Seq2Seq Model-Based Augmentation of Atmospheric Microwave Remote Sensing Data

• Main Authors: Wu, Peng, Liu, Zhifu, Wu, Changzhe
• Format: Conference Proceeding
• Language: English
• Subjects: Atmospheric modeling; Data Augmentation; Data models; Long Short-Term Memory; Microwave amplifiers; Microwave integrated circuits; Microwave Radiometer; Microwave theory and techniques; Recurrent Neural Networks; Sequence to Sequence; Temperature distribution; Temperature sensors
• ISSN: 2994-3124
• DOI: 10.1109/ICMMT61774.2024.10672036

Meteorological observation data play an important role in weather prediction. However, some areas are difficult to produce high-resolution background fields due to the lack of meteorological observation data, which affects the accuracy of meteorological prediction. In order to solve this problem, this paper proposes a method for amplifying atmospheric microwave remote sensing data based on a sequence-to-sequence model. In this study, the MP3000A microwave radiometer is used as the experimental platform, and 30330 temperature data at 58 altitudes from September 1 to October 30, 2023 in Harbin are selected as the experimental data. Firstly, two groups of sequence-to-sequence models were constructed with the long and short-term memory neural network and recurrent neural network as the basic unit respectively, and it was found that the validation loss of the first group of network structure was smaller through screening experiments. Secondly, the first group of models is used to amplify the low-resolution temperature data. The experimental results show that the difference between the amplified temperature data and the true value is very small, with a mean square error (MSE) of 1.8768 and a mean absolute error (MAE) of 0.9533. It indicates that the model can capture the temperature trend very well in realizing the temperature data amplification.