More effective Wildlife Conservation with AI & cloud

The Need

Since its introduction with the independence of Namibia in 1990, the concept of community-based conservation had improved both the conservation of endangered species and the livelihoods of the rural communities. Receiving direct economic gains from tourism on their land, local communities in Namibia, legally established in so-called 'Communal Conservancies', felt the positive impact on their life and thus bought in to the cause of wildlife conservation. Recently, however, the drop in tourism due to Covid-19 and progressing climate change put those achievements at risk.

Our client, a global NGO, introduced a new program to directly compensate conservancies for additional conservation efforts, which were to be validated by data. The conservancies contractually agreed on setting aside specific parcels of conservancy land for exclusive use by wildlife and committed to regular ranger patrols into these areas. In a pilot project, eight conservancies in the Zambezi region deployed camera traps to estimate species abundance and logged ranger patrols using a smartphone app. Further, satellite imagery was periodically inspected to identify prohibited human interference in specifically demarcated land corridors crucial to migratory routes.

Several existing stand-alone tools were used by the client for the processing of patrol data and camera trap images. However, they operated in isolation, requiring additional manual effort to consolidate and check for consistency. To facilitate rapid and cost-effective scalability, it was crucial that the technological solution for processing the data be low-maintenance, highly automated and cost-efficient (for example, not requiring manual human intervention for labeling camera trap and satellite images), so as to maximize donor funds disbursed to those in need, the conservancies.

Our Solution

The Deloitte solution is a data infrastructure built on the Amazon Web Services (AWS) cloud, which enables an efficient processing of data from various sources, either uploaded manually by various local participants, for example game guards, or retrieved automatically, such as the satellite imagery of the European Space Agency (ESA). 

Camera trap images are fed through a two-step computer vision pipeline using stead-of-the-art methods from the field of AI for conservation. First, the raw images are passed through an object detection model which detects the classes 'human', 'animal' and 'vehicle'. Detected animals are processed further by a second model, classifying them into around 50 relevant species. Using some heuristics to filter out duplication, individual animal sightings are counted over time to obtain a broad estimate of species abundance.

Machine learning automation extends to the view from the sky. Satellite images are regularly uploaded, and monthly composites derived from them in order to achieve cloud-free images. These are then fed to an advanced deep neural network (transformer model) trained by Deloitte in collaboration with the client to detect fire-cleared areas. New clearings (and thus unwanted human interference) are detected by comparing past and present images.

Together with patrol route and observation data from the smartphone app, the results of the two AI pipelines populate a file-based online storage solution (AWS S3) and feed the inputs to the payment rules of the NGO. Users are presented results of conservation efforts in the dynamic, mobile-friendly Quicksight dashboard, a cloud-native application deeply integrated into the AWS ecosystem. These are presented both in the form of KPI as well as the payments to be disbursed to communities. Customized add-ons to the Quicksight dashboard allow users drill-down to the original satellite and camera trap images.

Advantages/Benefits

• Reduced cost, brought by automation efficiencies, fully realizing the potential of camera trap and satellite images
• Speed, through continuous monitoring, enabling the NGO to intervene earlier in case of non-compliance
• Insight, by automatically combining data from various sources to create a holistic view of conservation
• Scalability, leveraging infrastructure largely based on a cloud-native 'Serverless Application Model' that grows with the project and requires only minimal maintenance
• Flexibility, via custom-built add-ons enhancing the functionality of the cloud-native dash-boarding solution to meet the specific needs of the NGO
• Attracting more donors through transparent performance and payment reporting

Ihre Ansprechpartner

Alexandros Melemenidis
Senior Consultant
Risk Advisory | Deloitte Digital
Tel.: +49 (0)69 756957677
amelemenidis@deloitte.de