In this study, a convection nowcasting method based on machine learning was proposed. First, the historical data were back-calculated using the pyramid optical flow method. Next, the generated optical flow field information of each pixel and the Red-Green-Blue (RGB) image information were input into the Convolutional Long Short-Term Memory (ConvLSTM) algorithm for training purposes. During the extrapolation process, dynamic characteristics such as the rotation, convergence, and divergence in the optical flow field were also used as predictors to form an optimal nowcasting model. The test analysis demonstrated that the algorithm combined the image feature extraction ability of the convolutional neural network (CNN) and the sequential learning ability of the long short-term memory network (LSTM) model to establish an end-to-end deep learning network, which could deeply extract high-order features of radar echoes such as structural texture, spatial correlation, and temporal evolution compared with the traditional algorithm. Based on learning through the above features, this algorithm can forecast the generation and dissipation trends of convective cells to some extent. The addition of the optical flow information can more accurately simulate nonlinear trends such as the rotation, or merging, or separation of radar echoes. The trajectories of radar echoes obtained through nowcasting are closer to their actual movements, which prolongs the valid forecasting period and improves forecast accuracy.

No abstract

By using various skill scores and spatial characteristics of spatial verification methods and traditional techniques of the model evaluation tool, the gridded precipitation observation, known as Climate Prediction Center Morphing Technique, gauge observation and three datasets that were derived from local, Shanghai, and Grapes models, respectively, were conducted to assess the 3 lead day rainfall forecast with 0.5 day intervals during the summer of 2020 over Central East China. Results have shown that the local model generally outperforms the other two for the most skill scores but usually with relatively larger uncertainties than the Shanghai model, and it has the least displacement errors for moderate rainfall among the three datasets. However, the rainfall of the Grapes model has been heavily underestimated and is accompanied with a large displacement error. Both the local and Shanghai model can effectively forecast the large-scale convection and rainstorms but over forecast the local convection, while the local model likely over forecasts the local rainstorms. In addition, the Shanghai model slightly favors over forecasting on a broad scale range and a broad threshold range, and the local model slightly misses the rainfall exceeding 100 mm. Generally, for a broadly comparative evaluation on rainfall, the popular dichotomous methods should be recommended when considering reasonable classification of thresholds if the accuracy is highly demanding. In addition, most spatial methods are suggested to conduct with proper pre-handling of non-rainfall event cases. Especially, the verification metrics including spatial characteristic difference information should be recommended to emphasize rewarding the severe events forecast under a global warming background.

By using various skill scores and spatial characteristics of spatial verification methods and tradi-tional techniques of the model evaluation tool (MET, V10.0.0), the gridded precipitation obser-vation CMPAV (V2.0) and three datasets that derived from local (LOC), Shanghai (SHA), and Grapes (GRA) model respectively are conducted to assess the 3 lead day rainfall forecast with 0.5-day intervals during summer of 2020 over central east China. Results have shown that LOC generally outperforms the other two for most skill scores but usually with relatively larger un-certainties than SHA, and it has the least displacement errors for moderate rainfall among the three datasets. However, the rainfall of GRA has been heavily underestimated and accompanied with large displacement error. Both LOC and SHA have shown almost equitable abilities in forecasting convection and rainstorms of the large area but with a slightly over-forecast of local convection, while LOC likely over-forecasts the local rainstorms. In addition, SHA slightly favors over-forecast on a broad scale range and a broad threshold range, and LOC slightly misses the rainfall exceeding 100 mm. Generally, for a broadly comparative evaluation on rainfall, the popular dichotomous methods should be recommended under considering reasonable classifi-cation of thresholds if the accuracy is highly demanded. And most spatial methods should be suggested to conduct with proper pre-handling of non-rainfall event cases. Especially, the veri-fications including spatial characteristic difference information could be recommended in a computationally sufficient environment.

Cloud modeling is one of the important and effective means in investigating cloud processes with sophisticated representations of cloud microphysics, and it can reasonably well resolve the time evolution, structure and life cycles of a single cloud and its systems. This paper presents a brief discussion and review of cloud models of three kinds, i.e., models with bulk parameterization, spectrum bin scheme and for cloud resolving, and some of their major applications are introduced including the influence of precipitation on cumulous dynamics, condensation growth and spectrum broadening, interaction between aerosol particles and clouds, together with cloud chemistry. Keywords-bulk parameterization scheme; spectrum bin model; cloud resolving model; aeroso; cloud chemistry I.978-0-7695-3563-0/08 $25.00