The ice water path (IWP) is an important cloud parameter in atmospheric radiation, and there are still great difficulties in its retrieval. Artificial neural networks have become a popular method in atmospheric remote sensing in recent years. This study presents a global IWP retrieval based on deep neural networks using the measurements from the Microwave Humidity Sounder (MWHS) aboard the FengYun-3B (FY-3B) satellite. Since FY-3B/MWHS has quasi-polarization channels at 150 GHz, the effect of the polarimetric radiance difference (PD) was also studied. A retrieval database was established using collocations between MWHS and CloudSat 2C-ICE (CloudSat and CALIPSO Ice Cloud Property Product). Then, two types of networks were trained for cloud scene filtering and IWP retrieval. For the cloud filtering network, the microwave channels show a capacity with a false alarm ratio (FAR) of 0.31 and a probability of detection (POD) of 0.61. For the IWP retrieval network, different combination inputs of auxiliaries and channels were compared. The results show that the five MWHS channels combined with scan angle, latitude, and the ocean/land mask of inputs of auxiliary variables perform best. Applying the cloud filtering network and IWP retrieval network, the final root mean squared error (RMSE) is 916.76 g m⁻², the mean absolute percentage error (MAPE) is 92 %, and the correlation coefficient (CC) is 0.65. Then, a tropical cyclone case measured simultaneously by MWHS and CloudSat was chosen to test the performance of the networks, and the result shows a good correlation (0.73) with 2C-ICE. Finally, the global annual mean IWP of MWHS is very close to that of 2C-ICE, and the 150 GHz channels give a significant improvement in the midlatitudes compared to using only 183 GHz channels.

中文翻译：

---

基于深度神经网络的风云三号B（FY-3B）卫星偏振测量微波测湿仪（MWHS）的冰水路径检索

冰水路径（IWP）是大气辐射中重要的云参数，其反演难度较大。近年来，人工神经网络已成为大气遥感中的一种流行方法。本研究使用风云三号 B (FY-3B) 卫星上的微波湿度探测仪 (MWHS) 的测量值，提出了基于深度神经网络的全球 IWP 检索。由于 FY-3B/MWHS 在 150 GHz 具有准极化通道，因此还研究了极化辐射差 (PD) 的影响。使用 MWHS 和 CloudSat 2C-ICE（CloudSat 和 CALIPSO Ice Cloud Property Product）之间的搭配建立了检索数据库。然后，针对云场景过滤和 IWP 检索训练了两种类型的网络。对于云过滤网络，微波通道的容量具有 0.31 的误报率 (FAR) 和 0.61 的检测概率 (POD)。对于 IWP 检索网络，比较了辅助和通道的不同组合输入。结果表明，5个MWHS通道结合扫描角度、纬度和辅助变量输入的海洋/陆地掩膜表现最好。应用云过滤网络和IWP检索网络，最终均方根误差（RMSE）为916.76 g m^-2，平均绝对百分比误差 (MAPE) 为 92 %，相关系数 (CC) 为 0.65。然后，选择 MWHS 和 CloudSat 同时测量的热带气旋案例来测试网络的性能，结果显示与 2C-ICE 具有良好的相关性（0.73）。最后，MWHS 的全球年平均 IWP 非常接近 2C-ICE，与仅使用 183 GHz 通道相比，150 GHz 通道在中纬度地区有显着改善。