Remote Sensing of Water and Environment
We have built a unique indoor optical remote sensing data acquisition system over the flumes at the Saint Anthony Falls Laboratory (SAFL) to study the impacts of optically complex waters on reflectance signals of water and environmental pollution. In this remote sensing facility, we can study the effects of bedload, suspended sediments, algal biomass, waves, and foams as well as plastic debris on the natural water reflectance signals. The system is equipped with a high-resolution ASD spectroradiometer (350-2500 nm), a source of light, a DSLR camera that images the subgrid fraction of debris within the field of view, and a sediment feeder.
Estimation and Forecast of Precipitation from Space
(image credit, NASA)
Our team develops modern machine and deep learning models for passive microwave retrievals of precipitation data from active and passive satellite observations. Coincidences of W-band radar from CloudSat satellite and microwave imagers onboard many weather satellites provide a unique opportunity to retrieve precipitation and its global phase change from rain to snow. The figure on the right-hand side shows estimates of precipitation at different phases over Greenland by our algorithm (second row) while the first row shows the brightness temperatures from the GPM radiometer at different frequencies and the last row represents the result of the Goddard Profiling Algorithm, ERA5 reanalysis and the radar onboard the GPM satellite. This algorithm used extreme gradient boosting decision trees for the detection of rain and snow in conjunction with a Bayesian retrieval scheme in the embedding space of a deep neural network for the estimation of the rates.
Our team also is developing deep learning models for the nowcasting of precipitation using satellite data. We demonstrated that for short-term forecasts, deep-learning models that learn from a sequence of satellite data can outperform global precipitation forecasts by the numerical weather models.
Remote Sensing of Soil-Snow-Vegetation Continuum
The snow is a radiometrically elusive medium however it is a lossless medium when it is dry. A small amount of water content can make the snow extremely lossy obscuring the bottom soil emission in microwave bands. Our team develops radiative transfer models and deep learning methodologies for remote sensing of soil freeze-thaw dynamics and vegetation water content over Arctic landscapes. For the first time, we could fill the gap in retrievals by the Soil Moisture Active and Passive Satellite (SMAP) over snow-covered surfaces.
From desire I rush to satisfaction, from satisfaction I leap to desire.
― Johann Wolfgang von Goethe
The top row shows the SMAP official product of soil moisture and vegetation optical depth and the bottom is the result of new retrievals by accounting for the multiple refractions withing the intervening snow layer.