Numerical weather forecast models have biases caused by insufficient grid resolution and incomplete physical processes, especially near the land surface. Therefore, the Japan Meteorological Agency (JMA) has been operationally post-processing the forecast model outputs to correct biases. The operational post-processing method uses a Kalman filter (KF) algorithm for surface temperature prediction. Recent reports have shown that deep convolutional neural networks (CNNs) outperform the JMA operational method in correcting temperature forecast biases. This study combined the CNN-based bias correction scheme with the JMA operational KF algorithm. We expected that the combination of CNNs and a KF would improve the post-processing performance, as the CNNs modify large horizontal structures, and then, the KF corrects minor spatiotemporal deviations. As expected, we confirmed that the combination outperformed both CNNs and the KF alone. This study demonstrated the advantages of the new method in correcting coastal fronts, heat waves, and radiative cooling biases.

In this study, we investigated the feasibility of rain enhancement by cloud seeding over a target area (the Sameura Dam catchment area, Kochi Prefecture) in early summer. The effects of salt micro-powder (MP) and hygroscopic flare (HF) seeding on the initial cloud microphysical structures were investigated using a detailed bin microphysics parcel model with background atmospheric aerosol data collected from ground-based observations conducted on the windward side of the target area and seeding aerosol data collected from the coordinated flights of seeding helicopter and in-situ measurement aircraft. Numerical seeding experiments showed that the size distributions of cloud droplets were broadened, and the onset of raindrop formation was accelerated by MP and HF seeding, although MP seeding showed more notable seeding effects than did HF seeding. MP seeding increased the mean droplet size and decreased the total number concentration of cloud droplets, whereas HF seeding had the opposite effect. Based on the relationship between the increase/decrease ratio of the cloud droplet number concentration and increase/decrease ratio of the surface precipitation by hygroscopic seeding obtained in previous studies, MP seeding had a positive seeding effect, whereas HF seeding had a negative effect. In the numerical seeding experiments, a range of variations in the number concentration and hygroscopicity of background aerosol particles, updraft velocity near the cloud base, the amount of seeding material applied, and the change in the physicochemical properties of the seeding aerosols to improve seeding effects were also considered. However, the outline of the results described above remained unchanged. These results demonstrate the possibility of increasing surface precipitation by MP seeding over the catchment. However, seeding a large amount of MP (NaCl) is necessary to enhance precipitation substantially. Simultaneously, considering the environmental impact is essential, as shown in our study.

Disasters caused by heavy rainfall associated with quasi-stationary line-shaped mesoscale convective systems (MCSs) frequently occur in Japan. Thus, highly accurate quantitative precipitation forecast (QPF) information that contributes to decision-making by municipalities to issue evacuation orders is necessary. To this end, we developed a blending forecasting system (BFS) for predicting heavy rainfall associated with MCSs. The BFS blends 1-h observed rainfall and forecasts of extrapolation-based nowcasting (EXT) in the first hour and numerical weather prediction (NWP) in the second hour, predicting 3-h accumulated rainfall (P3h) and its return period (RP) of up to 2 h ahead with a higher horizontal resolution (1 km) and higher-frequency updates (every 10 min) compared to the current operational systems. A blending technique with a spatial maximum filter for tolerating forecast displacement errors (BLEDE) was applied to the predicted rainfall of EXT and NWP. To improve the accuracy of the NWP, vertical profiles of water vapor obtained with two water vapor lidars (WVLs) were assimilated into the NWP. This combination predicted rare heavy rainfall with an RP of more than 10 years in the same city where flooding occurred for a heavy rainfall event associated with quasi-stationary line-shaped MCSs in southern Kyushu on 10 July 2021. The BFS yielded such forecast information 40 min earlier than the existing warning information, indicating the potential for providing a longer lead time for evacuation. The improvement in forecast accuracy was due to both BLEDE and WVL data assimilation (WVL-DA); however, the contribution of BLEDE was more than five times that of WVL-DA in terms of predicting the P3h for the threshold of 80 mm. Additionally, the sensitivity of the predicted rainfall to the background error covariance matrix in WVL-DA is also discussed.