- Drifft Newsletter
- Posts
- Applied AI and mapping technology in resource markets
Applied AI and mapping technology in resource markets
Where startups and researchers are finding success and where you should be looking
David here. In the next 7 min you’ll get:
💡What’s New: A massive update on AI and resource mapping
🤔 Opinion: Why industrial AI takes longer and how to start
🛠️ Tools & Data: Get a handle on geo data, analytics, and earth observation.
Want to feature your service or product in DRIFFT? Grab an ad spot here.
💡What’s new
AI in resources
AI is clearly making headway in land and resource characterization as well as spectral and geochemical analysis. More studies show how mineral prospecting with data science and monitoring mine expansion are possible. AMZN has jumped into earth science with its release of the Earthformer model for time-series predictions (github repo below). And IBM partnered with NASA on a satellite imagery foundation model. The Metals Comany and Kongsberg Digital are testing AI in undersea mining. Kobold’s story should be useful to VCs backing physical resource projects (see a short case study, then an in-depth look at how they broke in to mining, and finish with an interview with the CEO).
There are several open questions about AI’s future. Should it be opensource (see Senate testimony)? And should the focus be large multimodal models or small use-case models for IoT? Both are happening. AMZN thinks its important and paid $4B for Anthropic. And it appears Meta and Bing have formed an AI search partnership. When it comes to automating manual work, the elephant in the room is Google’s latest forray into robots. In all the hype, I find it’s a balanced view that is best.
“Take a large multimodal model. Fine-tune it on robot actions. That's it! You now have a robot that understands and manipulates visual and semantic concepts it has never experienced before.”
Mapping, Data, and Navigation
First there were dozens of free satellite data providers (more here). Now comes opensource with MAXAR. GlobeHe for drone images has a free tier too. Remarkable that the only cost for starting an earth observation practice is labor!
All this imagery is enabling new studies. India is using satellites to characterize minerals. Australia released a mapping application to interact with its national core sample library. You’ll also find data for AUS, CA, and US at the USGS (and don’t miss their science data catalogue). From my vantage, it seems that opensource is capable of providing all necessary solutions for satellite image analytics. But this doesn’t solve the problem of trusthworthy data. Platforms are emerging for centarlizing both data and analytics, like this latest Stanford startup for underground exploration. Along with these challenges comes the need for UIs to aid decision making.
Drones and LiDAR are finding uses above and below ground. DJI recently debuted a new payload drone at $17k. There’s even a hand-held lidar if you have legal hurdles. Baidu thinks it’s worth working on autonomous mapping. As lidar solutions mature, providers will pop up overseas and costs will drop.
In every area technologies are merging to give us improved versions of previous concepts like radar. Robots in remote locations will soon have better IMUs that don’t require GPS (a longer introduction here). While we wait for the science to mature, existing tech gets more accurate and cheaper using consumer-grade sensors with better algorithms. It’s the fusion of data and technologies in interesting ways that defines state-of-the-art.
Resources and tech
There’s a surprising amout of useful mining info on Wikipedia. This introduction to the mining cycle is a look at how new startups fit in.

a compressed view of the mining lifecycle
There’s more to resources tech than just exploration and recovery. Some questions / ideas startups can tackle include:
What is the best use of the land AFTER mining?
Could the combination of new tech and old mine tailings bring economic resources in the previous generation’s waste?
If injecting water into the ground to frac oil is bad, what does injecting CO2 do?
Then there’s water utilization as well as fleet management.
In 2018, BCG was early to call out tech’s potential for impact in resource markets. Then in 2021, they forecasted the range of productivity increases possible. Last year, they highlighted Canada’s opportunity due to that region’s specialization in mining (note: Canada became Kobold’s prooving ground).
In other news:
Business models of imagery data providers could be disrupted by superresolution techniques.
AI encoders compress images better than PNG, so could it help reduce data storage costs?
Will the next generation of batteries be absorbing nuclear power?
China not only leads in mining and mineral exports, but also in patents (>4k this year).
Finance
Here’s a video to watch if you feel lucky but know nothing about mining. And you should know what makes rare earth minerals rare. Hint: manganese may be emerging as an economic concern (DYOR).
Western states (especially NV) stand to gain if the battery metals run continues and commodities prices stay high. It’s possible to get in to resources simply buying land and mineral options, but it’s worth considering the various ways traders fail in options markets in addition to understanding mineral rights risks. Personally, I am interested in logistics. Regardless of where there’s a run on minerals, they’ve all got to be shipped! Also, startups take note–safety is a god-term that carries a strong funding mandate.
