Pilot release for eTEC-BIG pre-proposal

This release demonstrates the feasibility of ecological niche modelling using deep learning (ENM-DL). In comparison with the proposed research the following is lacking and beyond our present capabilities:

• The data set is for the Ungulates. This is a group of ±150 species. We aim for a data set that is 100 times larger, which will require scaling up in the data filtering workflow.
• The ENM-DL components are prototyped as iPython Notebooks whose memory requirements are already getting out of hand. This needs to be replaced with parallelized python library code to be run on SurfSARA server infrastructure.
• We need outside expertise on how best to assess how well the models fit the data. E.g. can AUC values be compared across species? How would we filter out bad models across 10,000 species?
• Extracting 'niche trait' values out of GIS layers is turning out to be an expensive operation, presumably because of inefficient implementation of the TIFF data handling. We need outside expertise on how to optimize that.