【摘要】 针对北极融冰期的海冰密集度反演，并改善全球导航卫星系统反射测量（GNSS-R）对海水的海冰密集度高估问题，本文提出一种利用机器学习算法生成高时空分辨率的融冰期海冰密集度估算方法，提取GNSS-R时延多普勒图（DDM）的特征参数，并结合海表温度数据建立LightGBM模型，将反演结果与参考海冰密集度值进行相关性分析和评估。本文的模型结果与OSI SAF的海冰密集度产品显示出较好的一致性，相关系数、平均绝对误差和均方根误差分别为0.965、0.061和0.090。该方法能够实现对北极海冰边缘区的海冰密集度高精度估计。更多还原

【Abstract】 In this paper, a high spatial-temporal resolution sea ice concentration estimation method for the Arctic melting season is proposed, aiming to improve the overestimation of sea ice concentration in seawater by the Global Navigation Satellite System-Reflectometry(GNSS-R). The method utilizes machine learning algorithms to extract feature parameters from the Delay Doppler Maps(DDM) obtained through GNSS-R and combines them with sea surface temperature data to establish a LightGBM model. The inversion results are then subjected to correlation analysis and evaluation against reference sea ice concentration values. The model’s performance is compared with the sea ice concentration product from OSI SAF, demonstrating good consistency, with correlation coefficient,mean absolute error, and root mean square error being 0.965, 0.061, and 0.090, respectively. This approach enables high-precision estimation of sea ice concentration in the Arctic marginal ice zone.更多还原