There are many use cases for AI to monitor waste and pollution, enabling timely intervention. A high-impact example is the use of GeoAI tools to monitor agricultural plastic. Farmers are increasingly relying on plastic in production due to its ability to protect crops, reduce pesticide consumption, extend growing seasons and increase yields by up to 60 per cent. Plastic covers for greenhouses can protect crops from adverse climate factors and animals, and can enhance solar radiation to create more favourable growing environments.
However, the mismanagement of large amounts of agricultural plastic can negatively impact the environment and agricultural yields. Given that most agricultural plastic products are single use, their frequent replacement generates high volumes of waste in landfills. The accumulation of film residue in agricultural soils directly contributes to negative soil properties and plant growth, such as reduced crop yield, plant height and root mass. Improperly disposed plastic in agricultural soil releases microplastics that pose harm to ecosystems, threaten food security and potentially harm human health.
Deep learning models can enable agriplastic modelling by analysing satellite and drone imagery to identify plastic-covered fields. For example, a study utilized multi-spectral remotely sensed imagery and classic ML classifiers to map the plastic cover in rural areas of southern Italy with an average classification accuracy of over 80 per cent. Another method revealed that AI-generated plastic mapping yielded more accurate results than hand-crafted methods that had a low generalization ability and high sensitivity to noise. Agriplastic mapping models developed in the 2024 ITU GeoAI Challenge featured various spectral indices relevant to agricultural productivity, such as a plastic index, vegetation and water indices to assess green cover and moisture, and a bare soil index.
AI for Good in focus: Multimodal adaptation of large language models for smart mobility in Africa
LLMs and vision-language models can deliver smart mobility solutions in Africa in order to address the region’s unique transportation challenges. On 29 April 2025, Dr. Ahmed Biyabani, Associate Teaching Professor at Carnegie Mellon University Africa, with a specialization in AI and digital transformation, led an AI for Good webinar exploring this topic. The webinar explored how LLMs can process various data sources such as GPS, images and time-series data, in order to predict congestion and optimize traffic flows. It highlighted the value of employing domain-adapted models trained on mobility-specific datasets from African cities, arguing that while general purpose LLMs perform well in broad tasks, they often require domain specific expertise in order to be applied in transport-related scenarios.
In order to effectively address this issue, the webinar explored how retrieval augmented generation can be used to improve LLM performance. Dr. Biyabani noted that this could be done by integrating real-time, localized data such as GPS, schedules, weather patterns, infrastructure maps and traffic warnings to assist in contextual decision-making. The webinar incorporated case studies from Kigali, Nairobi and Lagos to demonstrate how local data can improve model accuracy. It also proposed a three-stage system architecture - namely data collection (i.e curating Africa-specific datasets), model adaptation (i.e fine-tuning vision-language models and deploying retrieval-augmented generation for dynamic updates) and deployment (i.e a web-based application for real-time mobility and insights).
This framework provides the ability to improve traffic management, emergency response and public transportation services in low-resource settings. The webinar additionally explored issues such as computational resource requirements, data privacy, bias reduction and human oversight, in addition to emphasizing the need for access to high-quality, relevant data. It noted that this framework can be applied to sectors beyond transportation, such as healthcare and logistics, therefore demonstrating the applicability of multimodal AI in emerging economies. The webinar concluded with a call towards more collaborative data-sharing initiatives and the need to develop supportive policy frameworks which can help grow innovative solutions across the African continent.
