A Study on Deep Learning Approaches for Mars Weather Forecasting

Forecasting surface temperature over Mars is one of the most popular research areas in planetary science as it is essential to obtain information about a planet’s meteorological data without directly landing on its surface. This paper presents an in-depth exploratory analysis and prediction of Mars temperature as a time series problem. A Residual CNN-LSTM is devised to predict the temperature for the year 2018 based on the temperatures recorded in 2012–2017. The experimental results aim to explore four scenarios: predicting maximum temperature using terrestrial dates, predicting minimum temperature using terrestrial dates, and the remaining two aspects predict the same using Solar Days (SOL). The proposed model is compared with other commonly used approaches, and it shows superiority in terms of Root Mean Squared Error (RMSE).