Abstract Accurate, spatiotemporally continuous total precipitable water (TPW) data under all‐weather conditions are crucial for understanding water and energy cycles. This study introduces TPWDiff‐CB, a novel deep learning‐based TPW retrieval model that employs a generative diffusion model. TPWDiff‐CB effectively estimates TPW under all‐weather conditions by leveraging thermal infrared observations from the Advanced Himawari Imager aboard Himawari‐8. Specifically, when compared to radiosonde TPW, TPWDiff‐CB yields a correlation coefficient (R) of 0.98 and a root mean square error of 4.51 mm under all‐weather condtions. The model demonstrates exceptional performance, maintaining near‐identical accuracy under both cloudy and clear‐sky conditions. Its robust capability in learning the TPW distribution and performing spatiotemporal retrievals surpasses that of traditional machine learning model. These findings highlight TPWDiff‐CB’s high accuracy and its promising potential for applications in weather and climate research.

Read original article

Read original article