Machine learning is already widely established in the sciences and for global applications since many decades. It offers tremendous opportunities and new information, and can bring much progress to the sustainable management of natural resources and human well-being worldwide. But so far, it has been widely underused for science-based wildlife management questions, and which traditionally just ask very narrow and parsimonious questions; often applied just after the fact. Here I review the traditional wildlife analysis design, and how it compares to, and can be extended and updated with, machine learning methods.
At the Salford Analytics and Data Mining Conference in May 2012 Dr. Falk Huettmann presented spatial land- and seascape examples worldwide (e.g. Arctic, Antarctic, Hindu-Kush Himalayas, Oceans) using public data and Geographic Information Systems (GIS) for high profile peer-reviewed conservation management studies that involve large open access data sets (>20), data mining and non-parsimonious multi-species predictions for a pro-active management before management problems occur. He covered how machine learning - by itself or in parallel and with model ensembles - can improve information gain, predictions, speed, model selection, inference, hypothesis testing, convenience, and data access all in once. Climate predictions and policy implications are emphasized. Dr. Huettmann concluded with an outlook that shows how to implement machine learning in the modern biodiversity and wildlife management sciences, online, open access, with metadata, for adaptive management, and where the future of biodiversity and wildlife management related data analysis and its ‘best practice’ will soon be.
Here is a short clip: