Solar image parameter data from the SDO: Long-term curation and data mining

详细信息    查看全文

• 作者：M.A. Schuh^a ; ^{michael.schuh@cs.montana.edu" class="auth_mail" title="E-mail the corresponding author} ; R.A. Angryk^b ; P.C. Martens^c
• 关键词：Solar images ; Computer vision ; Data analysis ; Data mining ; Big data
• 刊名：Astronomy and Computing
• 出版年：2015
• 出版时间：November 2015
• 年：2015
• 卷：13
• 期：Complete
• 页码：86-98
• 全文大小：6446 K

文摘

The Solar Dynamics Observatory (SDO) mission captures thousands of images of the Sun per day, motivating the need for efficient and effective storage, representation, and search over a massive repository of data. This work investigates the general-purpose image parameter data produced by the SDO Feature Finding Team’s trainable module, which operates at a fixed six minute cadence over all AIA channels. The data contains ten numerical measures computed for each image cell over a class="mathmlsrc">class="formulatext stixSupport mathImg" data-mathURL="/science?_ob=MathURL&_method=retrieve&_eid=1-s2.0-S2213133715000979&_mathId=si22.gif&_user=111111111&_pii=S2213133715000979&_rdoc=1&_issn=22131337&md5=d1915c34d6a2a8dd8167155db475b82a" title="Click to view the MathML source">64×64class="mathContainer hidden">class="mathCode">$64\times 64$ grid for each image. We analyze all available data and metadata produced over the first three years and present comprehensive statistics and outliers while validating the cleanliness and usability of the data source for future research. We then utilize a database of automated solar event reports to create large-scale region-labeled datasets available to the public. We highlight the new-found potential for data-driven discovery by presenting several best-case labeling scenarios that establish a baseline for comparing machine learning classification and attribute (image parameter) evaluation results. Future work focuses on continued dataset curation and spatiotemporal data mining.