Machine Learning Capabilities using the NASA NCCS High Performance Data Analytics Services

The NASA Center for Climate Simulation (NCCS) is building a shared data analytic platform. The platform is co-located with both high performance computational capability and large volumes of NASA generated datasets from both observational platforms and Earth system simulations. A primary objective of the platform is to assist researchers performing in situ analytics and deploy ideas quickly into applications.

To prove the concept, we will present examples and architectures of several machine learning cases built on the platform, including: Neural Network workflow to predict Indian Monsoon precipitation with the reanalysis datasets and to explore mechanisms that constrain overfitting and evaluate computational improvement in GPU system. Maximum Entropy workflow to investigate the correlation between climate variability and bird habitat distribution, exploring the feasibility of Maximum Entropy for feature selection. Multiple regression workflows, including random forest regression, support vector regression, and neural network. An example will also be to derive the depth of lakes from remote sensing data. We will conduct inter-comparison among the different algorithms.

Through these exercises, we aim to accumulate experiences on constructing complex workflows and provide preferable data analytics services for the users.