(PhD09) In-Situ Simulation Data Compression for Climate Domain
AI/Machine Learning/Deep Learning
Scientific Software Development
TimeMonday, June 17th1pm - 6pm CEST
DescriptionThe need for storing of climate simulation data has been underlined by exploring of climate change. A climate model is a very complex multi-component system, which is ordinarily represented as multidimensional arrays of numbers with type of ﬂoating-point and contain a speciﬁc climate elements with such atmospheric parameters as temperature, humidity, precipitation, wind speed and power, and other. Usually dimensions of datasets are longitude, latitude, height and time. The amount of the data is growing exponentially, mostly at 40% to 60% year. Compression costs time for decoding and encoding data, but it reduces resource usage, data storage space and transmission capacity (throughput).
This poster represents first results and methodology for the PhD work on data compression for climate domain. It consists of such sections: motivation of the work on data compression for this field, goals, methodology, an overview of the most popular algorithms, and first evaluation results (SCIL as HDF5 filter with other filters). Impact of SCIL usage on overall application performance will be studied.