Breaking Barriers in Habitat Mapping for Disease Ecology with Accessible AI Foundation Models and Satellite Data
Many of the world’s most persistent and burdensome diseases—like dengue, malaria, schistosomiasis, and others—are environmentally mediated. Their transmission depends on the presence of specific microhabitats that support the survival and reproduction of vectors, reservoir hosts, or pathogens themselves. The Stanford Disease Ecology in a Changing World (DECO) program, based at the Center for Human and Planetary Health, leverages these relationships to develop predictors and ecological levers of disease intervention. However, studying these relationships computationally requires large-scale data on ground conditions, including land-use and climate data.
Through an ongoing collaboration with Earth Genome—an organization that combines the power of satellite imagery and advanced machine learning to enable a range of ecological analyses, our DECO team is developing high-resolution, up-to-date land-use maps capable of capturing the fine-scale habitat variation most relevant to disease transmission.
To date, three major challenges have constrained our efforts to scale microhabitat-level information into mechanistic and operational disease prediction and management tools:
- Low ecological precision in public land cover data: Many commonly used remote sensing datasets classify land into broad categories like “cropland” or “forest,” which are too coarse to distinguish ecologically important habitat types. For example, malaria risk might vary drastically between rice paddies and coffee plantations, yet these are often indistinguishable in broad-brush publicly available land use products.
- Outdated or static spatial data: If the land cover or habitat data disease risk maps are not regularly updated, they quickly become outdated—especially in rapidly changing environments. This limits our ability to anticipate or respond to disease outbreaks as new threats emerge and thus their utility in outbreak prevention and disease management.
- High technical barriers to habitat classification: Producing custom, high-resolution habitat maps typically requires advanced skills in remote sensing, machine learning, and computer vision. For most public health researchers and practitioners, applying these methods remains extremely time-consuming and technically inaccessible.
Earth Genome’s Earth Index platform is helping overcome long-standing limitations in land-cover mapping for health applications. Built on AI foundation models, it preprocesses satellite imagery, converts it into embeddings, and runs it through advanced classification pipelines to rapidly detect fine-scale microhabitats—at speeds several orders of magnitude faster than traditional workflows. The platform prioritizes accessibility, enabling users with minimal expertise in computer vision or remote sensing to create custom land-cover classifications. Its interactive, human-in-the-loop design allows outputs to be refined and tailored to specific needs through a point-and-click interface or custom coding. For public health, this means researchers and practitioners can access up-to-date, high-resolution, ecologically relevant spatial data tailored to specific disease systems without advanced technical training.
DECO is now using Earth Genome’s Earth Index to understand how different land uses predict dengue risk in rural areas of Costa Rica. Our goal is to identify how specific habitats—like built environments (e.g., informal settlements), plantations, or fragmented forests—sustain dengue transmission. Focusing on microhabitat-level predictors will improve both the spatial precision, development of ecological interventions, and practical utility of disease risk maps. This approach brings us closer to proactive and targeted disease management, which is particularly useful in underserved rural and tropical regions where personnel and resources for disease control are often limited.
Our team at DECO is excited about the potential of this platform to support real-time, spatially precise, and ecologically driven public health interventions, helping to close the gap between environmental change and human health and wellbeing.
