New particle formation event detection with Mask R-CNN

Abstract. Atmospheric new particle formation (NPF) is an important source of climate-relevant aerosol particles which has been observed at many locations globally. To study this phenomenon, the first step to identify whether NPF event occurs or not on a given day. In practice, identification performed visually by classifying non-event days from number size distribution surface plots. Unfortunately, day-by-day visual classification time-consuming and labor-intensive, process renders subjective results. detect events automatically, we regard signature (banana shape) all over world in plots as special kind object, deep learning model called Mask R-CNN applied localize spatial layouts their Utilizing only 358 human-annotated masks data Station for Measuring Ecosystem–Atmosphere Relations (SMEAR) II station (Hyytiälä, Finland), was successfully generalized three SMEAR stations Finland San Pietro Capofiume (SPC) Italy. addition detection (especially strongest events), presented method can determine growth rates, start times, end times automatically. The automatically determined rates agree with manually rates. statistical results validate potential applying proposed different sites, will improve automatic level analysis. Furthermore, analysis minimize subjectivity compared human-made analysis, especially when long-term series are analyzed comparisons between sites needed characteristics such thereby saving time effort scientists studying events.