A month work for one or two software engineers.

Two major tasks: A) A library to generate generate graph compressed data structure (compressed sparse adjacency, edge matrix or list ): 1- to read in parallel (distributed parallelism —mpi or netCDF-4 parallel) raw data of simulation models which is available in netCDF. The size of data is of 10s to 100's of GBs. 2- To generate time series from read model data, which is in 3-dimensions: variable, time and location (gridded data). So from data you will have big number of time series. The has to be chunked in the different processors in a distributed memory architecture. 3- Two parallel functions: a) The calculation of the Pearson Correlation Coefficient between all time series to generate an edge or adjacency list out of it. Here, a user specified parameter will define a threshold between 0 and 1,and considering it, the function will build a corresponding edge list (and it should be compressed since the resulting correlation matrix after applying the threshold will be sparse, i.e lots of values less than the user specified threshold). The second is a function doing the same but by calculating the mutual information rather than pearson correlations. B) Suggesting a compatible parallel graph analysis package. If not, paralleling igraph package using openMP for multicore systems (and also for GPGPUS). This package has to read the produced data structure from (A), and will construct a graph from it, and will analyze (in parallel) the graph (weighted, directed, unweighted, etc), for example: degree, betweenness, community detection, etc.