Doppler weather radar is an essential tool for monitoring and warning of hazardous weather phenomena. A large aliasing range ($r_{\mathrm{a}}$) is important for surveillance but a high aliasing velocity ($v_{\mathrm{a}}$) is also important to obtain storm dynamics unambiguously. However, the $r_{\mathrm{a}}$ and $v_{\mathrm{a}}$ are inversely related to pulse repetition time. This “Doppler dilemma” is more challenging at shorter wavelengths. The proposed algorithm employs a CNN (convolutional neural network), which is widely used in image classification, to tackle the velocity dealiasing issue. Velocity aliasing can be converted to a classification problem. The velocity field and aliased count can be regarded as the input image and the label, respectively. Through a fit-and-adjust process, the best weights and the biases of the model are determined to minimize a cost function. The proposed method is compared against the traditional region-based method. Both methods show similar performance on mostly filled precipitation. For sparsely filled precipitation; however, the CNN demonstrated better performance since the CNN processes the entire scan at once while the region-based method processes only the limited adjacent area.

Accurate precipitation forecasting is becoming increasingly important in the context of climate change. In response, machine learning-based approaches have recently gained attention as an emerging alternative to traditional methods such as numerical weather prediction and climate models. Nonetheless, many recent approaches still rely on off-the-shelf loss functions, and even the more advanced ones merely involve optimization processes based on the critical success index (CSI). The problem, however, is that CSI may become ineffective during extended dry periods when precipitation remains below the threshold, rendering it less than ideal as a criterion for optimization. To address this limitation, we introduce a simple penalty expression and reinterpret it as a quadratic unconstrained binary optimization (QUBO) formulation. Ultimately, the resulting QUBO formulation is relaxed into a differentiable advanced torrential (AT) loss function through an approximation process. The proposed AT loss demonstrates its superiority through the Lipschitz constant, forecast performance evaluations, consistency experiments, and ablation studies with the operational model.

Boundary layer (BL) observations are important for forecasting warm-season precipitation and air quality, as well as for more accurately representing BL processes in numerical weather prediction (NWP) and climate models. In the convective BL, organized forms of BL convection, such as horizontal convective rolls (HCRs) and cellular convections, arise depending on environmental conditions. Despite efforts to automate the classification of BL organizational mode, challenges persist and manual classification remains common. With advancements in technology, machine learning offers significant potential to improve classification. In this study, convolutional neural networks (CNNs) are employed to automate the classification of BL organizational mode, specifically cells and HCRs, using S-band radar reflectivity data from the Next Generation Weather Radar (NEXRAD) KTLX (Oklahoma City, Oklahoma) radar system. This classification enables the detection of cells and HCRs, as well as transition periods between cells and HCRs. The CNN-predicted classifications are compared to manual classifications from Stensrud et al. and Banghoff et al., achieving approximately 74% accuracy for cells and 82% for HCRs, with mismatches occurring primarily in mixed-mode cases. The proportion of cell and HCR modes identified by the CNN differed by less than 4% from the manual climatological analysis. Furthermore, the CNN also captured the duration, start time, end time, and transition time of cell and HCR modes, with a mean time difference of under 1 h, showing strong alignment with the manual classification. The CNN-based model could be applied in future research examining large sample sizes of BL modes, significantly reducing the human labor required in manual analysis. Significance Statement: The purpose of this study is to classify radar observed boundary layer organizational modes—namely, horizontal convective rolls (linear patterns) and cellular convections—using convolutional neural networks, a widely used machine learning technique for image classification. Previously, this classification was performed manually, which required significant human labor. The results demonstrate the potential for long-term climatological analysis.