5:15 PM - 7:15 PM
[AAS05-P09] Enhancement of Precipitation Mapping Accuracy in the Korean Peninsula using AI and Multi-source Dataset
Keywords:Precipitation, ASOS, GPM, Radar, LDAPS, Random Forest
The Korean Peninsula is located at the eastern tip of the Asian continent, and although it is relatively small at approximately 220,000 km², it is more than 70% mountainous and surrounded by the sea on three sides, making it a humid climate region. This geography results in pronounced seasonal and interannual variability in precipitation, which makes it vulnerable to extreme climate events such as droughts and floods. In recent years, climate change has intensified the irregularity of precipitation patterns, and there is a need for accurate and high-resolution precipitation maps that reflect the spatial and temporal variability of precipitation.
In this study, various datasets were integrated using a Random Forest (RF) machine learning algorithm to generate a high-resolution precipitation map that considers spatial and temporal variability. The main dataset used is ASOS data, which is a reliable ground observation from 105 meteorological stations, but it is collected as point measurements and has limitations in fully representing the distribution of precipitation. To address this, high-resolution radar precipitation data, Global Precipitation Measurement (GPM) Integrated Multi-Satellite Receiving (IMERG) data, Local Data Assimilation Prediction System (LDAPS) data, and Shuttle Radar Altimetry (SRTM) Digital Elevation Model (DEM) data were added to reflect high-resolution spatial and temporal variability and effectively represent the complex terrain of the Korean Peninsula.
The high-resolution precipitation maps generated in this study can improve the accuracy of precipitation distribution analysis and compensate for the limitations of conventional point estimation methods to more accurately reflect precipitation variability across the Korean Peninsula. It is also expected to be utilized in various fields such as disaster management, weather forecasting, and climate change adaptation strategies. In particular, since precipitation is an important natural mechanism that effectively removes fine particulate matter from the atmosphere, the precipitation maps in this study, which accurately reflect the spatial and temporal variability of precipitation, will contribute to improving the reliability of air quality forecasts.
In this study, various datasets were integrated using a Random Forest (RF) machine learning algorithm to generate a high-resolution precipitation map that considers spatial and temporal variability. The main dataset used is ASOS data, which is a reliable ground observation from 105 meteorological stations, but it is collected as point measurements and has limitations in fully representing the distribution of precipitation. To address this, high-resolution radar precipitation data, Global Precipitation Measurement (GPM) Integrated Multi-Satellite Receiving (IMERG) data, Local Data Assimilation Prediction System (LDAPS) data, and Shuttle Radar Altimetry (SRTM) Digital Elevation Model (DEM) data were added to reflect high-resolution spatial and temporal variability and effectively represent the complex terrain of the Korean Peninsula.
The high-resolution precipitation maps generated in this study can improve the accuracy of precipitation distribution analysis and compensate for the limitations of conventional point estimation methods to more accurately reflect precipitation variability across the Korean Peninsula. It is also expected to be utilized in various fields such as disaster management, weather forecasting, and climate change adaptation strategies. In particular, since precipitation is an important natural mechanism that effectively removes fine particulate matter from the atmosphere, the precipitation maps in this study, which accurately reflect the spatial and temporal variability of precipitation, will contribute to improving the reliability of air quality forecasts.