Wetlands are essential for fulfilling global commitments on topics of climate, biodiversity, and sustainable development. However, due to climate change and human disturbances, there are growing concerns and significant attention being directed towards the degradation of wetlands, which is manifested through both a loss in area and a reduction in their functionalities. This thesis presents data-driven, training-free computer vision algorithms designed to address the challenges of monitoring and interpreting wetland dynamics. Through the development of these methodologies, we aim to answer key research questions regarding the temporal and spatial patterns of wetland refilling, the impact of climate variability on wetland dynamics, and the influence of anthropogenic activities, particularly the initiation of wildfires, on wetland ecosystems.

First, the superior penetrating capabilities of L-band microwaves from the Cyclone Global Navigation Satellite System (CYGNSS) are leveraged to detect water through dense vegetation and cloud cover prevalent in the tropics. We developed a segmentation algorithm, coupled with spatial analysis tailored to CYGNSS data, to differentiate land and water. This effort culminated in the creation of the UC Berkeley Random Walk Algorithm WaterMask from CYGNSS (Berkeley-RWAWC) [rɔ:k], a quasi-global (37.4°N to 37.4°S), monthly waterbody map. This product significantly enhances the interpretation of seasonal and interannual variability in tropical hydrology, offering new perspectives for studies on these critical ecosystems. Building on this foundation, we advanced the algorithm towards operational hydrology by harnessing CYGNSS’ near-real-time capabilities. The temporal resolution of the product was refined to capture daily water dynamics, addressing the critical need for timely monitoring and response to increasingly frequent and extreme hydroclimatic events. The enhanced product's performance was validated in the Sudd wetlands, an area characterized by complex natural and anthropogenic influences on flooding. The demonstration underscored the potential of integrating cutting-edge remote sensing technologies with robust algorithms to enhance disaster response and management. Finally, the study focus shifts to the Pantanal, the world’s largest tropical wetland, which has recently been affected by frequent mega wildfires. Multi-object tracking algorithms were developed and customized to label and track fire patterns, enabling an in-depth analysis of fire events in wetlands. The final chapter investigates trends in fire occurrences, examines the interplay between fire regimes and wetland refilling patterns, and explores the resilience of the Pantanal against extreme events.