Cantabria Powers Next-Gen Geospatial Insights With NVIDIA

The pressure to do more with less pushed Cantabria’s team to enhance their data analysis capabilities with AI-based computer vision. They turned to Esri ArcGIS Pro software powered by NVIDIA RTX™ 4000 Ada Generation GPUs.

Professional workstations with RTX Workstation GPUs “have been one of our best investments,” said Gabriel Ortiz, chief of the service of cartography and GIS.

Gabriel and his team use the CUDA-accelerated GeoAI functionality of Esri ArcGIS Pro, along with U-Net, MaskRCNN, adversarial networks and many other AI model architectures for segmentation, super-resolution, perceptual analysis, and other computer vision tasks. The team was able to accelerate AI fine-tuning and inference cycles without the need for data center or cloud infrastructure.

Esri’s GeoAI computer vision coupled with RTX Workstation enabled the Cantabria GIS team to extract unique details from landscape imagery, including tree covers, buildings, crops, and the expansion of forests over time. AI accelerates GIS calculations on raster data, extracting data that would have been otherwise inaccessible.

The integration of NVIDIA CUDA with Esri ArcGIS GeoAI has empowered Cantabria’s geospatial team to generate high-resolution cartographic displays with unprecedented speed and precision, seamlessly blending open-source AI models within the Esri environment. This allows for 1:5,000 scale and controlled segmentation across diverse terrain types—including tree cover, scrub, pasture, urban zones, and bare land—unlocking new levels of detail in geospatial analysis.

Building on this momentum, Cantabria’s analysts have established advanced workflows to restore degraded aerial survey imagery from the 1950s. By harnessing CUDA-accelerated NVIDIA hardware, they apply U-Net transformation modules in combination with adversarial networks and the concept of perceptual loss and then evaluate the results using the VGG-16 AI model. This approach enables the transformation of aging grayscale imagery into vibrant, full-color restorations that are both visually convincing and preserve essential pixel-level geospatial data, ensuring historical records become a rich, usable resource for modern land management and planning.

As an example, Ortiz and his team aimed to improve five-year-old imagery that suffered from poor color balance, heavy shadows, and other limitations. Leveraging Esri’s raster analytics pipeline with NVIDIA GPUs, they rapidly iterated model versions and validated enhancement quality on large image blocks with a perceptual loss feedback loop. This workflow produced enhanced, visually convincing geospatial imagery that maintains critical macro features such as building edges, canopy boundaries, and transport corridors.

Tasks that previously required over a year and cost millions of taxpayer euros, often involving outsourcing, can now be completed by Ortiz’s small team in a few months. For example, a 1:5,000-scale mapping operation, such as building extraction, which once depended on extensive stereo plotting and manual quality assurance and control by dozens of professionals, was recently accomplished by a single data scientist using an NVIDIA-powered workstation in just three days.

Ortiz’s team has also applied these technologies for the early detection and tracking of ecological threats, most notably the invasive species Cortaderia selloana (Pampa grass), which threatens Cantabria’s local biodiversity. Applying their Esri GeoAI techniques powered by NVIDIA GPUs, the crew produced full-resolution maps, providing real-time insight into the species’ spread and land claims.

Similarly, their work classifying crop fields has enabled authorities to generate detailed fuel models, supporting more effective wildfire prevention and response.