水汽是大气中重要的组成部分, 实现大气中水汽含量的高精度反演对气象研究有着重要意义.本文利用FY-3D卫星微波成像仪(FY-3D/MWRI)在2019-2022年间每年7月份的亮温资料, 根据随机森林算法, 以太平洋区域ERA5水汽数据为参考, 分别建立海上晴空大气可降水的六通道、八通道随机森林反演模型(RF6、RF8).试验结果表明, 相较于经验回归反演模型, 随机森林反演模型精度有明显提高, 其中RF6模型精度提高了约22%, RF8模型精度提高了约28%.进一步, 将基于太平洋区域数据训练生成的RF6、RF8反演模型应用于北大西洋与南印度洋的大气可降水反演, 也得到了较好的结果.因此, 海上晴空大气可降水的随机森林反演模型具有较好的稳定性和普适性, 且RF8模型优于RF6模型, RF6模型优于经验回归模型.

Water vapor is an important part of the atmosphere, and it is of great significance to realize the high-precision retrieval of water vapor content in the atmosphere for meteorological research. This paper used brightness temperature data of FY-3D/MWRI in July from 2019 to 2022 annually, with the water vapor data of ERA5 on Pacific as references, established six-channel and eight-channel random forest retrieval models (RF6 and RF8) for total precipitable water in maritime clear sky, based on the random forest algorithm. The experimental results indicated that compared to the empirical regression retrieval model, the random forest retrieval model have improved accuracy obviously, with the RF6 model achieving an improvement of about 22% and the RF8 model achieving an improvement of about 28%. Furthermore, when applying the RF6 and RF8 models which trained based on Pacific region data to the North Atlantic and South Indian Ocean, positive retrieval results were also obtained. Considering all factors, the RF8 model outperforms the RF6 model, and the RF6 model outperforms the empirical regression model.