Digital Surface Modeling: How AI Converts 2D Images into 3D Insights

Digital surface models (DSMs) are nothing new, but as software and scanning equipment have become more sophisticated so have DSM’s. Digital terrain maps have many engineering applications, of course. but other sectors, such as agriculture, energy, and conservation, are finding uses for them as well.

When you add AI to the mix, however, the capabilities take a gigantic leap.

Though the technology is still evolving, models and digital twins that integrate artificial neural networks and digital surface data promise to write a brand-new chapter in geospatial engineering.

An artificial neural network is what it sounds like. It’s based on human learning and mimics the biological neurons in our brains. These networks must be trained to process data similar to the complex way we think. This type of machine learning is also known as “deep learning.”

Neural networks can be trained by using a variety of classic landform patterns and varied terrain from all over the globe. Satellite and aerial images from drones or other aircraft can be input, and the more the model absorbs the better it will be at getting the most out of the data it receives in the future.

Filling in the Gaps

Creating digital elevation maps (DEMs) has historically been a time-consuming, expensive, and complex process. Even using the most advanced LiDAR scanners and high-resolution photogrammetry, data voids are inevitable in rough terrain, particularly in peaks and valleys. This affects the quality and accuracy of the resulting maps and models created from the data.

This is one of the areas where deep learning techniques help neural networks fill in the gaps. “Big data” exists for mountainous terrain all over the world in the form of satellite images that can be used as samples for training neural networks to understand different types of terrain.

Models can then recognize patterns of topographical features in complex data sets and use the information to help fill in data voids with estimated height calculations. The trick is to train the network to reflect trends in local terrain in order to fill the data void and match up seamlessly. This is one of the areas where work still needs to be done in improving the technology, but so far results have been promising.

Never before possible, an AI model can even turn a single 2D image into a 3D Digital Terrain Map using monocular depth estimation. In monocular depth estimation, AI predicts the depth between the object and the camera for each pixel. This requires the model to learn and comprehend relationships between objects in the image and corresponding depth information. Adding to the complexity are factors such as shadows and the direction of the sun.

Being able to incorporate AI for the derivation of elevation data from orthophotos is a huge step toward making DEM models more cost-effective and accessible.

Uses for Digital Surface Models

The advent of integrating AI and DEMs is exciting news for engineers, geologists, cartographers, hydrologists, and more. Applying AI to digital surface mapping increases its applications for:

• Soil erosion mapping
• Disaster risk assessment
• Watershed management
• Infrastructure planning
• Urban development
• Natural resource management
• Environmental impact assessment
• Landform classification

Models can be used to simulate obvious events like flash flooding and landslides, but they can be trained to do far more nuanced and complex tasks like predict near-surface soil moisture or forecast how much solar irradiation occurs in a region of interest for developing solar photovoltaic facilities.

The ability of these models to help solve problems and make plans, to monitor change, and predict outcomes multiples their use infinitely.

Darling Geomatics Can Put AI to Work for You

Darling Geomatics has long been a leader in geospatial engineering. Their ability to apply cutting edge technology to your project can save time and money and boost efficiency. Find out how much intelligent 3D digital surface models can improve your workflow.