Research on nowcasting through dual-polarization weather radar data using deep learning approach is rare but worth exploring. This paper lightens a previous work, the MCT (Multivariate Channel Transformer) model, which leads to the design of the MSF (Multivariate Swin Fusion) model. The commonalities between the two are as follows: on one hand, both fuses several dual-polarization observables including reflectivity (Z), specific differential phase (Kdp ), and differential reflectivity (Zdr ) to more comprehensively consider meteorological particle features; on the other hand, they introduces the attention mechanism to more fully fuse multi-frame, multi-variate, and multi-scale features. In the experimental evaluation, this study first selects observation data from KMLB radar in FL, USA, and uses traditional optical flow method, deep learning TrajGRU method, etc. as controls. The results show that both MCT and MSF perform better than the control, and the 60min forecast scores of both are 8.78/9.31 for RMSE and 0.46/0.18/0.07 for CSI (20/35/45dBZ), and this conclusion is verified by case study. Further, the role of the attention mechanism is verified by ablation experiments.
more »
« less
Physically Explainable Deep Learning for Convective Initiation Nowcasting Using GOES-16 Satellite Observations
Convective initiation (CI) nowcasting remains a challenging problem for both numerical weather prediction models and existing nowcasting algorithms. In this study, an object-based probabilistic deep learning model is developed to predict CI based on multichannel infrared
- Award ID(s):
- 2019758
- PAR ID:
- 10534407
- Publisher / Repository:
- American Meteorological Society
- Date Published:
- Journal Name:
- Artificial Intelligence for the Earth Systems
- Volume:
- 3
- Issue:
- 3
- ISSN:
- 2769-7525
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
more » « less
Abstract The evolution of model-based cloud-top brightness temperatures (BT) associated with convective initiation (CI) is assessed for three bulk cloud microphysics schemes in the Weather Research and Forecasting Model. Using a composite-based analysis, cloud objects derived from high-resolution (500 m) model simulations are compared to 5-min
GOES-16 imagery for a case study day located near the Alabama–Mississippi border. Observed and simulated cloud characteristics for clouds reaching CI are examined by utilizing infrared BTs commonly used in satellite-based CI nowcasting methods. The results demonstrate the ability of object-based verification methods with satellite observations to evaluate the evolution of model cloud characteristics, and the BT comparison provides insight into a known issue of model simulations producing too many convective cells reaching CI. The timing of CI from the different microphysical schemes is dependent on the production of ice in the upper levels of the cloud, which typically occurs near the time of maximum cloud growth. In particular, large differences in precipitation formation drive differences in the amount of cloud water able to reach upper layers of the cloud, which impacts cloud-top glaciation. Larger cloud mixing ratios are found in clouds with sustained growth leading to more cloud water lofted to the upper levels of the cloud and the formation of ice. Clouds unable to sustain growth lack the necessary cloud water needed to form ice and grow into cumulonimbus. Clouds with slower growth rates display similar BT trends as clouds exhibiting growth, which suggests that forecasting CI using geostationary satellites might require additional information beyond those derived at cloud top. -
The task of nowcasting by deep learning using multivariate, rather than just reflectivity, is limited by poor interpretability. The previous experiment designed MCT (Multivariate Channel Transformer), a deep learning model capable of nowcasting with dual-polarization radar data. Four analytical methods are designed to further explore the contribution of polarization parameters: (i) Case studies of different meteorological processes. (ii) A permutation test ranking the significance of each variable. (iii) Visualization of the feature maps obtained by forward propagation of the input data. (iv) Data downscaling of polarimetric radar data. The results show that the polarization parameters serve as a guide to predict the location and shape of strong reflectivity, as well as the energy retention of strong echoes at 40-50 dBZ. The contributions of Zdr and Kdp are more evident in the prediction results after 30 min, and the importance of Kdp exceeds that of Zdr in case of strong convective weather.more » « less
-
more » « less
Abstract Accurate short‐term weather prediction, essential for many aspects of life, relies mainly on forecasts from numerical weather models. Here, we report results supporting strongly deep learning as a viable, alternative approach. A 3D convolutional neural network, which uses a single frame of meteorology fields as input to predict the precipitation spatial distribution, is developed based on 39‐years (1980–2018) data of meteorology and daily precipitation over the contiguous United States. Results show that the trained network outperforms the state‐of‐the‐art weather models in predicting daily total precipitation, and the superiority of the network extends to forecast leads up to 5 days. Combining the network predictions with the weather‐model forecasts significantly improves the accuracy of model forecasts, especially for heavy‐precipitation events. Furthermore, the millisecond‐scale inference time of the network facilitates large ensemble predictions for extra accuracy improvement. These results demonstrate the promising prospects of deep learning in short‐term weather predictions.
-
more » « less
Abstract Numerical weather prediction models often fail to correctly forecast convection initiation (CI) at night. To improve our understanding of such events, researchers collected a unique dataset of thermodynamic and kinematic remote sensing profilers as part of the Plains Elevated Convection at Night (PECAN) experiment. This study evaluates the impacts made to a nocturnal CI forecast on 26 June 2015 by assimilating a network of atmospheric emitted radiance interferometers (AERIs), Doppler lidars, radio wind profilers, high-frequency rawinsondes, and mobile surface observations using an advanced, ensemble-based data assimilation system. Relative to operational forecasts, assimilating the PECAN dataset improves the timing, location, and orientation of the CI event. Specifically, radio wind profilers and rawinsondes are shown to be the most impactful instrument by enhancing the moisture advection into the region of CI in the forecast. Assimilating thermodynamic profiles collected by the AERIs increases midlevel moisture and improves the ensemble probability of CI in the forecast. The impacts of assimilating the radio wind profilers, AERI retrievals, and rawinsondes remain large throughout forecasting the growth of the CI event into a mesoscale convective system. Assimilating Doppler lidar and surface data only slightly improves the CI forecast by enhancing the convergence along an outflow boundary that partially forces the nocturnal CI event. Our findings suggest that a mesoscale network of profiling and surface instruments has the potential to greatly improve short-term forecasts of nocturnal convection.
