detection of probabilistic Convective Initiation (CI) using Himawari-8 AHI, weather radar, and lightning data.

Convective rainfall can cause flash flooding with significant human and economic losses. In order to prevent such damages, monitoring and prediction of convective rainfall have been conducted with Automatic Weather System (AWS) and ground based weather radar data. However, these measurements cannot cover vast areas limiting spatial continuity. Geostationary satellite sensors observe clouds and storms over vast areas at very high temporal resolution ( 10 minutes). Thus, geostationary satellite remote sensing is an alternative way to predict and monitor convective rainfall. In general, interest fields such as brightness temperature at a specific spectral channel or the difference of brightness temperatures between two channels are considered important to identify Convective Initiation (CI). Existing CI algorithms use simple interest fields and their associated thresholds. However, such a simple thresholding approach might not be ideal to consider complicated characteristics of convective clouds. In this study, logistic regression and probabilistic random forest were evaluated to provide CI probability associated with various characteristics of convective clouds. Himawari-8 Advanced Himawari-8 Imager (AHI) data collected between June and August 2015 were used to detect CI. A quantitative validation of CI was conducted using weather radar and lightning data. Results show that an overall accuracy of CI detection by logistic regression is 84.5% when radar data was used as reference data and 0.5 was applied to the probability data as a threshold to make a binary classification, which is higher than that by probabilistic random forest (87.4%). The validation using lightning data produced a similar result with the radar-based assessment. However, the probability of detection (POD) of the logistic regression model was a bit lower than that of the random forest model due to the relatively large number of missed CI objects.