This marketing piece seems to argue that tech-generated effeciencies can help stabalize mineral prices. I leave it to you to decide if that’s a rational conclusion given the volatility introduced by a global battery electric rush. Keeping with that theme of marketing, I’ve realized that geo tech marketing copy sells to professionals instead of the public (IMO, dry and technical). But imagine the deals you might do if your tech startup could help companies market like this.
Finally, don’t miss this review of private equity activity in earth observation (with a recent deal to benchmark).
Jobs and training
Physical resources industries are becoming an interesting niche for software, AI, and robotics specialists. There’s a lack of STEM talent in the US with ongoing efforts to address shortfalls.
You can start with an intro book to GIS mapping and this introduction to geospatial analytics (with class notes). Google has a great reference for algorithms. You can even see them as a mindmap. If you like UIs, it’s helpful to review how mapping is done and to review algorithms to generate a world map as well as map tiles. Note: these skills are transferable to augmented reality. There’s also this youtube web mapping playlist. Many online courses can guide you (like python for geospatial, or an under-the-hood tutorial of google earth engine, as well as this NASA course on using space imagery). For those who love to work on databases, learn how to use Sedona (github below), Apache’s geospatial data engine. Bonus: use MSFT’s planetary computer data in a Jupyter notebook.
For those wanting to learn AI, consider a free course at DeepLearningAI. An up-and-coming area is porting large AI models to small IoT devices and MIT has a free lecture series to help. If you want to go all in, the next generation of mapping algorithms will run on neuromorphic hardware.
The age of working for a single company for years may be past. Even Upwork has a dial-a-geologist section if you want to take side gigs. Here are a set of interview questions for aspiring freelance geologists.
Skill up! Opportunities are bright and increasing every day in this industry.
🤔 Opinion
AI is overtaking consumer digital markets with such speed that it seems worthwhile to stress patience for resource business leaders.
Simply put: Industrial AI is not fast to develop, not cheap, and not easy. Payoff periods will be longer.
From my experience in automotive, things slow down when AI meets the real world. One reason is that AI systems observing the physical world encounter vagueries they don’t typically see in the digital world. Those become bugs to work out in your system. Another reason is that software and pure engineering cultures think differently. “Fail fast” isn’t a career-building idea when failure shuts down processing and costs millions of dollars.
These are the environments we encounter in resource industries.
If you’re deciding on an AI project at your company, you can expect the challenges and costs to multiply. Start small is my simplest advice. Be thinking in terms of 3 to 5yrs for initial development.
If you’re a CTO considering the best path forward, cast your eye on operations (billing, customer and vendor management, compliance, security, etc). These are low-hanging fruits for AI to contribute to profits.
🛠️ Tools and Data
Ames Stereo Pipeline: https://github.com/NeoGeographyToolkit/StereoPipeline NASA Ames Stereo Pipeline (ASP) suite of open source automated geodesy and stereogrammetry tools for processing stereo images captured from satellites, rovers, aerial cameras, and historical images, with and without accurate camera pose information. | ![]() |
Opengeos Github: https://github.com/opengeos A collection of gis tools for web-based gis and deep learning. | ![]() |
Open HSI: https://github.com/openhsi/openhsi Opensource hyperspectral imagery analytics package for evaluating satellite imagery. | ![]() |
Chemo Tools: https://github.com/paucablop/chemotools A collection of preprocessing tools and utilities for working with spectral data as well as developing machine learning models for predicting properties or classifying samples. Use after OpenHSI. | ![]() |
Chrieke’s Github: https://github.com/chrieke Fantastic set of repos on all things satellites, earth observation, and GIS. |
Earthformer: https://github.com/opengeos/earthformer a Python package for the Amazon space-time Transformer for Earth system forecasting. | ![]() |
Apache Sedona: https://github.com/apache/sedona a spatial computing engine to process spatial data at scale within modern cluster computing systems. | ![]() |
Special Mention: https://anvaka.github.io/map-of-github/#2/0.25/4.3, a rather interesting “Map of Github”.
Share DRIFFT
Know someone looking to level-up their AI and opensource game in mining? Just copy and paste this link: https://drifft.beehiiv.com
Thanks for reading! Want me to look into a particular topic? Email your suggestions and and I will dig.