242 points by antidnan 10 hours ago | 139 comments
folli 9 hours ago
An RF machine-learning model was developed to predict lithium concentrations in Smackover Formation brines throughout southern Arkansas. The model was developed by (i) assigning explanatory variables to brine samples collected at wells, (ii) tuning the RF model to make predictions at wells and assess model performance, (iii) mapping spatially continuous predictions of lithium concentrations across the Reynolds oolite unit of the Smackover Formation in southern Arkansas, and (iv) inspecting the model for explanatory variable importance and influence. Initial model tuning used the tidymodels framework (52) in R (53) to test XGBoost, K-nearest neighbors, and RF algorithms; RF models consistently had higher accuracy and lower bias, so they were used to train the final model and predict lithium.
Explanatory variables used to tune the RF model included geologic, geochemical, and temperature information for Jurassic and Cretaceous units. The geologic framework of the model domain is expected to influence brine chemistry both spatially and with depth. Explanatory variables used to train the RF model must be mapped across the model domain to create spatially continuous predictions of lithium. Thus, spatially continuous subsurface geologic information is key, although these digital resources are often difficult to acquire.
Interesting to me that RF performed better the XGBoost, would have expected at least a similar outcome if tuned correctly.
jofer 5 hours ago
However, kriging is really quite difficult to use with non-continuous inputs. RF is a lot more forgiving there. You don't need to develop a covariance model for discrete values (or a covariance model for how the different inputs relate, either).
aaronblohowiak 19 minutes ago
lordgrenville 6 hours ago
I would hazard a guess that with better tuning, XGBoost would still have won. (The paper notes that the authors chose a suboptimal set of hyperparameters out of fear of overfitting - maybe the same logic justifies choosing a suboptimal model type...)
levocardia 6 hours ago
I haven't read in detail what their validation strategy is but this seems like the kind of problem where it's not so easy as you'd think -- you need to be very careful about how you stratify your train, dev, and test sets. A random 80/10/10 split would be way too optimistic: your model would just learn to interpolate between geographically proximate locations. You'd probably need to cross-validate across different geographic areas.
This also seems like an application that would benefit from "active learning". given that drilling and testing is expensive, you'd want to choose where to collect new data based on where it would best update your model's accuracty. A similar-ish ML story comes from Flint, MI [1] though the ending is not so happy
[1] https://www.theatlantic.com/technology/archive/2019/01/how-m...
jandrese 7 hours ago
jofer 5 hours ago
Animats 6 hours ago
It's in a caldera in a mountain that I-80 bypassed to go through Winnemuca, Nevada. Nearest town is Mill City, NV, which is listed as a ghost town, despite being next to I-80 and a main line railroad track. The mine site is about 12km from Mill City on a dirt road not tracked by Google Street View.
Google Earth shows signs of development near Mill City. Looks like a trailer park and a truck stop. The road to the mine looks freshly graded. Nothing at the mine site yet.
It's a good place for a mine. There are no neighbors for at least 10km, but within 15km, there's good road and rail access.
diggernet 6 hours ago
Searching in Google Maps, Thacker Mine comes up as 40.58448942010599, -117.8912129833345. As you say, that is near I-80 and Mill City, and there is nothing there.
But Wikipedia says it's at 41.70850912415866, -118.05475061324945 in the McDermitt Caldera, nowhere near Mill City or I-80.
I'm thinking probably don't trust Google on this one. :)
Animats 5 hours ago
"Lithium Americas will contract with a bus company to drive workers an hour to the site for 10-hour work shifts, he added. An additional two hours will be added for transportation time. If you go to work on our project, you will have free room and board and free transportation to the site every day. You would get three free meals a day." If you're an unemployed coal miner in West Virginia, that might look good.
[1] https://www.nevadaappeal.com/news/2024/oct/12/nevada-operati...
CSSer 2 hours ago
mjrpes 5 hours ago
jeffbee 3 hours ago
Then I asked chatgpt and it refused to make a map but said that I should just look on the map for Thacker Pass, which is almost right but it also said I should look northeast of Winnemucca, which isn't correct. It's north and west.
Zero for two, for robots.
_heimdall 5 hours ago
It is interesting to see how much of this data could be modelled based on wastewater brines from other industries in the area, assuming we go on to mine the lithium it will say a lot if the ML predictions prove accurate.
One thing I couldn't tell, and its probably just a limitation of how much time I could spend reading the source paper, is what method would be needed to extract the bulk of the lithium expected to be there. If processing brine water is sufficient that may be easier to control externalities than if they have to strip mine and get all the overburden out of the way first.
jofer 5 hours ago
It's mining brine. I.e. the "mines" are basically deep water wells.
The limestone itself doesn't have any lithium. It's the water in the pores in the limestone that is relatively concentrated in lithium.
In most of these cases, you're already producing brines from the smackover formation as a part of existing oil and gas production, but the brine is being re-injecting after oil is separated from it. The idea is that it's better to keep those and evaporate them down for lithium production.
That does require large evaporation ponds, generally speaking, but it's not strip mining.
_heimdall 4 hours ago
As far as evap ponds go, are there usually chemicals or elements in the same brine water as lithium that is important when evaporating into the atmosphere?
jofer 2 hours ago
First and foremost, here are definitely lots of other salts. It is brine, after all. You produce a lot of halite (salt), gypsum, calcite, and all kinds of other evaporite minerals.
There are all kinds of things in smaller concentrations, though.
What comes out of a oil/water separator would need lots of additional processing before going to something like an evap pond. It's relatively hazardous stuff for a lot of reasons other than oil (e.g. it can be rather radioactive). It typically goes through quite a bit of additional processing unless it's being immediately reinjected.
lazide 45 minutes ago
pfdietz 3 hours ago
jofer 2 hours ago
pfdietz 58 minutes ago
jeffbee 3 hours ago
Do you have the same trepidation about aluminum, iron, dish soap, and table salt? I ask because the amount of "ripping open" involved in lithium production is like a speck in the eye of a whale compared to all the other mining. In terms of scale all existing and proposed lithium mines are teensy tiny by the standards of mines.
_heimdall 3 hours ago
jeffbee 3 hours ago
lazide 43 minutes ago
Which, for those of us that like moonscape, is a bit sad. But there is a lot of moonscape in that region, and there aren’t a huge number of moonscape fans. At least that are going to try to picket any projects. So overall, meh.
That area of Nevada is also pretty economically ‘challenged’, so why not.
tommykins 5 hours ago
Very good work - but typically we don't build prospectivity models this way (or rather we don't validate them this way anymore). Great to see the USGS starting to dip their toe back in this though, they and the GSC were long the leaders in this, but have dropped it on the last 5-7 years.