This is the fourth post of a four-part series showcasing how WRI’s Data Lab is addressing global challenges through innovative technology strategies and products. In each part, we focus on the strategies, specific products and underlying technologies shaping our approach. Our goal is to inspire collaboration across sectors and highlight how data-driven solutions can drive meaningful, scalable impact.
We see massive opportunity in solving global climate action challenges through AI, and we’re investigating two key areas where we believe AI can play a critical role. By partnering with leading organizations like Meta and Google, we’re excited to apply our expertise in impact product development alongside cutting-edge technological innovations. These efforts are designed to address critical gaps in climate data and action, and we’re eager to collaborate with like-minded organizations and individuals. If you’re interested in exploring these opportunities with us, we’d love to hear from you.
1. Foundational language AI models for low-tech communities
Generally, the benefits from the AI boom over the last several years have been accrued by the privileged demographics of those designing the models — the most popular foundational models are trained on mostly English text and require large amounts of (local or cloud) computing resources to train and run.
Many WRI’s data applications’ user demographics are different than the implicit (or explicit) target user of many popular models and AI tools. Many of our users live in the Global South, aren’t native English speakers and often have limited access to computing resources, whether it be cloud services or the latest smartphones.
We see significant opportunities for these users to benefit from AI models that speak their language, literally. There’s a significant gap to be filled with foundational models that utilize the latest demonstrated advances in on-device model efficiency, low-resource strategies for adapting to models to new languages and/or the inclusion of cultural nuances in model training to provide underserved users with useful AI models.
This potential can be seen in the field data collection context. If we can build models that work on the devices of researchers collecting field data in rural Africa, we can create a smoother user experience and ultimately have more efficient and accurate data collection results.
2. AI-powered satellite imagery analysis
Historically, analysis of satellite imagery was a very computationally intensive and expensive process, hence many workflows still utilize manual image interpretation. With advances in foundational AI models, pre-trained and with self-supervised learning capabilities, researchers can analyze satellite imagery and generate geospatial datasets faster and cheaper than ever before. Now, high-resolution, custom geospatial data insights are more accessible than ever.
For example, WRI’s Land and Carbon Lab has partnered with Meta to develop a groundbreaking 1-meter global tree canopy height dataset, utilizing Meta’s DINOv2 foundational AI model. This dataset detects single trees on a global scale, unlocking a number of opportunities for tracking land-use emissions and progress on various conservation and restoration projects.
We believe that we’re just scratching the surface of this technology and its potential applications. We’re particularly interested in exploring how these advances in imagery and dataset generation can build a better picture of land use around the world, and specifically the “fuzzy” borders between human land use and nature, which has historically been very challenging to understand at scale.
3. Cloud native geospatial communities and technologies
As an organization primarily focused on user-facing data and applications, using cloud infrastructure effectively is the foundation of all we do. Without the open data formats, standards and software packages, the near-real-time maps and scalable AI systems we build wouldn’t be possible. And that’s why we’re so excited about the growth and progress of the cloud native geospatial ecosystem. This work, led by communities like the Cloud-Native Geospatial Forum (CNG) and Open Geospatial Consortium (OGC) as well as many individuals and companies, is making rapid strides in standards, data formats, and infrastructure puts the promise of geospatial insights within reach for so many more.
We believe that this work has massive potential to make geospatial analysis not just better but cheaper, faster, and simpler. In turn, this will accelerate the speed of development and improve the adaptability and resilience of our systems. At WRI, our focus remains on building user-facing technology, but we are excited to increasingly integrate these tools and approaches into our work and share our insights and impact with the community.
Across this series, WRI’s Data Lab outlines the strategies, products and technologies we believe hold real promise for accelerating action across people, nature and climate. These approaches reflect where we see the biggest opportunities to close critical gaps.
While many of these concepts or strategies are already in development at the WRI Data Lab, others need collaboration, investment and technical expertise to move forward and realize their potential. By sharing these opportunities, we aim to spark dialogue and build partnerships with those equally committed to technology-driven solutions that address today’s urgent challenges.
If you’re working on any of these exciting areas — as a researcher, funder, policymaker or technologist — let’s talk. Reach out to us at datalab@wri.org.
Other posts in this series:
