Protecting Rainforests through Machine Learning

2019-12-08

Computer scientist David Dao has developed an intelligent algorithm that uses satellite images of rainforests and unmanned aerial vehicle (UAV) images to predict the next potential area of rainforest destruction. He will present his research at the upcoming climate conference in Madrid and launch a pilot project in Chile in January.


Dao, who is from Germany, is an expert in the field of machine learning and has developed an intelligent algorithm that can automatically analyze satellite and UAV images. This helps to reveal where forest cover is declining and the extent of the decline. They can even predict where rainforests will disappear in the near future, and the trick lies in how the algorithm reads the images.


Satellite and UAVs provide countless images of tropical rainforests at different heights, resolutions, and qualities. The commonality among them is that the areas they depict are not labeled or otherwise identified. Unlike maps, these places do not have easily recognizable labels such as names, forests, rivers, and roads that computer scientists would say. This means that computer algorithms cannot recognize what is forest cover and what is not forest cover.



Forest area is shrinking


As Dao explains, the algorithm reads sequences to identify forest areas and whether these areas are shrinking. These sequences are individual images strung together in chronological order, much like old film reels or comics.


From an aerial perspective, the resulting pattern resembles a fish skeleton, with a spine and small bones, hence the name "fishbone". By comparing these aerial images arranged in chronological order, the algorithm can determine how road systems and forest cover change over time.


This means that the intelligent algorithm does not need labels to generate an overall image indicating that the rainforest is shrinking. They can also predict where the most severe deforestation will occur next. The model is also applicable to deforestation near rivers and agricultural areas.




Trial run in Chilean rainforest


In a research project called Komorebi, David Dao has enlisted partners in the field, including the Chilean National Forestry Corporation (CONAF). In early 2020, a pilot project will begin in the Valdivian rainforest on the Pacific coast south of the capital, Santiago. Dao will test and refine his predictive algorithm in real rainforest conditions. His method not only detects overall decline in the rainforest but also identifies the most severely affected tree species. This is an important factor in climate change, as different types of trees store carbon dioxide at different rates, and one way to protect forests is to provide financial incentives to local residents to preserve trees as a carbon storage method instead of deforestation.


In the Chilean rainforest, they will study how to improve the accuracy of the predictive algorithm by combining satellite images with images captured by low-altitude UAVs. Unlike satellite images, UAV images can be accurate to within 30 centimeters. Dao says, "If there are UAV images, we can also observe changes in trees and discover changes in biodiversity."

RECENT AI NEWS

RECENT AI TOOLS