In a previous Advisor, we discussed how AI-powered platforms are transforming wildlife conservation and research by optimizing data collection and monitoring/analysis efforts and facilitating global collaboration. We covered key applications such as data management, animal monitoring/tracking, wildlife protection, and species identification automation. This Advisor continues that conversation by examining AI application development and exploring AI models used in conservation and research efforts.

A Multistep Process

Building AI applications for wildlife conservation and research is a multistep process that consists of: data collection and labeling, training and testing appropriate AI models, and deploying solutions to supporting field devices:

• Data collection. These AI applications require large amounts of data to train and test the models. Such data can include images or audio files of wildlife and information on their environment or habitat. The data is typically captured by cameras, audio recorders, drones, and satellites operating in the field. It can also be sourced from websites and social media platforms (i.e., images posted by animal lovers and “citizen scientists”).

• Labeling. Collected data is then labeled with specific information, which, depending on application requirements, can include type of species, location, behavior, date, environmental/habitat conditions, and so on. Data labeling may be performed by humans or by using automated techniques including AI.

• Training and testing. The labeled data is then split into training and testing datasets. The former is used to train the appropriate AI model(s). The latter serves to evaluate and assess the accuracy of the developed model when exposed to new data.

• Deployment. Deploying the model involves integrating it into various devices (e.g., field cameras, drones, satellites) and hardware and software platforms necessary for data processing and analysis. Next, the model is evaluated and refined using feedback gained from field trials.

AI Models Used in Wildlife Conservation & Research

AI models used in wildlife conservation and research applications are typically tailored to specific domains or scenarios, such as species identification and tracking, habitat monitoring, and biodiversity monitoring (tracking populations and behaviors of different species). Some of the common models employed for such applications include convolutional neural networks (CNNs), support vector machines (SVMs), and random forests. Some applications may use several different types of AI models.

CNNs

CNNs excel at image and video analysis, aiding in species identification and monitoring. CNN development for wildlife conservation applications involves training the model using large datasets of labeled images of different species so it learns to recognize patterns and features specific to each species. This capability lets the model accurately identify different animal species.

CNNs are particularly useful for monitoring wildlife populations and tracking endangered species more efficiently. Accurate identification of animal species is also key to understanding biodiversity richness and studying the impact of climate change on species distribution within a specific region. Google’s DeepMind division collaborated with Tanzania’s Serengeti National Park to develop an application for tracking wildlife populations that uses CNNs (among other models) to analyze images captured from field cameras.

SVMs

An SVM is a machine learning algorithm that offers robust classification capabilities to support various wildlife conservation and research efforts. Scientists and researchers use SVMs to classify species according to specific features extracted from images or audio. For such applications, SVMs are trained on labeled data where each data point is associated with a specific class (e.g., species). The model learns to classify new data points by finding the optimal boundary that separates different classes.

For example, researchers might use SVMs to classify bird species based on features extracted from song or call recordings. These features include aspects like frequency, duration, and patterns of the sounds. Once trained, the model can accurately classify new audio recordings based on these learned features, aiding in the identification and monitoring of various bird species.

Random Forests

Random forest models are used for applications like habitat suitability modeling and predicting species distribution. Random forests are groups of decision trees trained on various subsets of the data. Each tree makes a prediction, and the final output is the majority vote or average of all trees. (Admittedly, this is a simplification of what is actually a complex process, but it does offer the basic idea of SVM functionality.) This approach is useful for handling large datasets with many variables. For example, researchers might use random forests to predict the presence of species in different habitats based on environmental variables such as temperature, vegetation, and altitude.

Issues & Considerations

There are a number of issues to consider when applying AI to wildlife conservation and research:

• Data quality and volume requirements. AI model development requires large volumes of quality data for training and testing purposes. Acquiring sufficient amounts of such data may prove problematic in instances involving rare or hard-to-access wildlife. This can reduce model accuracy. Generating synthetic data to augment training datasets is one option that helps improve model accuracy in such situations.

• Image/audio quality. This is a particularly pressing problem. Poor-quality images from trail cameras, drones, satellites, or mobile devices can lower model accuracy. The same goes for audio recordings.

• Privacy. Cameras, particularly those on drones, may raise privacy issues for individuals and disturb wildlife.

• Model maintenance. Wildlife behavior and habitats can change rapidly, requiring updating and retraining of AI models.

• Costs and implementation hassles. Purchasing, implementing, and maintaining AI technology, including drones and cameras, can be expensive. The growing availability of cloud platforms and services for wildlife conservation and research, like Wildlife Insights, helps alleviate the high costs and headaches associated with AI implementation.

Conclusion

AI applications are revolutionizing wildlife conservation by enabling efficient monitoring, protection, and research of diverse species. The examples discussed here represent only a portion of the AI techniques transforming this field. However, they offer a good overview of the kinds of AI techniques that organizations can use to further their research efforts. Finally, I’d like to hear your opinion on using AI for wildlife conservation and research. As always, your comments will be held in strict confidence. You can email me at experts@cutter.com or call +1 510 356 7299 with your comments.