Leveraging heterogeneous data and machine learning for mapping connectivity and infrastructure provision in Amazon

Abstract The Amazon region faces the critical challenge of balancing economic development with urgent environmental preservation. This study leverages advanced geospatial analytics and machine learning to map and analyze both physical and digital connectivity across the Amazon. We integrate heterogeneous datasets—including road networks, navigable waterways, building footprints, internet, and electricity access—within a uniform H3 spatial indexing system. Our analysis exposes stark infrastructure disparities: only 33% of the region has adequate road access. While 51% is served by navigable waterways, this infrastructure remains critically underutilized, with just 621 docks identified. Digital connectivity is similarly fragmented, with 42% of the area lacking sufficient data for assessment. K -means clustering reveals four distinct connectivity profiles: high-connectivity urban hubs (5%), medium-connectivity rural zones (60%), low-connectivity pristine forests (26%), and isolated remote settlements (9%). SHAP analysis identifies transportation accessibility as the primary driver of development patterns, followed by internet access and proximity to essential services. Utilizing these insights, we developed a machine learning-derived infrastructure provision index, identifying 12.4% of the Amazon as critical development zones requiring targeted intervention. This comprehensive mapping framework provides an essential evidence base to guide sustainable development strategies that enhance livelihoods while safeguarding the Amazon’s ecological integrity.