ResearchPortfolio: Nabin Malakar
Welcome to my research portfolio page. This is a page of my ongoing research activities.
Lands Surface Temperature and Emissivity Product for NASA (MOD21)
I am proud to be part of the team which developed a Land Surface Temperature and Emissivity (LST&E) product for NASA was developed at JPL.
The pdf link:
https://emissivity.jpl.nasa.gov/downloads/examples/documents/MOD21_ATBD_Hulley_v2.2a.pdf
Aerosol Optical Depth
Aerosol are tiny particles in the atmosphere. These particles are measured by the extent they deplete the transmission of light by absorption or scattering, known as aerosol optical depth (AOD). The AOD index is measured from both the ground stations as well as from the remote sensing methods. However, when inter-compared measurement from these two methods are not free from bias. We developed machine learning approaches to study the cause of bias, and to study the relevant variables towards explaining the bias. (paper is available here) The method is further extended to study the global distribution of small particles (PM2.5).
We also applied data fusion of active and passive sensor to obtain regional PM2.5 map. More descriptions and paper link http://www.nabinkm.com/2014/01/assessing-surface-pm25-estimates-using.html
One of my mentee was rewarded the first prize (News, link).
Downscaling Regional Climate Meteorological Forecasts
We have applied machine learning technique to downscale meteorological variables of interest over the northeast USA. The results were presented in the machine learning conference organized by NYAS in NYC. Please see an example below.
Inferring Incomplete datasets using EOF
Remote sensing data are often incomplete because of various factors such as cloud cover. We have applied EOF methods to infer the incomplete data sets.
Please follow the following link to watch a demo.
Inferring Incomplete datasets using EOF: Application to Ionospheric data
Using the technique of EOF method and time series information, we could infer incomplete ionosphere data. Paper here ...
Principle-based Automation
I have always been fascinated by the power of Machine-learning techniques and its possible application to societal benefit. My PhD research involved the use of Bayesian and Maximum entropy techniques and decision-theoretic approach to experimental design at its core. I worked on the algorithm design for autonomous decision-making system for model-based exploration. Basically, it involved a cost function or utility to decide on possible alternative actions. We considered entropy as a measure and were able to arrive at a principle-based collaboration scheme for two intelligent agents.
My paper on sensor characterization using mixture of Gaussian model is available here (pdf 110K). I also designed the entropy-based search algorithm for experimental design (http://arxiv.org/abs/1008.4973, and http://arxiv.org/abs/1111.3421).
Pelagic habitat Classification: GOM
I have been working on projects analyzing remote sensing satellite data and applying various machine-learning techniques. In conjunction with Dr. Rebecca J. Allee (NOAA), we analyzed the remote sensing data to classify the water column component in the Gulf of Mexico. We were able to provide a consistent machine learning classification to the remote sensing data and establish a correspondence between the Coastal and Marine Ecological Classification Scheme (CMECS) and the ML results. The results are available at URL: http://www.tinyurl.com/SOMGOM2
This helped delineate the water preference for commercial fish species. This could be used by decision makers for conservation, and fishermen to decide on where to go for fishing as well.
Figure showing the ML classification of North Gulf of Mexico. The result was used to delineate the water type preference for commercial fish species.
In UT Dallas, I worked on the societal applications of Machine Learning to build conceptual understanding of physical phenomena from real-world data. I have been working with big as well as collaboration projects. Please click on the list for details.
Oil Sleek from RS data
Oil sleek detection from space is a challenging task due to the fact that the reflective properties of oil on water are similar. Sun-glint methods are often used to visually identify the sleek. We explored ML techniques to identify the oil sleek in Gulf of Mexico, and were successful to some extent. In the future, integrating the social media with remote sensing data will help establish an alert system to detect the oil sleek.
Poster: https://docs.google.com/viewer?a=v&pid=sites&srcid=ZGVmYXVsdGRvbWFpbnxuYWJpbmttfGd4OjNlNDU1ODdiOTkwMzc5ZjA