Near real-time LOD prediction using ConvLSTM model through integrating IGS rapid LOD and effective angular momentum

• Published in: Geo-spatial Information Science
• Main Authors: Kehao Yu, Xiaoya Wang, Zhao Li, Jian Wang, Lihua Li, Weiping Jiang, Kai Yang, Jingwen Long, Yunzhao Li
• Format: Article
• Language: English
• Published: Taylor & Francis Group 2025-03-01
• Subjects: ConvLSTM; LOD prediction; effective angular momentum; IGS rapid LOD; Earth orientation parameters
• Online Access: https://www.tandfonline.com/doi/10.1080/10095020.2025.2471432
• ISSN: 1009-5020; 1993-5153

Length of day (LOD), a critical component of Earth orientation parameters (EOP), represents variations in Earth’s rotation rate. It is very difficult to predict accurately due to the effects of atmosphere, ocean, hydrology, the Earth’s internal interactions and so on. The international Earth rotation and reference systems service (IERS) EOP C04 series, derived from four space geodetic observations, could offer high accuracy and smooth EOP product. However, this product typically has a latency of about 30 days. It is not adequate for fields requiring strict real-time data processing and applications, such as precise tracking and navigation of interplanetary spacecraft, global navigation satellite system (GNSS) meteorology, real-time precision orbit determination of artificial satellites, real-time kinematic (RTK) positioning and so on. To address the aforementioned issues, we propose an algorithm for predicting LOD that adopts a convolutional long-short-term memory (ConvLSTM) method with different base sequence lengths based on the LOD series from the IERS EOP C04, effective angular momentum (EAM) datasets and GNSS near-real-time (NRT) LOD data from the International GNSS Services (IGS) Rapid Products. Compared to the most accurate models used by participants in the Second Earth Orientation Parameters Prediction Comparison Campaign (2nd EOP PCC), when GNSS NRT data is not used, the proposed model improves LOD ultra-short-term (1–10 days) prediction accuracy by 29.72% and medium- to long-term (60–360 days) prediction accuracy by 11.86%. After incorporating GNSS NRT data, the short-term (10–30 days) LOD prediction accuracy improves by 55.07%. It is shown that the ConvLSTM model, integrated with GNSS NRT data and EAM datasets, could significantly enhance the forecast accuracy of LOD across various time spans. This advancement enriches the Earth’s rotation prediction models and holds potential benefits for real time applications such as real-time satellite orbit determination, extreme weather analysis, RTK technology and so on.