Evaluating CNN-BiLSTM Models for LOD Predictions

The term “Earth Orientation Parameters” (EOPs) refers to parameters that indicate variations in how the Earth rotates. Numerous real-life applications, such as spacecraft navigation, weather forecasting, and disaster management, depend on correct, short-term forecasts of Earth orientation parameters (EOP). The hardest component of the EOP to forecast is the LOD (Length Of Day), which is a measure of variations in Earth's rotation rate. This study aims to evaluate CNN-BiLSTM to forecast and model the time series of LOD. The methodology can be described as follows: The trend components are first removed using polynomial curve fitting and least-square methods. After that, the obtained residuals are used in the CNN-BiLSTM model for predictions. The predictions are then evaluated for mean absolute error using different input sizes.