NN‐MLT Model Prediction for Low‐Latitude Region Based on Artificial Neural Network and Long‐Term SABER Observations

• Published in: Earth and space science (Hoboken, N.J.) Vol. 10; no. 6
• Main Authors: Lingerew, Chalachew, Jaya Prakash Raju, U., Guimarães Santos, Celso Augusto
• Format: Journal Article
• Language: English
• Published: American Geophysical Union (AGU) 01.06.2023
• Subjects: MLT; model performance; NN‐MLT; NRLMSIS2‐0; prediction; TIMED/SABER
• ISSN: 2333-5084; 2333-5084
• DOI: 10.1029/2023EA002930

The low‐latitude mesosphere and lower thermosphere (MLT) regions are distinct and, highly turbulent transition zones in Earth's atmosphere. The scarcity of reliable measurements makes continuous monitoring of these areas challenging. Therefore, the necessity for studies focused on the MLT region cannot be overstated, as they are essential for developing effective models that meet the accuracy requirements of satellite‐based observations. The neural networks NN‐MLT model, developed using 15 years of Thermosphere, Ionosphere, and Mesosphere Energetics and Dynamics/satellite, equipped with Broadband Emission Radiometry (SABER) observed temperature data spanning from January 2006 to December 2020, employs neural network techniques. The data set was split, with 90% used for training and the remaining 10% allocated for prediction. The model's validation was tested with two other partitions (80(20) and 70(30)). The 90(10) partition, exhibiting a high correlation coefficient (R), low standard deviation (σ), and low root mean square error (RMSE), demonstrated the model's good performance. As clearly shown from statistical metrics (R, RMSE, mean, and σ) at three specific altitude levels (60, 75, and 90 km), the NN‐MLT model's performance aligns closely with the empirical model (NRLMSISE2‐0) and SABER observations. The NN‐MLT model displays a high R (0.74) and low RMSE (4.35 K) at 60 km, indicating its effective performance compared to the other two heights of 75 and 90 km. The NN‐MLT model's spatiotemporal variability in MLT temperature prediction agrees well with the SABER data at all altitudes, particularly at 60 km. While the NN‐MLT model accurately captures the seasonal variations of MLT temperature, the analysis leads to the conclusion that it consistently outperforms the empirical model and aligns closely with observations. Plain Language Summary The middle atmosphere serves as a connection between the lower and upper atmospheres through wave dynamics. The mesosphere and lower thermosphere (MLT) region, often called the “ignosphere,” is challenging to study due to the scarcity of reliable and continuous measurements. In this paper, the first attempt to develop a low‐latitude NN‐MLT model based on neural network techniques is presented using 15 years of SABER observation data. The model effectively captures the spatiotemporal variations and fits well with the standard empirical model (NRLMSIS2‐0) and SABER observations. Key Points A neural networks‐mesosphere and lower thermosphere (NN‐MLT) model based on artificial NN is developed using Thermosphere, Ionosphere, and Mesosphere Energetics and Dynamics/satellite, equipped with Broadband Emission Radiometry observations to predict MLT temperature variability The newly developed NN‐MLT model shows potential in accurately predicting MLT temperature variability over low latitudes The NN‐MLT model can significantly enhance the prediction capabilities for temperatures in the MLT