While we are breaking new science frontiers with Ecology AI, we are actively improving the reliability of survey data and associated scientific outputs of drone-based monitoring surveys. AI can be a useful tool to enhance the reliability of detections and identifications, as well as greatly increase the efficiency of manually post-processing imagery. However, it is vital that the outputs are relevant and answer the needs of the initial intent of the survey.
One of our recent internal research investments has been on developing a statistical method to convert AI detections of various fauna into a reliable estimate of density and population for each species. Current ecological statistical approaches are geared to derive population estimates from human-derived count data. Using these same statistical methods on AI detections do not currently provide robust results.
Through the WildAI partnership with Fujitsu Australia and Indigenous Precision Services, we conducted a long-range survey close to the Victoria/South Australia border, in collaboration with Parks Victoria. The survey area is known for being home to both red and grey kangaroos, which set the ideal field experiment to develop and test our statistical modelling for deriving a population estimate for these two species from the AI detections.
