Improving Electron Density Predictions in the Topside of the Ionosphere Using Machine Learning on In Situ Satellite Data
Improving Electron Density Predictions in the Topside of the Ionosphere Using Machine Learning on In Situ Satellite Data
复制标题
DOI：
10.1029/2022sw003134
复制
发表时间：
2022-08
期刊：
影响因子：
--
通讯作者：
S. Dutta;M. Cohen
中科院分区：
文献类型：
--
作者：
S. Dutta;M. Cohen
对地球电离层建模是预测空间天气的一个关键组成部分，而空间天气又会影响无线电波传播、导航和通信。这项研究的重点是利用卫星数据，特别是来自国防气象卫星方案的数据，预测电离层顶面的电子密度，国防气象卫星方案是一个由19颗卫星组成的集合，这些卫星在极轨道上绕地球运行了不同的时间，时间完全涵盖了1982年至今。开发了一个人工神经网络，并在两个太阳周期的数据（113卫星年）上进行了训练，沿着全球驱动因素和指数，如F10.7，行星际磁场和Kp，以生成电子密度预测。我们在随后6年的数据（26个卫星年）上测试了该模型，发现相关系数为0.87。一旦经过训练，该模型可以预测在给定当前/最近地磁条件下由纬度和经度指定的任何位置处的顶侧电子密度。我们通过与在相似高度轨道上运行的DEMETER卫星的数据进行比较，并将其作为真实电子密度值的独立来源，验证了该模型。将我们的结果与国际参考电离层进行比较，我们发现我们的模型在低纬度到中纬度以及安静和中度干扰的地磁条件下工作得更好，但不适用于高度干扰的条件。
Modeling the Earth's ionosphere is a critical component of forecasting space weather, which in turn impacts radio wave propagation, navigation and communication. This research focuses on predicting the electron density in the topside of the ionosphere using satellite data, in particular from the Defense Meteorological Satellite Program, a collection of 19 satellites that have been polar orbiting the Earth for various lengths of times, fully covering 1982 to the present. An artificial neural network was developed and trained on two solar cycles worth of data (113 satellite‐years), along with global drivers and indices such as F10.7, interplanetary magnetic field, and Kp to generate an electron density prediction. We tested the model on six years of subsequent data (26 satellite‐years), and found a correlation coefficient of 0.87. Once trained, the model can predict topside electron density at any location specified by latitude and longitude given current/recent geomagnetic conditions. We validated the model via comparison with data from the DEMETER satellite which orbited at a similar altitude, and taking that as an independent source of true electron density values. Comparing our results to the International Reference Ionosphere, we find that our model works better at low to mid‐latitudes, and for quiet and moderately disturbed geomagnetic conditions, but not for highly disturbed conditions.