Research

Growing global populations have increasingly engaged in unsustainable practices, leading to the alarming depletion of natural resources, disruption of biodiversity, and climate disturbances. These actions pose significant threats to the well-being of future generations. In order to comprehend and address these land transformations effectively, the application of imaging technologies has facilitated the observation and comprehension of these complex changes and underlying drivers. By harnessing such technologies, we can design sustainable solutions and make informed decisions to mitigate these challenges. Notably, microwave remote sensing data, specifically Synthetic Aperture Radar (SAR), has emerged as a game-changing tool, providing near real-time imaging capabilities regardless of weather conditions or time of day. However, the sheer volume of SAR image data acquired from various satellites in orbit exceeds the capacity of skilled professionals available for analysis, hindering the derivation of valuable insights.

To address this disparity, my research focuses on machine learning-based approaches that process vast amounts of global-scale data and autonomously extract intricate insights from SAR imagery, thereby reducing the need for extensive human involvement. My ambition lies in developing robust machine learning methodologies that leverage radar remote sensing images to derive accurate, reliable, and actionable information pertaining to large-scale land changes and their underlying drivers. By enabling this advancement, I aim to bridge the gap between the growing need for comprehensive analysis of SAR data and the limited availability of skilled experts, ultimately facilitating sustainable land management practices.

The Land Use Change Alert (LUCA) system represents the culmination of this research vision, translating years of methodological development into an operational platform that addresses global forest monitoring challenges at scale. LUCA embodies the principles outlined above: it autonomously processes daily Sentinel-1 SAR acquisitions across the entire globe using advanced machine learning pipelines deployed on Google Earth Engine and Google Cloud Platform, delivering near-real-time forest land use change alerts without the constraints of cloud cover or daylight. By integrating sophisticated change detection algorithms with scalable cloud infrastructure, LUCA demonstrates how machine learning can unlock the full potential of SAR data streams, providing conservation organizations, government agencies, and research institutions with actionable intelligence for forest monitoring, illegal deforestation detection, and law enforcement. This system exemplifies the transition from theoretical research to practical impact, enabling data-driven decision-making for sustainable land management at a scale previously unattainable through manual analysis.

Looking ahead, LUCA will evolve as new technological capabilities emerge. One of the most significant developments on the horizon is the arrival of new radar satellite data from NISAR, a joint mission between NASA and ISRO. This satellite will dramatically reduce the time between observations, allowing LUCA to shift from biweekly to weekly alerts—a critical improvement because faster detection enables faster response by authorities on the ground. The launch of NISAR is particularly transformative, as its L-band radar will enhance our ability to characterize changes in forest structure, improving detection of both deforestation and forest degradation. Beyond temporal improvements, we are working to deepen LUCA’s analytical capabilities by moving from simple change detection to causal attribution—understanding not just where forest loss is occurring, but what type of change (deforestation, degradation, or management practices) and why it is happening (agriculture, infrastructure development, or logging). This advancement involves integrating SAR data with optical imagery from Sentinel-2, Landsat, and PlanetScope, with AI playing a central role in differentiating between drivers of forest loss and mapping forest regrowth.

I’m currently working on implementing intelligent prioritization algorithms that use cloud computing and AI to automatically rank and highlight the most urgent deforestation events, such as those occurring near protected areas or in regions with high risk of illegal activity, making the system more actionable for decision-makers. Finally, LUCA’s monitoring framework is being expanded beyond traditional forests to other critical ecosystems, including urban tree cover, mangroves, peatlands, coastal zones, savannas, and grasslands. This expansion will enable tracking of degradation in ecologically significant but underrepresented landscapes such as Brazil’s Cerrado and African savannas, broadening the system’s impact on global environmental conservation.