EAGLE - Earth Intelligence Studio

Welcome to the Destination Earth EAGLE platform documentation. EAGLE is the Earth Intelligence Studio, a visual geospatial workflow platform for satellite imagery analysis and AI-powered Earth observation.

Figure 1: The EAGLE public landing page on the Destination Earth platform.

Overview

EAGLE provides a no-code, drag-and-drop canvas for building, executing, and monitoring AI-powered processing pipelines on Earth observation data. Users design workflows by connecting nodes on an interactive canvas, running inference, fine-tuning models, and exploring results - all without writing code.

Figure 2: The empty EAGLE workflow canvas with the Node Palette on the left and quick-start prompts in the centre.

Figure 3: The Node Palette grouped by category (Data Sources, Preprocessing, AI Models, Inference, Tools, Outputs).

Core Platform Capabilities

EAGLE is built around a visual workflow paradigm:

• Visual Workflow Builder - Design geospatial processing pipelines by dragging, connecting, and configuring nodes on an interactive canvas powered by React Flow.

• AI Model Inference - Run pretrained models for flood/water detection, burn scar detection, crop classification, solar farm and rooftop solar panel detection, change detection, cloud and shadow detection, field boundary delineation, and text-prompted segmentation with SAM3.

• Model Fine-tuning - Fine-tune flood/water, burn scar, crop classification, and Satlas solar farm models on your own labelled data using one of four foundation-model backbones.

• Satellite Data Access - Integrated Copernicus Data Space access for Sentinel-2 search, download, and processing. GeoTIFF, Zarr, and NetCDF upload are supported.

• Interactive Segmentation (SAM3) - Segment Anything Model 3 integration for text-prompted geospatial segmentation.

• Spectral Analysis - Interactive per-pixel and grid-based spectral profile extraction with reference spectra, spectral indices (NDVI, NDWI, SAVI, EVI, NDBI, BSI), and statistics.

• Zonal Statistics - Polygon-based aggregate statistics (17+ measures including percentiles, variance, and frequency metrics) over raster imagery with choropleth visualization.

• Super Resolution - Upscale Sentinel-2 10 m imagery to 2.5 m using the OpenSR latent diffusion model.

• Explainable AI (GradCAM) - Visual heatmap overlays on all segmentation models showing which regions drove each prediction.

• LLM Workflow Assistant - AI advisor powered by Mistral for natural-language workflow generation and error recovery (“Fix with AI”).

• Model Registry - Versioned lifecycle management of fine-tuned models with promotion, comparison, rollback, and tagging.

• Workflow Sharing - Public read-only share links with configurable expiry for collaborating on workflows and outputs.

• Result Visualization - Interactive map previews, GeoTIFF tile serving, timeline playback, and downloadable GeoJSON / GeoTIFF outputs.

Foundation Model Backbones

Core segmentation tasks (flood/water, burn scar, crop classification) can be run on four backbone architectures:

• Prithvi EO v2 (IBM / NASA) - 600 M-parameter Vision Transformer pretrained on 4.2 M global Sentinel-2 / Landsat time-series samples. Ships with pretrained task checkpoints for immediate inference.

• TerraMind v1 (IBM / ESA) - Any-to-any multi-modal foundation model using all 12 Sentinel-2 L2A bands. Requires fine-tuning before use.

• DOFA Large (Dynamic One-For-All) - Wavelength-adaptive foundation model that works with any sensor via sensor presets (Sentinel-2, Sentinel-2 Full, Landsat-8/9, RGB). Requires fine-tuning before use.

• Satlas Swin-V2-B (Allen AI) - Swin Transformer V2-B with Feature Pyramid Network, specialised for infrastructure and environmental mapping on Sentinel-2. Ships with a pretrained solar-farm checkpoint.

Available AI Models

Inference:

• Flood / Water Detection - Prithvi, TerraMind, or DOFA backbone. Prithvi ships with a pretrained checkpoint.

• Burn Scar Detection - Prithvi, TerraMind, or DOFA backbone. Prithvi ships with a pretrained checkpoint.

• Crop Classification - Multitemporal (3 timesteps) with Prithvi, TerraMind, or DOFA backbone. Prithvi ships with a pretrained checkpoint.

• Solar Farm Detection - Pretrained Satlas Swin-V2-B for utility-scale solar farms on Sentinel-2 imagery.

• Solar Panel Detection - Mask R-CNN instance segmentation (geoai-py) for rooftop and ground-mounted panels on VHR / aerial RGB imagery.

• Field Boundary Detection - YOLO11 (Delineate Anything) for agricultural parcel delineation, with GeoJSON vector output.

