Search

Machine learning algorithms should be tested for use in quantitative precipitation estimation models of rain radar data in South Korea because such an application can provide a more accurate estimate of rainfall than the conventional ZR relationship-based model. The applicability of random forest, stochastic gradient boosted model, and extreme learning machine methods to quantitative precipitation estimation models was investigated using case studies with polarization radar data from Gwangdeoksan radar station. Various combinations of input variable sets were tested, and results showed that machine learning algorithms can be applied to build the quantitative precipitation estimation model of the polarization radar data in South Korea. The machine learning-based quantitative precipitation estimation models led to better performances than ZR relationship-based models, particularly for heavy rainfall events. The extreme learning machine is considered the best of the algorithms used based on evaluation criteria. [ABSTRACT FROM AUTHOR]

Hallasan Mountain is located at the center of Jeju Island, Korea. Even though Hallasan Mountain has a height of just 1,950 m, the temperature during the winter decreases below −20 degrees Celsius. On the contrary, the temperature on the coastal areas remains just above freezing. Therefore, large snowfalls in the mountain and rainfall in the coastal areas are very common in Jeju Island. Most of the rain gauges are available around highly populated coastal areas, and snow measurements are available at just four locations on the coastal areas. Therefore, it is practically impossible to distinguish the rainfall and snowfall in Jeju Island. Fortunately, two radars (Seongsan and Gosan radars) operate on Jeju Island, which fully covers Hallasan Mountain. This study proposes a method of using both the radar and rain gauge information to map the snowy region in Jeju Island, including Hallasan Mountain. As a first step, this study analyzed the Z-R and Z-S relationships to derive a fixed threshold of radar reflectivity to separate snowfall from rainfall, and, in the second step, this study additionally considered the observed rain rate information to implement the problem of using the fixed threshold. This proposed method was applied to radar reflectivity data collected during November 1, 2014, to April 30, 2015, and the results indicate that the method considering both the radar and rain gauge information was satisfactory. This method also showed good performance, especially when the rain rate was very low. [ABSTRACT FROM AUTHOR]

The microphysical characteristics of a mesoscale convective system (MCS) during a summer monsoon of South Korea are investigated using the generalized drop size distributions (DSD) that are derived from S-band dual-polarization radar data. The characteristics parameters of generalized DSDs (generalized number concentration, N0′ and generalized mean diameter, Dm) are directly calculated from DSD's two moments without any assumption on the DSD model. Relationships between ZDR and generalized DSD parameters normalized by ZH are derived in the form of the polynomial equation. Verification of the retrieved DSD parameters is conducted with the 2-D video disdrometer (2DVD) located about 23 km from the radar. The standard deviations (SD) of retrieved DSD parameters are about 0.26 for log N0′, and about 0.11 for Dm because of the variability of DSDs. The SD of the retrieved log N0′ from the dual-polarimetric measurement reaches to about 0.46 (almost double) for 11 rain events while the accuracy of retrieved Dm is quite higher (~0.19). This higher error in retrieved log N0′ is likely attributed to the larger discrepancy in radar-observed and DSD-calculated ZDR when ZH is low. This retrieval technique is applied to a mesoscale convective system (MCS) case to investigate the Lagrangian characteristics of the microphysical process. The MCS is classified into the leading edge and trailing stratiform region by using the storm classification algorithm. The leading edge dominated by strong updraft showed the broad DSD spectra with a steady temporal increase of Dm throughout the event, likely because of the dominant drop growth by the collision-coalescence process. On the other hand, the drop growth is less significant in the trailing stratiform region as shown by the nearly constant Dm for the entire period. The DSD variation is also controlled by the new generation of drops in the leading edge and less extent in the trailing stratiform during the early period when precipitation systems grow. When the system weakens, the characteristic number concentration decreases with time, indicating the new generation of drops becomes less significant in both regions. [ABSTRACT FROM AUTHOR]