Numerical weather prediction (NWP) models are fundamental in meteorology for simulating and forecasting the behavior of various atmospheric variables. The accuracy of precipitation forecasts and the acquisition of sufficient lead time are crucial for preventing hazardous weather events. However, the performance of NWP models is limited by the nonlinear and unpredictable patterns of extreme weather phenomena driven by temporal dynamics. In this regard, we propose a Self-Supervised Learning with Probabilistic Density Labeling (SSLPDL) for estimating rainfall probability by post-processing NWP forecasts. Our post-processing method uses self-supervised learning (SSL) with masked modeling for reconstructing atmospheric physics variables, enabling the model to learn the dependency between variables. The pre-trained encoder is then utilized in transfer learning to a precipitation segmentation task. Furthermore, we introduce a straightforward labeling approach based on probability density to address the class imbalance in extreme weather phenomena like heavy rain events. Experimental results show that SSLPDL surpasses other precipitation forecasting models in regional precipitation post-processing and demonstrates competitive performance in extending forecast lead times. Our code is available at https://github.com/joonha425/SSLPDL.

Accurate short-term precipitation forecast is of social and economic significance for preventing severe weather damage. Deep learning has been rapidly adopted in nowcasting based on weather radar, which plays a key role in preventing dangerous weather conditions such as torrential rainfall. However, the limited observation range of the radar imposes constraints on shorter forecast lead times. Securing a sufficient lead time for timely flood warnings and emergency responses is crucial. Here, we propose a novel GAN-based framework that combines radar and satellite data to extend forecast lead time. First, we tokenize the satellite image to align with radar dimensions and combine the satellite and radar data. We then apply positional encoding to add positional information. Second, we design the self-conditioned generator to estimate distributions of various rainfall intensities. Finally, we employ Gaussian Fourier features to map the input noise into a continuous representation. The proposed framework realistically and accurately produces time series images of various precipitation types. Furthermore, our multisource data-driven system outperforms numerical weather prediction at forecasts of up to 6 hours in South Korea.

Summary This study aims to recommend metadata for building a high ranking in Search Engine Result Page (SERP) by considering Search Engine Optimizations (SEO). For online marketing, it is important to place their websites on the top rank in a result of search engines. However, on‐page techniques of traditional SEO do not have logical foundation to select metadata. Metadata is an important element to prioritize of websites when search engine indexing for user queries. Thereby, for online marketing, this study proposes a method for recommending metadata, which consists of two steps: i) combining keywords and metadata from high‐ranked websites, and ii) evaluating the importance of terms based on semantic relevance. First, terms are selected with influential keywords and metadata by using their frequency and weight. Second, prioritize the terms according to semantic relevance based on a competitive learning model. We evaluated the validity of the proposed method by using three queries in Google. Experimental results demonstrate that it increases traffic of a website, by using terms, which are high‐ranked websites and semantic relevance.