The engineering problem
A municipal planning authority needs to identify unauthorized construction activity across a 200 km² urban area over the past five years. A mining company needs quarterly vegetation health reports from multispectral satellite data. A defence intelligence unit needs terrain analysis for a 50 km route corridor — classified, air-gapped, no external connectivity.
Each scenario requires the same fundamental pipeline: acquire imagery, process it through domain-specific algorithms, extract intelligence, and deliver verified results that can support decisions with legal or operational consequences.
Why traditional approaches are insufficient
Processing scale. A single Sentinel-2 scene is 1 GB. A five-year time series for a metro area is hundreds of scenes. Desktop GIS cannot process this volume without manual batching, and manual batching introduces errors and gaps.
Tool coordination. Change detection requires image alignment, spectral normalization, AI segmentation, spatial statistics, and cartographic output — five distinct tool domains that typically require five separate software environments configured by different specialists.
No audit trail. When a planning decision is challenged in court, can you prove which satellite images, which algorithm version, and which parameters produced the change detection map? Manual GIS workflows provide no inherent provenance.
How Studio executes geospatial intelligence
Intent: "Detect building-level construction changes in Guntur, Andhra Pradesh from 2017 to 2022. Output: interactive dashboard with change overlay, statistical heatmaps, and downloadable GIS datasets."
Studio decomposes this into an execution plan:
1. Imagery acquisition → Sentinel-2 time series (2017-2022), cloud-filtered
2. Preprocessing → atmospheric correction, co-registration, temporal alignment
3. Change detection → spectral difference analysis + AI building segmentation
4. Hotspot identification → spatial clustering of change areas
5. Statistical analysis → ward-level aggregation, growth rate computation
6. Visualization → interactive map dashboard with layer controls
7. Export → GeoTIFF, shapefiles, PDF report
8. Verify → hash all outputs, record full execution lineage
The Execution Fabric orchestrates:
- GPU provisioning for AI segmentation models (deep learning inference)
- Sequential dependency management (preprocessing must precede analysis)
- Large dataset transfer (hundreds of GB of imagery)
- Checkpoint and resume for multi-day processing campaigns
The Spatial Intelligence Workbench provides:
- QGIS 3.x + PostGIS for full-stack geospatial analysis with spatial SQL
- AI-powered change detection and land classification models
- 50+ GeoAI capabilities via the GeoAI & Spatial Capability Pack
- GPU-accelerated deep learning segmentation and object extraction
The GeoAI & Spatial Capability Pack includes:
- Change Detection — building footprint and land-use change from multi-temporal imagery
- Terrain Analysis — slope, aspect, viewshed, and route optimization
- Object Extraction — automated feature identification from aerial and satellite imagery
- Land Classification — supervised and unsupervised spectral classification
- Route Optimization — network analysis with terrain and accessibility constraints
- Spatial Analytics — hotspot identification, cluster analysis, spatial statistics
Verified output: Every GeoTIFF, every shapefile, every statistical table receives SHA-256 hashing and full execution lineage. The dashboard itself is an artifact — its data provenance traces back through every processing step to the original satellite scenes. The execution receipt provides a traceable record of how the analytical output was produced.
Live evidence
The Guntur Change Detection Dashboard is the public demonstration of this exact pipeline. It processes real Sentinel-2 data through the Spatial Intelligence Workbench and delivers an interactive map with building-level change detection, ward-wise statistics, and heatmap overlays. It is open source, browser-native, and fully reproducible.
This is not a mockup or wireframe. It is a working output of the execution approach described here.
Deployment relevance
Geospatial intelligence work frequently involves classified data (defence), commercially sensitive data (mining, agriculture), or sovereignty-restricted data (government). The same GeoAI Capability Pack is designed to run in:
- Cloud — for open-source satellite data and commercial projects
- Sovereign — for government agencies requiring data to remain within national boundaries
- Air-Gapped — for defence and intelligence applications with zero internet connectivity
The Portable Appliance deployment mode supports field teams processing drone survey data in areas without connectivity — the same execution architecture on transportable, encrypted hardware.
This is Part 4 of the How Nebula Cloud Executes Work series.
Previous: Part 3 — Executing Engineering Design Next: Part 5 — Executing ML Workflows · Part 6 — Executing Scientific Computing
Related: Spatial Intelligence Workbench · GeoAI Capabilities · Execution Fabric · Demo Catalogue