The field of weather prediction has seen a surge in the application of deep learning techniques, particularly for precipitation nowcasting - predicting the levels and locations of precipitation in the near future. However, most existing deep-learning-based approaches for precipitation nowcasting only consider radar and/or satellite images as inputs, neglecting meteorological observations collected from ground weather stations which are sparsely located. In this paper titled "Deep-Learning-Based Precipitation Nowcasting with Ground Weather Station Data and Radar Data," Jihoon Ko, Kyuhan Lee, Hyunjin Hwang, and Kijung Shin propose a novel attentive method called ASOC that effectively exploits ground-based meteorological observations from multiple weather stations. The proposed method captures temporal dynamics of the observations and contextual relationships between them. It can be easily combined with existing image-based precipitation nowcasting models without changing their architectures. The authors demonstrate that such a combination improves the average critical success index (CSI) of predicting heavy (at least 10 mm/hr) and light (at least 1 mm/hr) rainfall events at 1-6 hr lead times by 5.7%, compared to the original image-based model using radar images and ground-based observations around South Korea collected from 2014 to 2020. This research is set to appear at the 17th International Workshop on Spatial and Spatiotemporal Data Mining (SSTDM-22).
Scientists are using computers to predict the weather better. They use pictures from space and radar, but sometimes they forget to look at information from weather stations on the ground. Some scientists made a new way of looking at this information called ASOC that helps make better predictions. This new way looks at how things change over time and how they are related to each other. They tested it in South Korea and found that it made predictions more accurate without changing the computer program too much. They will talk about their work at a special meeting for people who study maps and data (SSTDM-22).
Definitions
- Weather prediction: guessing what the weather will be like in the future
- Deep learning techniques: ways for computers to learn by themselves
- Precipitation nowcasting: predicting where and when rain or snow will happen soon
- Radar and satellite images: pictures taken from space or with special machines on Earth that show where clouds are
- Meteorological observations: information collected by instruments about temperature, wind, humidity, etc.
- Attentive method: a way of looking carefully at something important
- Temporal dynamics: how things change over time
- Contextual relationships: how things are connected to each other in a certain situation
- Critical success index (CSI): a measure of how well predictions match reality
Deep-Learning-Based Precipitation Nowcasting with Ground Weather Station Data and Radar Data
The field of weather prediction has seen a surge in the application of deep learning techniques, particularly for precipitation nowcasting - predicting the levels and locations of precipitation in the near future. In this paper titled "Deep-Learning-Based Precipitation Nowcasting with Ground Weather Station Data and Radar Data," Jihoon Ko, Kyuhan Lee, Hyunjin Hwang, and Kijung Shin propose a novel attentive method called ASOC that effectively exploits ground-based meteorological observations from multiple weather stations. This research is set to appear at the 17th International Workshop on Spatial and Spatiotemporal Data Mining (SSTDM-22).
Background
Most existing deep learning based approaches for precipitation nowcasting only consider radar images or satellite images as inputs, neglecting meteorological observations collected from ground weather stations which are sparsely located. The authors wanted to develop an approach that could exploit both radar data as well as ground station data to improve accuracy in predicting heavy (at least 10 mm/hr) and light (at least 1 mm/hr) rainfall events at 1-6 hr lead times.
Methodology
The proposed method captures temporal dynamics of the observations and contextual relationships between them. It can be easily combined with existing image-based precipitation nowcasting models without changing their architectures. The authors used South Korea's observational dataset collected from 2014 to 2020 for training their model. They compared their model's performance against an original image based model using radar images along with ground based observation data around South Korea.
Results
The authors found that their proposed method improved the average critical success index (CSI) by 5.7%, compared to the original image based model using radar images and ground based observation data around South Korea collected from 2014 to 2020 when predicting heavy (at least 10 mm/hr) and light (at least 1 mm/hr) rainfall events at 1-6 hr lead times .
Conclusion
This research demonstrates that combining existing image based models with ASOC improves accuracy in predicting heavy (at least 10 mm/hr) and light (at least 1 mm/hr) rainfall events at 1-6 hr lead times by 5.7%. This new approach provides a more accurate way of forecasting rainfalls while also taking into account sparsely located meteorological observations collected from ground weather stations which were previously neglected by most deep learning approaches for precipitation nowcasting
Recently, many deep-learning techniques have been applied to various weather-related prediction tasks, including precipitation nowcasting (i.e., predicting precipitation levels and locations in the near future). Most existing deep-learning-based approaches for precipitation nowcasting, however, consider only radar and/or satellite images as inputs, and meteorological observations collected from ground weather stations, which are sparsely located, are relatively unexplored. In this paper, we propose ASOC, a novel attentive method for effectively exploiting ground-based meteorological observations from multiple weather stations. ASOC is designed to capture temporal dynamics of the observations and also contextual relationships between them. ASOC is easily combined with existing image-based precipitation nowcasting models without changing their architectures. We show that such a combination improves the average critical success index (CSI) of predicting heavy (at least 10 mm/hr) and light (at least 1 mm/hr) rainfall events at 1-6 hr lead times by 5.7%, compared to the original image-based model, using the radar images and ground-based observations around South Korea collected from 2014 to 2020.
