Project Overview –GeoLLM.jpeg
A browser-based geospatial intelligence platform that combines interactive 3D terrain visualization, DEM-based geospatial processing, Earth Observation foundation models, deep learning-based terrain segmentation and Large Language Models (LLMs). The platform enables users to upload and explore DEM data, visualize terrain in 3D, perform terrain analysis, classify terrain characteristics using the Prithvi-EO-2.0 foundation model and interact with geospatial information through Qwen LLM-based conversational assistant.
Users upload elevation data, view it as a rotatable Three.js mesh with 15+ derived layers (slope, hillshade, TRI, hydrology, etc.), click for coordinates and DigiPin (India), explore flood/risk tools, and chat with an LLM whose answers are grounded in real terrain statistics and optional NASA Prithvi segmentation not generic chat.
Problem Statement
Traditional GIS tools (QGIS, ArcGIS) are powerful but:
  • Steep learning curve and complex desktop workflows
  • Hard for non-experts (students, planners, field teams) to get a quick read of landform
  • Statistics, 3D visualization, and natural-language Q&A are usually split across tools
  • Relating answers (“where’s the highest peak?”) back to a live map is often manual
There is a gap for a web-first system that combines DEM processing, 3D exploration, hydrology/risk views, and terrain-grounded AI in one loop: load → look → ask → see the answer on the map.
Proposed Solution
A web application that:
  1. Ingests DEMs (GeoTIFF, ASC, SRTM .hgt, etc.) via a FastAPI + GDAL backend with QGIS-aligned stats
  2. Renders a fast 512×512 preview mesh in the browser with React + Three.js
  3. Computes client-side derivative layers and server-side full-resolution terrain analysis
  4. Supports click-to-pin location (lat/lon, DigiPin, Nominatim area names)
  5. Adds flood/hydrology and risk-zone exploration
  6. Exposes a chat assistant constrained to on-screen terrain context (and optional Prithvi JSON / GTChain workflows)
  7. Caches processed DEMs (PostgreSQL + disk) and optionally hooks into QGIS via a plugin
Major Features
  • Interactive 3D terrain viewer with orbit, pan, zoom, and optional auto-rotate
  • Vertical exaggeration, multiple colormaps, wireframe overlay, and light/dark theme
  • DEM upload and processing for GeoTIFF, ASC, SRTM .hgt, and other GDAL rasters
  • Full-resolution elevation statistics (min, max, mean, median, stddev, p10, p90)
  • Fast 512×512 preview mesh for smooth browser rendering
  • 15+ terrain layers: elevation, hillshade, slope, aspect, contour, TRI, TPI, roughness, ridge/crest, valley depth, sinks, lineament, viewshed, TWI, flow accumulation
  • Backend GDAL terrain analysis for slope, aspect, TRI, and curvature (QGIS-aligned)
  • Click-to-pin with latitude, longitude, elevation, DigiPin (India), and area name
  • Flood and hydrology tools: water-level simulation, TWI, flow accumulation, risk zones
  • GeoLLM chat assistant for natural-language questions about the loaded terrain
  • AI-powered terrain feature extraction and classification using the Prithvi-EO-2.0 foundation model.
  • Adaptation of the Prithvi vision encoder for eight DEM-derived terrain channels: elevation, slope, sin(aspect), cos(aspect), TPI, TRI, roughness, and hillshade.
  • Extreme-point actions: pin highest peak, lowest point, or steepest slope from chat
  • Analytics panels with elevation/slope histograms and layer-aware statistics
  • Export options: PNG screenshot, CSV data, JSON stats report, printable HTML report
  • DEM caching with PostgreSQL for faster re-uploads
  • QGIS plugin to process the active raster and open the 3D viewer
 
Technologies and Tools
  • Frontend: React 18, Vite 5, Tailwind CSS 3
  • 3D Graphics: Three.js, React Three Fiber, Drei
  • Charts / UI: Recharts, react-dropzone
  • Backend: Python 3, FastAPI, Uvicorn
  • Geospatial Processing: GDAL, QGIS Python bindings, NumPy, rasterio
  • Database: PostgreSQL, SQLAlchemy
  • AI / Deep Learning: PyTorch, Prithvi-EO-2.0.
  • AI / LLM: OpenAI-compatible API (e.g. Ollama), Prithvi terrain segmentation
  • Location Services: OpenStreetMap Nominatim, DigiPin
  • Infrastructure: Docker, docker-compose
  • Desktop Integration: QGIS plugin
Practical Applications
  • Teaching DEM and terrain analysis concepts to students
  • Quick landform assessment for planners and researchers
  • Landslide and steep-slope risk screening
  • Flood and hydrology awareness using water-level and wetness layers
  • Field planning with DigiPin and reverse-geocoded location lookup
  • Generating terrain reports and screenshots for presentations
  • Bridging desktop GIS (QGIS) with a browser-based 3D viewer

Let's Build the Future With Us

Whether you are a researcher, startup, industry partner, government organisation, academic institution, or student, IITTNiF provides opportunities to innovate, collaborate, develop technologies, and create national impact.