a minimum operating temperature beyond which they would ground flights or reschedule, having information about how the temperatures would be would bring in a lot of value and cost savings.
Hence, the analysis aims to provide a forecast of both the maximum and minimum temperatures monthly for aircraft so that they can better schedule their flights. As major airliners have global operations the analysis has been done for three major cities: Tokyo, Paris and Los Angeles. The model provides a 12-month forecast as airlines are well equipped with the information. Also, as they require the data well in advance to plan the schedules the model will provide a 6-month lead time.
Data Description: The data that is available is for three major cities: Tokyo, Paris and Los Angeles. The locations have been chosen strategically as they have one of the largest airline traffic in the world and have a number of airline operations and service centers. Data has been collected from 2000 to 2012 and consists of both maximum and minimum temperature for each city. There is also monthly data for each of the year (without any missing values and in degree Celsius) that will be used for the analysis.
Each of the plots have evident time series components present (Appendix 1). In addition to noise and level at some points for the plots, all of them have seasonality and a slightly varying level. All these factors will be considered while analyzing the data and then decide on which model would be used to predict the temperatures. The data has been obtained from the website of the Department of Statistics at The University of Auckland.
Forecasting Goal & Model Used: The goal is to forecast the maximum and minimum monthly temperatures for each of the cities over a period of 12 months. The model also accounts for the 6- month lead time that needs to be provided for the airliners. As there are different time series components present in the plot for the three cities, the analysis has shown that Holt-Winters Additive model would be used for Los Angeles, while Linear Regression model will be used for Tokyo and Paris. The RMSE and MAPE with the seasonality naïve prediction have been compared with those of the chosen models and it is found that they perform better on both prediction and validation data.
Conclusions & Recommendations: The 18-month forecast that is provided to the airliners can be used to make any schedules or operations optimizations in advance. This not only provides a better- informed decision for the airliners but also ensures that revenues leakages and cost overruns are avoided and saved. Airliners can use the data and compare them with the operating temperatures of their aircraft and then make decisions.
However, there are some implications when there the temperatures are under or over-forecasted. For each half a degree of Celsius predicted below, there is another minute added to the flight. This adds to extra time on air increasing fuel consumption. Also, if there is data for daily temperatures, predictions can be made on a daily basis that will be of higher value to the stakeholders. Since airline prices are dynamic and change regularly based on how the temperatures fluctuate (as mentioned earlier differing temperatures affect flying times and eventually change fuel prices), providing accurate prediction is important. To avoid the costs of wrong forecasting, the models will be run every month and provide updated forecasts for the airliners.