The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Back to feedPost
Unpopular opinion: 90% of the complex Machine Learning models built for satellite remote sensing are completely unnecessary.
We see fancy Deep Learning architectures deployed just to map basic surface water, vegetation stress, or land cover—when a well-calibrated physical index (NDVI, NDWI) combined with simple thresholding or physical terrain constraints (like HAND) delivers:
🔹 10x faster execution
🔹 100% explainable physics
🔹 Zero black-box hallucination
High resolution and neural networks look impressive in a portfolio, but if your model ignores terrain elevation or physical domain mechanics, it’s just over-engineered noise.
Sometimes the best GIS analysis is the simplest one rooted in physics. 🌍
Are we relying too much on AI for spatial problems that simple remote sensing physics already solved? Let’s discuss. 💬👇
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Related posts
Check it out
New case study live!! 🚀
Forsite brings together aerial LiDAR, pipeline patrol, wildfire science, and professional forestry under one roof, giving operators across North America a single, credentialed partner for the complete intelligence picture their decisions depend on.
How do you make complex real estate offer comparisons effortless?
A wall I walked past turned into my latest experiment with photogrammetry.
I took a quick, rough video scan while out on a walk, mostly to see what I could get out of it. What impressed me was how little time it took to get something usable (some cleanup would be necessary for a production shot).
Still experimenting, but there’s something exciting about turning an everyday surface into a 3D asset without making a whole project out of it.
This is wild — from a rough walk scan to that usable mesh so fast! The moss texture retention is crazy realistic.
What did you use for the scan? Polycam/RealityScan + Blender cleanup? I've been thinking about using photogrammetry for real-world clinic interiors to build more...
