Precipitation Passive Microwave Retrieval
Estimation of precipitation from space is one of the most exciting uses of earth remote sensing. The upwelling earth radiation in microwave bands contains spectral signatures that allow us to measure global precipitation from space. In the past few years, our team has developed a new passive retrieval algorithm to obtain improved estimates of precipitation from space using passive radiometric measurements provided by the Tropical Rainfall Measuring Mission (TRMM) and Global Precipitation Measurement (GPM) satellites.
Our retrieval algorithm, called ShARP (Shrunken Locally Linear Embedding for Passive Microwave Retrieval of Precipitation), relies on manifold learning via Bayesian sparse approximation and promises improved passive retrieval of precipitation, especially over land and at the vicinity of coastlines. Some key outcomes of the ShARP retrievals are shown in the figures below in comparison with the standard TRMM-2A12 passive retrieval products during the entire calendar year 2013. It can be seen that ShARP promises improved retrievals over arid and semi-arid (e.g., Sahara Desert) regions and mitigates the commonly observed over-estimation of rainfall over snow-covered land surface (e.g., Tibetan highlands).
A software package, using MATLAB computing language, has been developed. The package can be freely downloaded and used for research purposes.The codes are primarily tested using MATLAB (R), 2014-15. The package requires the "signal processing" and "optimization" toolboxes (see, the readme file). The current version of the algorithm is memory intensive as it uses spectral and rainfall dictionaries containing a large number of brightness temperatures and their coincident rainfall profiles. A random access memory of at least 16 GB is recommended depending on the size of the employed dictionaries for the current version. An ongoing effort is underway to produce a parallel version of the algorithm to make more efficient. Please read more ...
Hurricane danielle, 2010
A squall line, GA, 2013
Cyclone Sidr, 2007
Top panel shows all active retrievals of the TRMM overpasses in 2013. The image shows mean values of more than 5200 of the TRMM overpasses. Active PR-retrievals are considered as a relative reference in our comparison.
Bottom panel compares the standard 2A12 passive retrievals with the results by the ShARP algorithm. It is shown that how the ShARP algorithm can reproduce the radar active products from passive observations of multi-spectral brightness temperature values. As is clear, ShARP promises improved retrievals over arid and semi-arid climate (e.g., Sahara Dessert) and reduced retrieval biases over cold mountainous regimes covered with snow (e.g., Himalayas and Tibetan Highlands).
Compressive Earth Observatory
The Compressive Earth Observatory (CEO) is a new conceptual framework that uses Compressive Sensing (CS) theory for the efficient estimation and sampling of land atmosphere state variables and fluxes from space. In a paper, published in Geophysical Research Letters, (Ebtehaj et al., 2015) using the retrievals of Atmospheric Infrared Sounder (AIRS) on board of NASA’s Aqua satellite, we demonstrated that: 1) the geophysical fields such as temperature and moisture fields are sparse in the wavelet domain, throughout the depth of atmosphere and 2) using a small set (30%) of random samples of temperature and moisture fields, the CS theory enables us to recover the entire field with high degree of accuracy. The main messages are: a) we may be able to design a next generation of sensors that allow to collect a smaller number of samples without compromising the accuracy of the earth observatory systems. b) With current sensing protocols, we may be able to design compatible and operationally viable random sampling schemes that enable significant reduction of the sampling density from space, leading to increased life span of the spacecraft, reduction in latency time of data transfer, and speedy retrievals for early warning systems.
Iterative reconstruction of an eight-day (2/10-10/01/2002) average surface skin temperature [Kelvin] using CS theory and data provided by the NASA's AIRS instrument. The down-sampled field only contains 35% of randomly chosen pixels. The energy of the reconstruction error is less than 0.01% of the total energy of the original temperature field. This reconstruction uses a stationary wavelet transform as the sparsifying operator and relies on the Fast Iterative Shrinkage Thresholding (FISTA) for solving the CS problem. Please read more...