By Dr David Marnevick

South Africa is a biodiversity hotspot, boasting important mammal species globally.

To assist with species conservation, we are continuing with our project to provide a multi-scale inventory of South Africa’s mammals using cutting edge technologies, citizen science, artificial intelligence (AI) and indigenous knowledge.

By using around 1 400 camera traps, we have recorded mammal species presence, behaviour and interactions in their natural habitat. To date, we have classified almost two million images from our focal sites, making this a large and valuable dataset.

While we continue to place cameras, sort images and work with citizen scientists through the Zooniverse platform, the next phase of our project is using AI to help classify our two million images. This step involves using citizen science classifications to train and refine a machine learning model that will accurately tag camera trap images. This accelerates the image classification process so that these we can understand population trends, spatial distributions and behavioural interactions (to name a few) that will help to manage these species, populations and ecosystems.

Critical to this phase is up-skilling South African researchers to implement, run and manage this AI component of the project. Working with our partners at the University of Minnesota’s Lion Center, our South African postdoctoral fellow is currently based in the USA and receiving training on the AI models through the Minnesota Supercomputer Institute (MSI). Simultaneous to this training in the USA, the postdoctoral fellow is also attending a remote training course at the Centre for High Performance Computing (CHPC) in Cape Town in order to replicate the models on the South African supercomputer cluster (known as Lengau). In this way, the project is dedicated to up-skilling South Africans to make this a truly African led project using African based infrastructure and resources.

Excitingly, by the end of 2020 we will start transferring all image processing, storage, machine learning and reporting directly to the CHPC, marking a transition of skills to African soil. This will also create another network hub that will reduce bottlenecks in our data pipeline.

Other key achievements in the project have been refining the citizen science reports to improve image classification accuracy. We are also working with UMN on a large-scale Africa-wide collaboration to refine the machine learning models. We have also successfully barcoded 43 mammal species.

While the COVID-19 pandemic has challenged people across the globe, our team has been working tirelessly to ensure that we continue to understand South Africa’s incredible biodiversity.