New technology leverages state-of-the-art AI capabilities to model and map the natural environment

Mathematical Geosciences (2022) DOI: 10.1007 / s11004-021-09988-0 “width =” 800 “height =” 462 “/>

An overview of the deep neural network architecture visualized using NN-SVG software (LeNail 2019). For each observation, Input A feeds an image of the surrounding terrain into the stack of convolution layers (displayed as a horizontal block). At the same time, input B feeds the observation position variables to the fully connected layer. These two branches of the network are then concatenated and fed through two fully connected layers (shown as vertical blocks), from which the two parameters of the Gaussian distribution are output. credit: Mathematical earth science (2022). DOI: 10.1007 / s11004-021-09988-0

A team of experts, including Charlie Kirkwood from the University of Exeter, has created a sophisticated new approach to more detailed and accurate modeling of the Earth’s natural features.

The new approach recognizes complex features and aspects of the terrain that go far beyond the capabilities of traditional methods and can be used to generate high quality environmental maps.

Importantly, the new system can also pave the way for unleashing new discoveries of relationships within the natural environment. This may help address some of the larger climate and environmental issues of the 21st century.

The study is published in the journal Mathematical earth scienceAs part of a special issue on geostatistics and machine learning.

Modeling and mapping the environment is a time-consuming, time-consuming and costly process. The cost limits the number of observations that can be obtained. That is, creating a comprehensive spatially continuous map relies on bridging the gap between these observations.

Scientists use a variety of sources to generate terrain elevation data and Satellite image.. However, traditional modeling methods require users to manually design predictive features from these datasets. For example, generate tilt angles and curvatures from terrain. Elevation data Hopefully these will help explain the spatial distribution of the mapped variables.

However, scientists believe that within the natural environment, there are likely to be more subtle relationships that may be overlooked by models based on traditional manual feature engineering approaches.

The pioneering new AI approach developed in this research raises environmental information extraction as an optimization problem. By doing so, you can use traditional modeling techniques to automatically recognize and use relationships that may go unnoticed and unavailable to humans.

This approach not only improves the quality of the map, but also unleashes the potential for AI to discover new relationships in the natural environment, while eliminating extensive trial and error experiments in the modeling process.

Charlie Kirkwood, a graduate student at the University of Exeter, said: “To help make decisions, we need a model that is reliable and at the same time provides the most specific answer possible. That is, our estimate, in this case a prediction in an unmeasured location. ”

“Our AI approach is set within the Bayesian statistical framework, which quantifies these uncertainties and feeds them directly to confidence intervals, excess probabilities, and decision-making processes. It can provide a variety of uncertainty measurements, including, which is provided while leveraging the available information more effectively than traditional approaches. This can be seen in the map details. ”


Predictive logs (CaO) interpolated from geochemical observations of river sediments throughout the UK using auxiliary information provided by digital elevation models. This map shows the average predicted distribution of deep neural networks that captures the complex relationships between terrain features and log (CaO). credit:Mathematical earth science (2022). DOI: 10.1007 / s11004-021-09988-0

The new approach was demonstrated using river sediment calcium concentration observations from the British Geological Survey’s Environmental Geochemical Baseline Survey (G-BASE) project.

The distribution of calcium in the environment, which is solely important for its effect on soil fertility, is primarily controlled by geology. Different types of rocks contain different proportions of calcium, but they are also controlled by surface hydrological processes.

Therefore, calcium offers a challenging use case for the AI ​​approach. It is necessary to learn to recognize and utilize features related to both rock geology (eg, textures of different terrains, sloping breaks) and surface hydrology (eg drains, river canals). I have.

According to scientists, this method arguably reveals Britain’s geology at a new level of detail, thanks to the information-extracting power of the new AI approach, even though it depicts only one element, calcium. , Created a very detailed and accurate map. The team believes that the combination of research skills, expertise and data resources from its partners, the University of Exeter, the Met Office and the British Geological Survey, will bring a new dawn to the practice of environmental mapping in the age of AI.

Professor Gavin Shaddick of the University of Exeter said: He creates a new methodology that directly addresses key issues in mapping environmental information. The resulting methodological advances can be used to create detailed maps of various environmental hazards and may provide a wealth of sources for both scientists and decision makers. “”

Garry Baker, Interim Chief Digital Officer of the British Geological Survey, said: Environmental studies and how it can utilize the extensive dataset available to everyone through the National Geoscience Data Center and the wider NERC, and UKRI data repositories. ”

Dr. Kirstine Dale, Principal Fellow of Data Science at Met Office and Co-Director of the Center for Environmental Intelligence Excellence, commented on the value of this work: The important thing is to emphasize what can be achieved by working across disciplines. In this case, mathematicians, meteorologists, and computer scientists can come together to deepen their knowledge of nature in ways not possible in a single discipline. ”

Researchers develop artificial intelligence tools for environmental research

For more information:
Charlie Kirkwood et al, Bayesian Deep Learning for Spatial Interpolation in the Presence of Auxiliary Information, Mathematical earth science (2022). DOI: 10.1007 / s11004-021-09988-0

Quote: New technology utilizes state-of-the-art AI features, nature acquired from https: // on March 16, 2022 Model and map the environment (March 16, 2022). -capabilities.html

This document is subject to copyright. No part may be reproduced without written permission, except for fair transactions for personal investigation or research purposes. Content is provided for informational purposes only.

New technology leverages state-of-the-art AI capabilities to model and map the natural environment

Source link New technology leverages state-of-the-art AI capabilities to model and map the natural environment

Show More

Related Articles

Back to top button