基于随机森林算法的中国及周边区域电离层f_(o)F_(2)预测模型

A random forest-based prediction model for ionospheric f_(o)F_(2) in China and surrounding regions

电离层F2层的临界频率(f_(o)F_(2))的平方与峰值电子密度(N_(m)F2)成正比,是影响GNSS性能的关键参数之一,提升电离层f_(o)F_(2)的预测精度对于优化GNSS广播电离层模型性能并提升GNSS的定位精度具有重要意义.本文基于中国及周边区域的18个测高仪台站和COSMIC(constellation observing system for meteorology,ionosphere,and climate)掩星观测数据,综合考虑世界时、年积日、地理位置、太阳和地磁活动等多维特征,利用随机森林(random forest,RF)算法构建了电离层f_(o)F_(2)预测模型.通过与国际参考电离层(international reference ionosphere,IRI)-2020模型对比分析,验证了该模型的预测精度.研究结果表明,与IRI国际无线电咨询委员会(International Radio Consultative Committee,CCIR)和IRI国际无线电科学联盟(International Union of Radio Science,URSI)模型相比,RF模型的平均绝对误差(mean absolute error,MAE)分别降低了14.81%和17.11%,均方根误差(root mean squared error,RMSE)分别降低了11.21%和13.14%.此外,该模型在不同纬度、地方时、太阳活动和地磁活动条件下,均展现出优于IRI-2020的预测精度.本研究不仅有效提升了中国及周边区域电离层f_(o)F_(2)的预测精度,还为提高GNSS的准确性和可靠性奠定了重要基础.

The square of the critical frequency of the ionospheric F_(2) layer(f_(o)F_(2))is proportional to the peak electron density(N_(m)F_(2))and serves as a crucial parameter affecting the performance of Global Navigation Satellite Systems(GNSS).Enhancing the prediction accuracy of f_(o)F_(2) is essential for optimizing GNSS broadcast ionospheric models,thereby improving the positioning accuracy of GNSS.This study develops an ionospheric f_(o)F_(2) prediction model for China and its surrounding regions using the random forest algorithm,based on data from 18 ionosonde stations of the China Research Institute of Radiowave Propagation and COSMIC occultation observations.The model incorporates multiple features,including Universal Time,day of the year,geographic location,solar,and geomagnetic activities.A comparative analysis with the International Reference Ionosphere(IRI-2020)model validates the prediction accuracy of our model.The results indicate that the random forest model reduces the mean absolute error by 14.81%and 17.11%,and the root mean square error by 11.21%and 13.14%,compared to the IRI CCIR and IRI URSI models,respectively.Additionally,the model exhibits superior prediction accuracy under various latitudes,local times,solar,and geomagnetic activity conditions when compared to IRI-2020.This research not only significantly enhances the f_(o)F_(2) prediction accuracy for China and its surrounding regions but also lays a critical foundation for improving the accuracy and reliability of GNSS globally.