[Scientists use machine learning to predict diversity of tree species in forests]
A collaborative team of researchers led by Ben Weinstein of the University of Florida, Oregon, US, utilized machine learning techniques to create detailed maps of over 100 million trees from 24 different sites across the United States. Their study, published in PLOS Biology on July 16, highlights the significance of these maps in enhancing conservation efforts and ecological projects.
In the past, ecologists have relied on manual surveys of small land plots to understand forest ecosystems. However, this approach lacks the ability to capture the true variability of entire forests. Other methods, while covering larger areas, struggle with accurate categorization of individual trees.
To address this challenge, researchers trained a deep neural network using images of the tree canopy and other sensor data captured from planes. This training set included data from 40,000 individual trees sourced from the National Ecological Observatory Network.
The deep neural network demonstrated an impressive classification accuracy of 75 to 85% for common tree species. Furthermore, the algorithm could identify vital information such as whether a tree is alive or dead.
Interestingly, the network performed exceptionally well in areas with less dense tree canopy, excelling in categorizing conifer species like pines, cedars, and redwoods. It also showed greater accuracy in regions with lower species diversity.
Researchers have made their model predictions available on Google Earth Engine to support further ecological research endeavors. By sharing their findings, they aim to contribute to a deeper understanding of forest ecology and ecosystem dynamics.
Our objective is to offer researchers comprehensive maps of tree species diversity in various ecosystems across the US. These canopy tree maps can be regularly updated with new data, enabling continuous improvement in predictions, the authors stated.
This groundbreaking research showcases the profound potential of machine learning algorithms in revolutionizing forest monitoring and conservation practices. By harnessing the power of technology, scientists are paving the way for more effective and sustainable resource management strategies.
