Geophysical TIFF images for machine learning algorithms
If anybody has geophysical data that he/she would like me to try on the machine learning algorithms, then please send them to me by the end of next Wednesday (or end of week at most) but let me know by Tuesday if you have such data.
Here are what you can send:
- Lithology (no road nor rivers nor anything else to be saved on the TIFF map except for lithology). The TIFF image can have integer values for the different lithologies
- Distance to Fault TIFF
- Distance to Dyke TIFF
- Distance to Fold TIFF
- Distance to Contact TIFF
- Radiometric TIFF images (e.g. Kth, POT, THO, URA)
- Magnetic TIFF images (e.g. AS, RTP, RTP 1VD, TILT)
- Gravity TIFF images
- AOI (Area Of Interest) TIFF image (having integer values for say: unknown, area not of interest, area of interest). This image can be created based on drillhole grade values (see below).
- Regolith TIFF map (its chemistry indicates the bedrock extent and the type of mafic or felsic rocks it sits on, hence a good indication of mineralization)
- Elevation TIFF map (influences the soil properties because of slopes and rain precipitation rate)
- Ground-water TIFF map (due to the deep penetration and interaction with deeply buried deposits such as porphyry copper).
- pH TIFF map
2 - Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) TIFF images:
- AIOH content
- FeO content
- Gypsum Content
- Quartz index
3 - Hyperspectral TIFFs:
- AI Smectite
- Ferric Iron MGOH
- Green Vegetation
- Kaolin content
- Mica Abundance
- Water content
Make sure you eliminate the faults post to mineralization from the 'Distance to faults' TIFF image UNLESS a secondary enrichment occurred after these faults have come in place. Likewise for 'Distance to Folds'.
For the AOI image, you can define the integer values based on thresholds of a given chemical element (e.g. Gold values < 0.8g/t will be "area not of interest", 0.8 <= Gold =< 1.5 will be "area of interest", Gold > 1.5g/t will be "area of high interest")
It's also good to have a TIFF image that contains values of an anomalous chemical element (e.g. values of a pathfinder such as Zn which is related to sulphide mineralization, hence, an indicator to potential base metal deposits) that come from soil samples. If you have:
- Lots of soil sample measures: assign each pixel the closest value to its center or the single value that falls within it and give a value of -999 to the rest of pixels.
- Very few soil sample measures: first, run an interpolation over a small area (with a buffer if you wish) that contains the soil samples (don't extrapolate much otherwise the machine learning will be learning more from estimates than from actual values). Second, make sure the pixels that contain actual values inside them receive the actual values the closest to their centers (i.e. overwrite the estimates for these pixels by the actual values falling inside them and pick the closest to the center in case more than one actual value falls inside a pixel).
If the soil samples were sampled in B horizon, then another map of lag samples and/or rock chip (from rock exposures) samples will be good along with soil samples map for better correlation with the aerian non-penetrating images.
A TIFF image that splits the pixels containing the values of the anomalous chemical element into two or three spatial zones is primordial to have. Label 70% of pixels as 'zone 1' and the rest label them 'zone 2' (or, in case of three zones, 15% each for 'zone 2' and 'zone 3'). I need at least two zones, the 70% one serves for training and the 30% for testing. At least zone 2 or zone 3 must have a spatial contact with zone 1 so I can merge it with the latter in case I need to increase the size of the training data.
Thank you.