[A-CG37] 衛星による地球環境観測

[ACG37-P23] Development of Fugitive Dust Early Wanning Model

*Chih Chao Ho¹、Tsu-Chiang Lei^2,1、Yi-Shiang Shiu²、Chih-Hsiung Chang¹、Lan-Chieh Pi³、Chia- Chuan Hsu³ (1.Construction and Disaster Prevention Research Center,Feng Chia University、2.Department of Urban Planning and Spatial Information,Feng Chia University、3.Water Resources Planning Institute, Water Resources Agency, Ministry of Economic Affairs)

キーワード：Fugitive Dust, LightGBM, Telemetry

The bare land of riverbed in the Taiwan Basin increases gratly during drought season erver year. Accompany the prevailing northeast monsoon, tiny sand and dust are easily lifted by strong winds, which seriously affects the health and life quality of nearby residents. In order to effectively predict and suppress fugitive dust, the study use hydrometeorological, air quality monitoring data, and river enviromental telemetry information to devolope a fugitive dust early warnning model by LightGBM. The model provides the probability of PM10 exceeding the standard in the next 48 hours.Among model, the river enviromental telemetry information, the proportion of bare land area of different river reach, was produced by SCL product of sentinel-2 satellite. The Zhuoshui River Basin in central Taiwan is used as case study in this study. The performance of classification is represented by the confusion matrix, and it is used to estimate the seven indicators of accuracy, precision, sensitivity(or recall), specificity, F1-score, balanced accuracy, and Matthews correlation coefficient. The results shows most of indicators for PM10 exceeding the standard in next 48 hours have excellent performance. Only the specificity in Xuguang, Yixian, Erlin, Mailiao, and Lunbei is slightly lower, which means that the PM10 early warning event has a slightly higher probability of being misjudged as exceeding the standard.