• Cloud Detection - OmniCloudMask 4-class segmentation (clear / thick cloud / thin cloud / shadow) for 10-50 m imagery.

• Change Detection - SAM-based bi-temporal comparison between two co-registered images.

• SAM3 Text Segmentation - Zero-shot segmentation from natural-language prompts using Segment Anything Model 3.

Fine-tuning (Training):

• Flood / Water, Burn Scar, and Crop Classification training pipelines with Prithvi, TerraMind, or DOFA backbones.

• Satlas Solar Farm fine-tuning for domain adaptation on custom labelled data.

Project-Based Workflow

All work within EAGLE is organized through a hierarchy:

• Project - A user-scoped container for workflows, data, and results for a specific initiative.

• Workflow - A visual pipeline of connected nodes saved within a project.

• Job - A single execution of a workflow, tracked with per-node status and GPU-aware queue management, recoverable after pod restarts.

Figure 4: The “Welcome to Eagle” modal prompting the user to select an existing project, create a new one, or open the demo.

Figure 5: The Create Project dialog before any details are entered.

Figure 6: The Create Project dialog with a name and description filled in.

Documentation Sections

Contents:

• Eagle Service User Manual
  • 1. Introduction to the Eagle Service
    • 1.1. Service Overview: The Earth Intelligence Studio
    • 1.2. Core Platform Capabilities
    • 1.3. Understanding the Project-Based Workflow
  • 2. Getting Started
    • 2.1. Authentication
    • 2.2. Creating Your First Project
    • 2.3. The Canvas Interface
    • 2.4. Keyboard Shortcuts
  • 3. Building Workflows
    • 3.1. Adding Nodes to the Canvas
    • 3.2. Node Categories
    • 3.3. Connecting Nodes
    • 3.4. Configuring Nodes
    • 3.5. Workflow Templates
  • 4. Node Reference
    • 4.1. Data Sources
    • 4.2. Preprocessing
    • 4.3. Training (AI Models)
    • 4.4. Inference
    • 4.5. Outputs
  • 5. Foundation Models
    • 5.1. Overview
    • 5.2. Backbone Architectures
    • 5.3. Pre-trained Model Training Data
    • 5.4. Choosing the Right Model
  • 6. Band Requirements
    • 6.1. Prithvi (6 bands)
    • 6.2. Satlas (9 bands, Solar Farm)
    • 6.3. DOFA Sensor Presets
    • 6.4. Multitemporal Crop Classification
  • 7. Executing and Monitoring Jobs
    • 7.1. Submitting a Workflow
    • 7.2. The Operations Log
    • 7.3. Job Queue Management
  • 8. Working with Results
    • 8.1. Previewing Results
    • 8.2. Downloading Results
    • 8.3. File Browser
  • 9. Advanced Features
    • 9.1. Zonal Statistics
    • 9.2. Spectral Analysis
    • 9.3. AI Workflow Assistant
    • 9.4. Model Registry
    • 9.5. Sharing Workflows
    • 9.6. Interactive Tour
  • 10. API Reference
  • 11. Support and Contact
• Support
  • Contact Us
    • General Inquiries
    • Operational Support
  • Response Times
  • What to Include
  • Additional Resources

Quick Start Guide

To start using EAGLE:

1. Sign In - Authenticate via the Destination Earth Keycloak SSO.

2. Create a Project - Use the project selector in the header to create a new project.

3. Try a Demo - Open the Help Guide (F1 or ?) and pick Flood Detection Demo (Emilia-Romagna 2023) from the Workflow Templates gallery. Choose an output folder on the Sentinel-2 node and click Run.

4. Build Your Own - Drag nodes from the sidebar onto the canvas, connect them by dragging from output to input handles, and configure each node’s parameters.

5. Submit Job - Click Run in the header to execute the workflow and monitor progress in the Operations Log.

6. View Results - Explore outputs in the preview panel (maps, charts, GradCAM heatmaps) and download GeoTIFF / GeoJSON exports.

Figure 7: The interactive onboarding tour, accessible from the “Tour” button in the top-right corner of the canvas.

Figure 8: A configured Sentinel-2 Data Source node showing date range, cloud-cover threshold, band selection and resolution.

Figure 9: Close-up of the Sentinel-2 node header indicating that a region and inputs have been selected.

Figure 10: An example end-to-end change-detection workflow connecting two Raster Inputs, Band Selectors and a Change Detection inference node.

Figure 11: The change-detection results panel offering Comparison, Enhanced, Instances and Report previews together with download links.

Figure 12: A side-by-side split comparison highlighting high-confidence changes between the earlier and later Sentinel-2 acquisitions.

License

Copyright © 2026 Destination Earth. All rights reserved.

Indices and tables

• Index

• Module Index

• Search Page