Dr Chris Yeomans 🇪🇺 (@SWgeoscience) header
Dr Chris Yeomans 🇪🇺 (@SWgeoscience) profile picture

Dr Chris Yeomans 🇪🇺

@SWgeoscience

#Geoscientist #metals #geothermal #machinelearning #datascience

 South West, England
swgeoscience.github.io
Joined May 2012
  • Tweets2.5K
  • Following929
  • Followers1.2K
Replying to @palaeobeas @hsrhughes @nom_D_plumes
@palaeobeas @hsrhughes @nom_D_plumes
3
We conclude that the #TiltDerivative is superior for its consistency across step changes in the data and its ability to identify lineaments off of major axes consistently. It's not perfect though as we identify some artefacts that require removing!
Three rose diagrams where A) is the original population of lineaments for the TDR transform; B) is the population of spurious lineaments; C) the final lineament population for the TDR transform following spatial filtering.

ALT Three rose diagrams where A) is the original population of lineaments for the TDR transform; B) is the population of spurious lineaments; C) the final lineament population for the TDR transform following spatial filtering.

3
Orientations can be vastly different, too! Where some are tightly grouped around major axes - the shortest path for lineaments to fit through pixels. The hillshade transform is a particular culprit, but commonly used! #geology #StructuralGeology #ImageAnalysis
Equal-area wedge rose diagrams for lineament populations derived from each feature extraction technique. The major NW-SE trend is apparent in all plots but with variable clarity. Plots are based on guidelines by Sanderson and Peacock (2020).

ALT Equal-area wedge rose diagrams for lineament populations derived from each feature extraction technique. The major NW-SE trend is apparent in all plots but with variable clarity. Plots are based on guidelines by Sanderson and Peacock (2020).

1
2
Different filters and transforms give different responses when using a semi-automated lineament detection method. Some capture more lineaments, some fewer, and at different depths (or elevation). It's particularly important when there are step changes in the data. #ImageAnalysis
Kernel Density Estimate (KDE) plot for the depths at which lineaments are sensed for each method where GRAD = gradient filter; LP3 = 3 x 3 Laplacian; LP5 = 5 x 5 Laplacian; HS = Hillshade and TDR = tilt derivative. The TDR-derived lineament population shows a clear deviation from others where lineaments at depths > -28 m are detected for this transform but not other operators.

ALT Kernel Density Estimate (KDE) plot for the depths at which lineaments are sensed for each method where GRAD = gradient filter; LP3 = 3 x 3 Laplacian; LP5 = 5 x 5 Laplacian; HS = Hillshade and TDR = tilt derivative. The TDR-derived lineament population shows a clear deviation from others where lineaments at depths > -28 m are detected for this transform but not other operators.

1
1
:: NEW PAPER THREAD :: Check out our new paper on using different transforms and filters in semi-automated #lineament analysis. It is #OpenAccess for the next six months or available on @EarthArXiv doi.org/10.1016/j.jsg.2021.1… A #BiscuitCorner publication. #geology #SciComm
A subset of the semi-automated lineament populations highlighting performance over the platform edge where the tilt derivative captures the most consistent lineament set. Derived from A) magnitude of gradient filter; B) magnitude of Sobel filter; C) 3× 3 Laplacian filter; D) 5 x 5 Laplacian filter; E) hillshade transform; F) tilt derivative (TDR) transform.

ALT A subset of the semi-automated lineament populations highlighting performance over the platform edge where the tilt derivative captures the most consistent lineament set. Derived from A) magnitude of gradient filter; B) magnitude of Sobel filter; C) 3× 3 Laplacian filter; D) 5 x 5 Laplacian filter; E) hillshade transform; F) tilt derivative (TDR) transform.

2
19
6
50
My cat wouldn’t shut up so I stepped away from the desk expecting a five minute fuss before she got bored... I may need to adapt my work environment today. Wish I’d brought stuff closer 😩 #CatsOfTwitter #WFH #office #cat
Image from Dr Chris Yeomans 🇪🇺 (@SWgeoscience)
6
Replying to @palaeobeas
@palaeobeas
1
Replying to @ZJAyres
@ZJAyres
2
I don’t normally do #ThinSectionThursday but here’s a CL image from a fault zone in southwest England and shows different #quartz generations, grey-yellow and blue, and some rather jazzy bands of what I’ve tentatively called hematite (orange, likely altered). Not a petrologist.
Image from Dr Chris Yeomans 🇪🇺 (@SWgeoscience)
2
3
21
I've always wanted to subscribe but never found the funds to do it... won't need to be finding the budget for that in the future
1
Whilst clearing space on my laptop I accidentally deleted something really important that stopped my email client from working. The suggested fix did not work. I have never been happier 🎉🤩🎉 #EmailFreeFriday
Image from Dr Chris Yeomans 🇪🇺 (@SWgeoscience)
8
Replying to @AcademicChatter
Imagine a world where Elsevier convince the general public to fund open access through crowdfunding. This guy would definitely be behind it. 😂
1
3
Replying to @geologyjess
I feel that. I also hope you got some good bonus stuff for levelling up 🤗
Replying to @Kate_Littler
*spends weekend reassessing career decisions*
However, t-SNE (t-distributed stochastic neighbour embedding) brought everything together. t-SNE takes all the information and preserves data structure whilst reducing the information to just two variables (embeddings). Now the #clusters are clear! doi.org/10.1016/j.gsf.2020.1…
Top is t-SNE embedding showing the lovely clusters. Below is location data from Hughes et al. (2015). The numbers point to specific localities of interest (see paper!)

ALT Top is t-SNE embedding showing the lovely clusters. Below is location data from Hughes et al. (2015). The numbers point to specific localities of interest (see paper!)

1
1
To make this a bit more manageable, Principal Component Analysis (PCA) reduced the number of variables and was used as inputs to a k-means clustering algorithm (#unsupervised learning). The #clusters showed some fairly nice correlations when plotted with the original #elements
Element scatter plots showing original data coloured by the clusters derived using Principal Component Analysis. Group 3 consistently shows high spread but is distinct from the more closely associated Group 1 and 2.

ALT Element scatter plots showing original data coloured by the clusters derived using Principal Component Analysis. Group 3 consistently shows high spread but is distinct from the more closely associated Group 1 and 2.

1
The main thrust of the paper is reducing complex #geochemical datasets into fewer variables that make it easier to analyse whilst retaining the information from lots of geochemical elements. @nom_D_plumes started with a rather nice correlation matrix - note Pd-Cu association.
Correlation matrix of all the input elements and how they relate to each other. Lots of information and perhaps the best way to get a quick look at what is associated with what, but it's not that useful for further analysis.

ALT Correlation matrix of all the input elements and how they relate to each other. Lots of information and perhaps the best way to get a quick look at what is associated with what, but it's not that useful for further analysis.

1
1
Replying to @jameshickey77
I see your shock and I raise you some Tory cronyism.
Image from Dr Chris Yeomans 🇪🇺 (@SWgeoscience)
Load more