deep learning

Deep Learning And Dirty Air | Earth Wise

May 6, 2021 By EarthWise Leave a Comment

Poor air quality is a major global problem. According to the World Health Organization, exposure to air pollution is linked to the premature deaths of an estimated seven million people every year. In fact, 9 out of 10 people breathe air that contains more pollutants than what the WHO considers safe. Air pollution is the fourth largest threat to human health, trailing only high blood pressure, dietary risks, and smoking.

But predicting pollution levels at a given place and time remains challenging. According to a new study recently published in the journal Science of the Total Environment, scientists are turning to deep learning to improve air quality estimates.

According to researchers, satellite observations and ground observations both measure air pollution, but both have major limitations. For example, satellites may collect data at the same time and at the same location each day, but they miss how emissions may vary throughout the day. Ground-based observations from weather stations do continuously collect data, but they only do so in a limited number of locations.

As a result, scientists have turned to deep learning – a type of machine learning – to analyze the relationship between satellite and ground-based observations of nitrogen dioxide around Los Angeles. Nitrogen dioxide is associated with emissions from traffic and power plants. The researchers were able to rely on the learned relationship to take daily satellite observations and create hourly estimates of atmospheric nitrogen dioxide levels in approximately three mile grids.

According to the research team, this study could be repeated for other greenhouse gases, and applied to different cities and regions – or even whole continents.

**********

Web Links

Scientists turn to deep learning to improve air quality forecasts

Air Pollution

Photo, posted November 4, 2019, courtesy of Ninara via Flickr.

Earth Wise is a production of WAMC Northeast Public Radio.

Tracking Endangered Species From Space | Earth Wise

March 4, 2021 By EarthWise Leave a Comment

Scientists at the University of Bath and the University of Oxford in the UK have developed a technique for remotely surveying elephants and other wildlife using satellite images and deep learning. The technique has the same accuracy as human counts done on the ground or from low-flying airplanes.

The new computer algorithm can analyze high-resolution satellite images and detect African elephants in both grasslands and forests. Previous techniques for monitoring wildlife from space were limited to homogenous habitats, such as the case of tracking whales in the open ocean.

On-the-ground or airplane surveys to monitor animal numbers are expensive and time-consuming. Satellites can collect nearly 2,000 square miles of imagery in a few minutes thereby eliminating the risk of double counting and reducing a process that previously took weeks to just a few days. The use of satellites also eliminates the logistical problems of monitoring species populations that cross international borders.

Accurate monitoring is important for efforts to save endangered species. There are only 40,000 – 50,000 African elephants left in the wild. It is essential to know where the animals are and how many there are in various locations. The new method is able to count elephants in mixed ecosystems, such as savannah and forests, where tree cover would previously have made satellite tracking difficult.

With satellite imagery resolution increasing every few years, it should be possible to see ever-smaller things in greater detail. The new algorithm works well for elephants; it may eventually become practical to track animals as small as an albatross from space.

**********

Web Links

A New Way to Track Endangered Wildlife Populations from Space

Photo, posted March 15, 2008, courtesy of Michelle Gadd/USFWS via Flickr.

Earth Wise is a production of WAMC Northeast Public Radio.

A Breakthrough In Animal Identification

December 25, 2018 By EarthWise Leave a Comment

Researchers from the University of Wyoming have developed a computer model that can identify wild animals in camera-trap photographs with remarkable accuracy and efficiency.

This breakthrough in artificial intelligence (AI), detailed in a paper recently published in the scientific journal Methods in Ecology and Evolution, represents a significant advancement in the study and conservation of wildlife. According to the paper’s authors, “the ability to rapidly identify millions of images from camera traps can fundamentally change the way ecologists design and implement wildlife studies.”

This study builds on previous research from the university in which a computer model analyzed 3.2 million images captured by camera traps in Africa. The A-I technique called deep learning categorized animal images at a 96.6% accuracy rate. This was the same accuracy rate as teams of human volunteers achieved, but the computer model worked at a much more rapid pace.

In the latest study, UW researchers trained a deep neural network on a powerful computer cluster to classify wildlife species using 3.37 million camera-trap images of 27 different animal species. The model was tested on nearly 375,000 images at a rate of about 2,000 images per minute. It achieved a 97.6% accuracy rate, which is likely the highest accuracy to date in using machine learning for wildlife image classification.

Artificial intelligence has been used in environmental science in other ways as well. For example, AI has been used to increase agricultural yields in farm fields and to help predict extreme weather.

Maybe artificial intelligence can prove to be a game changer for the environment.

**********

Web Links

Researchers Successfully Train Computers to Identify Animals in Photos

Photo, posted January 8, 2012, courtesy of the U.S. Fish and Wildlife Service via Flickr.

Earth Wise is a production of WAMC Northeast Public Radio.