Sensitivity of different optimization solvers in LSTM algorithm for temperature forecast over Mars at Jezero Crater landing site

Abstract

Exact forecast of surface temperature over MARS is important and critical. Surface temperature is fundamental to the environmental parameter that has a direct impact on designing and operating the land rovers that explore the MARS planet. In this paper, We used well known long Short-Term Memory (LSTM) algorithm to build a data-driven model to predict the surface temperature over the planned landing site Jezero Crater for Mars 2020 Rover. The data-driven model is built using a dataset based on the Mars Climate Database (MCD) which derived from the Global Climate Model (GCM) simulations for MARS. The temporal availability of this data from martian year 24 to 33. we evaluated the effect of the three different optimization solvers on surface temperature prediction over the landing site Jezero Crater for two different numbers of epochs. The solver that provides the lowest RMSE is used to predict the surface temperature over the landing site from martian year 34 to martian year 36.