Radiosonde data collected from 83 stations in China from January to December 2012 were used to evaluate Moderate Resolution Imaging Spectroradiometer (MODIS) near-infrared (NIR) and thermal infrared (IR) total precipitable water vapour (PWV) products. The results indicate that MODIS NIR PWV products shows better agreement with radiosonde data than with IR PWV products, with the correlation coefficients up to 0.95. The root mean square errors (RMSEs) of NIR PWV range from 2 to 8 mm with different stations, which shows significant regional differences over China. The mean RMSE is about 5.03 mm (~35%) with a positive deviation of 2.56 mm (~18%), indicating the occurrence of a slight overestimation. Moreover, MODIS IR PWV during night-time has a better agreement with radiosonde PWV than that during daytime. The mean RMSE of IR PWV during daytime was ~6.02 mm (~42%), with a positive deviation of 1.54 mm (~11%). The mean RMSE of IR PWV during night-time was ~5.81 mm (~40%), with a negative deviation of approximately −0.04 mm (~0.25%). Both the NIR and IR PWV products during daytime tend to be higher than radiosonde PWV.

The Visible and Infrared Spin-Scan Radiometer (VISSR) onboard the Fengyun-2 (FY-2) satellite can provide valuable thermal infrared observations to help create a precipitable water vapor (PWV) product with high spatial and temporal resolutions. The current FY-2/VISSR PWV product in operation is produced by using a traditional two-band physical split-window (PSW) method, which produces low quality results under dry atmospheric conditions. Based on the sensitivity characteristics of FY-2F/VISSR water vapor channel and two split-window channels to atmospheric water vapor, this study developed a new, robust operational PWV retrieval algorithm for FY-2F to improve the operational precision of the current PWV product. The algorithm uses a modified three-band PSW method, which adds a scale for the water vapor channel in the improved three-band PSW method. Integrated PWV products from the radiosonde data in 2016 are used here to validate the precision of the PWV retrieved by the modified three-band and traditional two-band PSW methods. The mean bias, root mean square error (RMSE), and correlation coefficient of the PWV retrieved by the modified three-band PSW method are 0.28 mm, 4.53 mm, and 0.969, respectively. The accuracy is much better than the PWV retrieved by the two-band method, whose mean bias, RMSE, and correlation coefficient are 12.67 mm, 29.35 mm, and 0.23. Especially, in mid- or high-latitude regions, the RMSE of the PWV is improved from 10 to 2 mm by changing the inversion in the two-band method to the modified three-band PSW method. Furthermore, the modified three-band PSW results show a better consistency with the radiosonde PWV at any zonal belt and season than the two-band PSW results. This new algorithm could significantly improve the quality of the current FY-2F/VISSR PWV product, especially at sites where the actual PWV are lower than 15 mm.

Nowcasting is a critical technology for disaster prevention and mitigation, and the accuracy of radar echo extrapolation directly impacts forecasting performance. In most deep learning-based models, accurately predicting heavy precipitation remains a challenging task. Focusing on the region of China, this study proposes an improved model based on residual and attention mechanisms—RA-UNet—for precipitation nowcasting with a lead time of 3 h. The model introduces the residual neural network (ResNet) and the convolutional block attention module (CBAM) to integrate multi-scale features into the U-Net encoder–decoder architecture, enhancing its ability to capture the spatiotemporal evolution of precipitation systems. Meanwhile, depthwise separable convolutions are employed to replace conventional convolutions, significantly improving computational efficiency while preserving model performance. To evaluate the model’s performance, experiments were conducted using 6 min resolution radar echo data from China in 2024, with comparisons made against the optical flow (OF) method and the U-Net model. The experimental results show that RA-UNet demonstrates significant advantages in 3 h forecasting: its mean absolute error (MAE) is reduced by approximately 7%, the false alarm rate (FAR) decreases by about 20%, and it outperforms the comparison models in metrics such as the critical success index (CSI) and structural similarity index (SSIM). Notably, RA-UNet effectively mitigates intensity degradation in long-term forecasts, successfully predicting the trend of >40 dBZ strong echo cores in two typical cases and significantly improving the premature dissipation problem of precipitation fields. This study provides a new approach to refined forecasting of complex precipitation systems, and future work will combine multi-source data fusion with physical constraint mechanisms to further enhance precipitation event prediction capabilities.

This study proposed an instantaneous summer air temperature (T_air) estimation model using the Himawari-8 Advanced Himawari Imager (AHI) brightness temperatures (BTs) in split-window channels and other auxiliary data. Correlation analysis and stepwise linear regression were used to select the predictors for T_air estimation. Nine predictors such as AHI BTs in channels 14 and 15, elevation, precipitable water vapor (PWV), and relative humidity (RH) were finally selected. Stepwise linear regression and neural network (NN) methods were applied to construct summer T_air estimation models over China, respectively. The estimated T_air by linear and NN models was evaluated using the observed T_air from 272 meteorological stations over China. The results showed that AHI BTs in channels 14 and 15, elevation, PWV, and RH were more important for T_air estimation than other predictors. The accuracy of the NN models was better than the linear models. The correlation coefficient (R), root mean square error (RMSE), and bias were 0.97, 1.72°C, and 0.04°C, respectively, for the NN model and were 0.89, 3.28°C, and 0.07°C, respectively, for the linear model. About 75.6% of the T_air differences by the NN model were within 2.0°C, and even 45.8% were within 1.0°C. The performance of the T_air estimation model for each site was also investigated, and the accuracy of T_air estimation in southeast China is better than northwest China.