Nighttime Remote Sensing
Nighttime VIIRS observations shed unique insight on Earth at night. This has a variety of applications in studying land, atmospheric, ocean and cryospheric processes at a very high temporal resolution enabling both near-near time analyses for monitoring natural hazards as well as targeted science analysis from higher level derived products. Some areas of nighttime analysis using VIIRS observations include urbanization studies, power outage and electricity access, disasters and recovery, socioeconomic trends, light pollution, anomalous thermal events such as forest fires, gas flares, volcanoes, lunar-illuminated smoke from fires, lunar-illuminated nighttime clouds, aurora, lightning, etc.
I am using machine learning and signal processing techniques to develop algorithms for maximal utilization of nighttime VIIRS observations and the Black Marble product suite. These are centered around using unsupervised learning for detection and mapping of anomalies such as fires, gas flares, power outage and anomalous trends in human activity as well as lunar illuminated mapping of nighttime features such as clouds, lightning and smoke.
Multitemporal Analysis of Multispectral Satellite Image Time-Series for Land Surface Change Recovery Detection
MODIS observations acquired over the last two decades has resulted in a unique dataset that is ideal for monitoring long term trends in different Earth system components. The focus of this study has been to develop a change detection framework for dense multispectral land surface reflectance time-series. This includes the formulation of i ) a novel time varying frequency model of land surface relfectance time-series characterizing change and post-change recovery rate, ii) unsupervised and interpretable change representations for clustering events and explaining change decisions. Change events considered in the study include both sudden (forest fires, floods) and gradual (droughts, coastal land gain) events.