2020 ESA Annual Meeting (August 3 - 6)

OOS 37 Abstract - Machine learning for decision support in wildlife conservation and land management

Wednesday, August 5, 2020: 3:45 PM
Dan Morris1, Siyu Yang1, Caleb Robinson1 and Nebojsa Jojic2, (1)Microsoft AI for Earth, (2)Microsoft Research
Background/Question/Methods

Conservation planning is a complex process that requires stakeholders to balance sparse evidence, financial constraints, political issues, and social factors. This type of multivariate integration is uniquely suited to human intelligence, and high-level planning remains extremely challenging for automated systems, even when those systems are playing a supporting role. However, ecologists and conservation planners often spend inordinate amounts of time distilling actionable insights from raw data, hours that could instead be put into these higher-level tasks. Therefore, we argue that one of the most substantial contributions that AI can make to conservation planning is accelerating the transformation from raw to actionable data, freeing valuable time for integrative tasks. In this talk, we will discuss ongoing work in using AI to accelerate (1) wildlife population surveys and (2) land cover surveys.

Results/Conclusions

Specifically, we will look at progress and challenges – algorithmic, infrastructural, and human factors – associated with accelerating the processing of sensor-based biodiversity surveys, through which we can accelerate (though not yet automate) the integration of data from camera traps, aerial images, acoustic devices, and other sensors. Then we will look at parallel challenges in the field of land cover mapping from aerial and satellite imagery. Both of these areas represent opportunities for AI and cloud-based processing tools to dramatically accelerate conservation workflows, allowing domain experts to maximize the time they spend on those tasks that are uniquely suited to human intelligence.