Scribd
Upload
36 views2 pages

Earth Space Science

The Ames Intelligent Data Understanding Group uses data mining and analysis techniques to support Earth and space science. They build on previous machine learning research at NASA Ames to create tools that help scientists understand and assimilate data. Their methods include using models trained on rich data to estimate values for other datasets, predicting photometric redshifts from broad band photometry, and analyzing data at different scales using self-dissimilarity to explore structures present. These novel approaches help scientists answer questions more efficiently.

Uploaded by

kishore13
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
36 views2 pages

Earth Space Science

The Ames Intelligent Data Understanding Group uses data mining and analysis techniques to support Earth and space science. They build on previous machine learning research at NASA Ames to create tools that help scientists understand and assimilate data. Their methods include using models trained on rich data to estimate values for other datasets, predicting photometric redshifts from broad band photometry, and analyzing data at different scales using self-dissimilarity to explore structures present. These novel approaches help scientists answer questions more efficiently.

Uploaded by

kishore13
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

Intelligent Data Understanding for Earth and

Space Science
The Ames Intelligent Data Understanding Group uses its expertise in Data Mining & Data
Analysis to support the fields of Earth & Space Science.

Background Figure 1
The Intelligent Data Understanding (IDU) Group is Greenland, from MODIS year 2000 day 140 time 1830
collaborating with domain scientists to answer true channel 6.
pressing scientific questions in their fields and aid in
knowledge discovery. The group builds on the
significant successes and reputation that previous
Machine Learning researchers have had at NASA
Ames. Our main goal is to create tools and methods to
aid in the assimilation and understanding of scientific
data to best advance NASA's missions. We also
advance scientific data understanding through
research, and develop methodology for assisting in the
efficient and cost effective collection of scientific
data.

Research Overview

Virtual Sensors- Using Data Mining Techniques to


Efficiently Estimate Remote Sensing Spectra
Various instruments are used to create images of the
earth and other objects in the universe in a diverse set
of wavelength bands. Many applied science questions Figure 2
that are relevant to the Earth Science remote sensing AVHRR prediction from year 2000, day 140, time 1839
community require analysis of enormous amounts of using a Multilayer Perceptron
data that were generated by instruments with disparate
measurement capabilities. The IDU Group addresses
this problem using Virtual Sensors. Virtual Sensors is
a method that uses models trained on spectrally rich
data to fill in unmeasured spectral channels in
spectrally poor data. We demonstrate this method by
using models trained on the high resolution Terra
Moderate Resolution Imaging Spectroradiometer
(MODIS) instrument to estimate what the equivalent
of the MODIS 1.6- m channel would be for the
National Oceanic and Atmospheric Administration
Advanced Very High Resolution Radiometer
(AVHRR/2) instrument. Simulation of the 1.6 micron
channel would improve the ability of the AVHRR/2
sensor to detect clouds over snow and ice.

NASA Ames Research Center January 2006


Intelligent Data Understanding for Earth and Space Science

Novel Methods for Predicting Photometric Redshifts Figures 3 (a & b)


from Broad Band Photometry using Virtual Sensors: Self dissimilarity results for satellite images. The original
We calculate photometric redshifts from the Sloan Digital images are shown on the left (pixel values have been
Sky Survey Main Galaxy Sample, The Galaxy Evolution digitized to either 0 or 1) and self dissimilarity signatures
Explorer All Sky Survey, and The Two Micron All Sky are shown on the right. Even with the crude pixel
Survey using two machine learning methods. We utilize thresholding the two images are seen to have different
the broad-band photometry from the three surveys structure at different scales. In particular image 3(b) has
alongside Sloan Digital Sky Survey measures of structures at very large scales that differ significantly from
photometric quality and galaxy morphology. Our first the small scale structure. This may be due to the relative
method draws from the theory of ensemble learning while lack of clouds in the second image.
the second employs Gaussian process regression both of
which allow for the estimation of redshift along with a 200

400

measure of uncertainty in the estimation. These two 600

methods are compared to a well known Artificial Neural 800

1000

Network training-set method and to simple linear and 1200


3(a)
quadratic regression. We also demonstrate the need to 1400

1600

provide confidence bands on the error estimation made by 1800

both classes of models. A key contribution of our group is 2000


200 400 600 800 1000 1200

to quantify the variability in the quality of results as a


function of model and training sample. We show how
simply choosing the "best" model given a data set and
200

400

model class can produce misleading results. We show that, 600

using additional quality and morphology indicators rather 800

1000
3(b)
than only the Sloan Digital Sky Survey broad-band 1200

imaging data, redshift accuracy can be improved by 10s of 1400

1600

percent. Our work can be expanded to other photometric 1800

surveys where sufficient redshift calibration objects exist. 200 400 600 800 1000 1200 1400 1600

Relevance to Science Mission


Self-Dissimilarity A New Data Analysis Tool Directorate
We are developing a family of novel methods to probe Data mining & data analysis techniques developed by the
aspects of data sets usually left unexamined by machine Intelligent Data Understanding Group will enable space &
learning algorithms. By mimicking the process that goes earth scientists to better answer important questions in
on under a microscope as the magnification is changed, we their fields. Our methods help scientists obtain results
probe a data set at different scales to build models of the sooner, extracting new information from old records. We
structures present in the data at different scales. By minimize their costs by using more effective methods to
comparing the resultant models at different scales get the information they need.
information flow between scales can be explored.
The method has been applied to one- and two-dimensional Points of Contact:
artificial and real images. Work is under way analyzing the
data for the cosmic microwave background recently Dr. Ashok Srivastava (Principal Scientist and Group
obtained from the Wilkinson Microwave Anisotropy Probe Leader)
(WMAP). The method promises a new tool through which 650-604-2409; Ashok.N.Srivastava@nasa.gov
to probe datasets. When applied to a data archive the http://ti.arc.nasa.gov/people/ashok
method could be used to index the data. In analysis of
remote sensing data, Self-Dissimilarity could reflect Dr. David Wolpert
hidden levels of bio-diversity; indicate progressive die- 650-604-3362; David.H.Wolpert@nasa.gov
backs before they become apparent or reflect changes to http://ti.arc.nasa.gov/people/dhw/
spatial distribution of climates.
Group Web Page:
http://ti.arc.nasa.gov/datamining

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy