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
MSF: One Lightweight Deep Learning Nowcasting Method with Attention Mechanism using Dual-Polarization Radar Observations
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
- Award ID(s):
- 2239880
- NSF-PAR ID:
- 10515827
- Publisher / Repository:
- IEEE
- Date Published:
- ISBN:
- 979-8-3503-2010-7
- Page Range / eLocation ID:
- 3807 to 3810
- Format(s):
- Medium: X
- Location:
- Pasadena, CA, USA
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Real-time and accurate precipitation estimation is critical for environmental protection and water resources management. Compared to traditional methods, i.e., radar reflectivity (Z) and rainfall rate (R) relations, relying on local raindrop size distributions, the deep learning model can fit the functional relationship between radar observations and rainfall rate measurements. However, the black-box nature of deep learning models makes it difficult to explain the physical mechanisms behind their results. To address this problem, this study proposes DQPENet, a deep learning model for polarimetric radar QPE utilizing dense blocks. We employ a permutation test to understand the relative importance of different radar data input variables. Additionally, we propose a regression importance value (RIV) method for the precipitation estimation task to visualize feature importance regions. Our experimental results show that radar reflectivity and specific differential phase at the lowest elevation angle are the two most important observables for the model’s precipitation estimation. Furthermore, we find that radar data closer to the rain gauge are more influential on the model’s results, indicating that the deep learning model is able to capture the underlying physical mechanism of atmospheric data.more » « less
-
Abstract Dual-frequency millimeter-wavelength radar observations in snowfall are analyzed in order to evaluate differences in conventional polarimetric radar variables such as differential reflectivity ( Z DR ) specific differential phase shift ( K DP ) and linear depolarization ratio (LDR) at traditional cloud radar frequencies at Ka and W bands (~35 and ~94 GHz, correspondingly). Low radar beam elevation (~5°) measurements were performed at Oliktok Point, Alaska, with a scanning fully polarimetric radar operating in the horizontal–vertical polarization basis. This radar has the same gate spacing and very close beam widths at both frequencies, which largely alleviates uncertainties associated with spatial and temporal data matching. It is shown that observed Ka- and W-band Z DR differences are, on average, less than about 0.5 dB and do not have a pronounced trend as a function of snowfall reflectivity. The observed Z DR differences agree well with modeling results obtained using integration over nonspherical ice particle size distributions. For higher signal-to-noise ratios, K DP data derived from differential phase measurements are approximately scaled as reciprocals of corresponding radar frequencies indicating that the influence of non-Rayleigh scattering effects on this variable is rather limited. This result is also in satisfactory agreement with data obtained by modeling using realistic particle size distributions. Observed Ka- and W-band LDR differences are strongly affected by the radar hardware system polarization “leak” and are generally less than 4 dB. Smaller differences are observed for higher depolarizations, where the polarization “leak” is less pronounced. Realistic assumptions about particle canting and the system polarization isolation lead to modeling results that satisfactorily agree with observational dual-frequency LDR data.more » « less
-
more » « less
Abstract. Radar observations of winter storms often exhibit locally enhanced linear features in reflectivity, sometimes labeled as snow bands. We have developed a new, objective method for detecting locally enhanced echo features in radar data from winter storms. In comparison to convective cells in warm season precipitation, these features are usually less distinct from the background echo and often have more fuzzy or feathered edges. This technique identifies both prominent, strong features and more subtle, faint features. A key difference from previous radar reflectivity feature detection algorithms is the combined use of two adaptive differential thresholds, one that decreases with increasing background values and one that increases with increasing background values. The algorithm detects features within a snow rate field rather than reflectivity and incorporates an underestimate and overestimate of feature areas to account for uncertainties in the detection. We demonstrate the technique on several examples from the US National Weather Service operational radar network. The feature detection algorithm is highly customizable and can be tuned for a variety of data sets and applications.
-
Image muting of mixed precipitation to improve identification of regions of heavy snow in radar datamore » « less
Abstract. In winter storms, enhanced radar reflectivity is often associated with heavy snow. However, some higher reflectivities are the result of mixed precipitation including melting snow. The correlation coefficient (a dual-polarization radar variable) can identify regions of mixed precipitation, but this information is usually presented separately from reflectivity. Especially under time pressure, radar data users can mistake regions of mixed precipitation for heavy snow because of the high cognitive load associated with comparing data in two fields while simultaneously attempting to discount a portion of the high reflectivity values. We developed an image muting method for regional radar maps that visually de-emphasizes the high reflectivity values associated with mixed precipitation. These image muted depictions of winter storm precipitation structures are useful for analyzing regions of heavy snow and monitoring real-time weather conditions.
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1109/IGARSS52108.2023.10283277
