CSCI 8970 – Colloquium Series – Fall 2010 – First Event
Discovery of Patterns in Global Earth Science Data using Data Mining
Monday, September 13, 2010
|
Presenter: |
Vipin Kumar |
His lecture was interesting and informative, but it left us with a challenge. To truly take advantage of the data that is increasingly being gathered, more specific and encompassing algorithms need to be developed. Also, one can anticipate that with the success of the program, future more accurate and detailed satellites and imagery systems will be developed.
The lecture talked about the challenges in analyzing Eco-Climate date. During his lecture he showed comparisons of global sea surface temperatures, a 3D rendering of the globe and how “green” different locations are around the world. The satellite permits data from fires around the world to be analyzed including the extent of the damage.
He illustrated the applications of data mining. The satellite can analyze the changes in forests which account for 20% of greenhouse gas emissions (2nd only to fossil fuel). Twenty five percent of climate change production can be provided by terrestrial carbon. The “carbon offset” notion takes root on the REDD program within the United Nations. REDD stands for Reduce Emissions from deforestation Destruction. The goal of the program is to determine where, when, ad why land cover changes occur. The program has become increasingly popular, and even made TIME’s list of 50 best inventions on 2009.
The system obtains data from forest fires, droughts, floods, logging, as well as reforestation. The system will help quantify the carbon impact of these changes, and understand their relationship. While the state of the art algorithms for land cover change detection did not scale up well to analyzing data at the global level, this satellite scans the whole surface of the earth every single day. This billion dollar a year project from NASA not only allows their scientists to analyze the data but the data has been made freely available to other scientists across the world.
Existing algorithms have problems, because currently there is too much noise on the images. As a result his group has developed an algorithm to analyze the data. He demonstrated the capability of the system by illustrating some of the global changes taking place from logging in Canada, shrinking of lakes in Africa, the impact of Katrina, reforestation in China, and Deforestation in the Amazon. Deforestation in the Amazon accounts for 50% of all humid tropical forest clearing. That amount is 4 times as much as the next country, Indonesia, which accounts for 12.8% of tropical forest clearing. Despite its high hopes, Indonesia’s attempt to increase its bio fuel industry through deforestation and planting, has proved disastrous.
After presenting data from his research, Dr. Kumar challenged the audience by encouraging the development of better algorithms, multi-resolution and multi-variate analysis, and data quality improvement. During his lecture he also provided an overview of a newer project promoting a data driven approach. NSF Expedition intends, through the collaboration of various universities, to utilize the physics based model and find what can be achieved with this model, and what can be built in top of it. The program is promoting the need to study the ecosystem as a complex environment.
