• Subscribe

At Science Node, we need your help. We have ALMOST reached our fund-raising goal. In order to maintain our independence as a source of unbiased news and information, we don’t take money from big tech corporations. That’s why we’re asking our readers to help us raise the final $10,000 we need to meet our budget for the year. Donate now to Science Node's Gofundme campaign. Thank you!

What machine learning can tell us about climate change

Speed read
  • One of the most important ecosystems on the planet, the Amazon rainforest is threatened by rising temperatures
  • NASA scientists’ new method analyzes relationships between variables like temperature, precipitation, and vegetation growth
  • Same method can be used to look at other climactic relationships and their impacts on the earth

The Amazon rainforest is the largest tropical ecosystem on the planet, accommodating more than half the world’s animal species and producing flora that eventually gets manufactured into 25 percent of all modern medicines. Unfortunately, this abundance is under threat due to climate change.

<strong>Biodiversity stronghold.</strong> The Amazon rainforest covers most of the Amazon river basin in South America, encompassing more than 2 million square miles. It stretches through nine nations, with the majority contained in Brazil, Peru, and Colombia. It is one of the most biodiverse tracts of tropical rainforest in the world. Courtesy NASA. Numerous studies have investigated how rising temperatures may be altering the Amazon ecosystem, specifically plant growth. However, Kamalika Das, a machine learning expert and senior research scientist at NASA's Ames Research Center, has found that many of the existing results contradict each other.

“It was very interesting to me that these conflicting studies are being published in very reputable journals. This is solid work,” Das says. “So why is it that people are coming to different conclusions, even though there is nothing wrong with their techniques? What is it that they're missing?”

Das decided to look at the problem from a larger perspective than the micro-region analyses her peers were conducting.

“When I went in, I wanted to see which conclusion I was siding with,” Das says. “Somehow we ended up finding results that fit both, which was a big surprise for us.” 

<strong>Decision tree.</strong> Hierarchical model with five partitions illustrated as a decision tree, with leaves color-coded to match the geographic area corresponding to each partition of the Amazon region on the map. For each leaf, we show the five most important model features, sorted according to their contributions toward explaining the vegetation variance within those partitions. LST and TRMM indicate land surface temperature and precipitation, respectively, with the numerical indices referring to the season index. Courtesy Kamalika Das, NASA/Ames; Marcin Szubert, University of Vermont.Instead of compartmentalizing the issues as other studies had done, Das conducted a large-scale regression study. With the help from the Pleiades supercomputer at Ames' NASA Advanced Supercomputing Facility and the NASA Earth Exchange data pools, she developed optimization-based models that analyzed the relationships between various independent and dependent variables like temperature, precipitation, land elevation, and vegetation growth.

“It’s not temperature alone that affects vegetation; everything works together,” Das says. “Studying the correlation itself is not enough to understand the dynamics that are actually in play in this region. It's not the final answer.”

While the models produced by this study represent nonlinear relationships between all of these variables, they give a first glimpse of the vital relationships between climactic factors and vegetation growth in the Amazon. They also set the foundation for future studies by informing researchers about which variables are the most important, paving the way for causal analysis and predictions.

A project as complex as this presented its fair share of challenges, says Das. This was the first time she had ever combined techniques from machine learning and genetic programming, an optimization technique inspired by biological evolution. Handling all of the data proved to be difficult.

<strong>Prediction errors.</strong> This image shows (a) spatial patterns of prediction errors of an Amazon region GP-tree model, averaged over the years 2003–2010. Positive and negative values indicate underestimated and overestimated vegetation predictions, respectively. The model captures almost 80% variation in the data, a 10% improvement over a linear model. (b) Differences in prediction errors between linear and nonlinear models. Green color indicates locations where the model provides more accurate predictions. Courtesy Kamalika Das, NASA/Ames; Marcin Szubert, University of Vermont.“It comes from various sources, and the products have gone through multiple levels of processing before we even get to look at them. There are multiple assumptions built into each processing model, so the products are very diverse from each other,” Das explains. “It's not just getting one variable from one location or one file and getting another variable from another file and putting them together. A lot of work goes into getting everything to actually fit together.”

But now that the hard work has been done, the system Das developed can be used to look at problems outside of the rainforest, such as air quality monitoring. Many factors impact a place’s air quality, such as building density, amount of greenspace, and number of vehicles. All of these variables can be analyzed using a similar regression study to make predictions about air quality and carbon levels in the atmosphere.

Das’s research is just one of the ways we can continue to learn about how climate change is impacting our world, and the strides we can take to stop it.

“I get a lot of satisfaction in trying to answer questions related to the environment and the general health of the earth,” says Das. “I just love to get all these answers, which I feel are very important for us to know so we can act accordingly—or at least know what’s coming ahead of us.”

Read more:

The research team included Marcin Szubert and Josh Bongard from University of Vermont and Anuradha Kodali, Sangram Ganguly and Kamalika Das from NASA Ames Research Center.

Join the conversation

Do you have story ideas or something to contribute? Let us know!

Copyright © 2019 Science Node ™  |  Privacy Notice  |  Sitemap

Disclaimer: While Science Node ™ does its best to provide complete and up-to-date information, it does not warrant that the information is error-free and disclaims all liability with respect to results from the use of the information.

Republish

We encourage you to republish this article online and in print, it’s free under our creative commons attribution license, but please follow some simple guidelines:
  1. You have to credit our authors.
  2. You have to credit ScienceNode.org — where possible include our logo with a link back to the original article.
  3. You can simply run the first few lines of the article and then add: “Read the full article on ScienceNode.org” containing a link back to the original article.
  4. The easiest way to get the article on your site is to embed the code below.