Hyperspectral Compositional Mapping Tool
This guide provides a step-by-step walkthrough to using Marigold's latest new feature for hyperspectral processing, using the Reko Diq porphyry system as a case study.
Introduction to Marigold’s Hyperspectral Compositional Mapping Tool
This tool was developed to help make characterization and quantification of hyperspectral mapping easier and more consistent for geologists who are not remote sensing experts. It includes both the assessment of mineral group and mineral species relative abundance as well as compositional differentiation for selected mineral groups and species members. This is a tool that will help the geologist build on the compositional mapping results derived from multispectral data, and to help them derive good quality repetitive analyses from infinitely more complex hyperspectral data. This enables characterization of minerals and mineral groups that are relatively widespread across mineral deposit types, addresses the mapping needs of some of the most common and currently explored for deposits (such as epithermal and porphyry systems), generating results that are immediately useful for targeting, but equally can be readily validated using in-scene spectral data. Only select mineral groups / species have been developed to-date, although more are planned and slated for development in the future, supported by exploration, research and documented by recognized research in the industry.
Guide to Use of the Marigold Compositional Mapping Tool
The tool can be found under the Hyperspectral drop-down menu in the Marigold Mineral Processing Toolbox. There is also an ‘Advanced’ version of the tool planned (to be released in future).
The objective of this tool is NOT to identify that a specific mineral is present, rather, it is to characterize and quantify the relative abundance and distribution of the mineral and its compositional variation, as measured spectrally. The presence of a mineral should be estimated first using multispectral (MS) methods, then confirmed spectrally using in-scene or spectral library spectra using and supervised classification and spectral unmixing algorithms before the Hyperspectral Compositional Mapping Tool is employed. This tool does not work with Multispectral (MS) data such as Sentinel-2, ASTER or the Fused BEC.
Step 1 - Verify the status of the selected input Hyperspectral (HSI) dataset.
EnMap, PRISMA, EMIT, or other airborne or spaceborne hyperspectral datasets have all been demonstrated to work effectively with the tool. Regardless of the dataset, it should be cleaned using the Data Statistical Analysis Tool to remove any bands with 0 or no-data values prior to use in the tool. Typical data input: cleaned EnMap_Hyperspectral data [cEnMap_HSI].
Step 2 - Ensure that all required masking has been done for the HSI data.
This typically includes masking for water and vegetation, but may also include masking for cloud, cloud shadow, snow, ice or topographic shadow. Typical input data [cEnMap_vwmsk_HSI].
Step 3 – Under the Hyperspectral drop-down menu, select the compositional mapping tool.
This will open a settings menu. Within the settings, select the input dataset [cEnMap_vwmsk_HSI], the base data name for the products i.e. [Compositional Mapping] or [CompMap_] as you choose, and the mineral to be mapped (e.g., white mica). This will automatically initiate the computation, and the results will be added to the Marigold processing workflow raster data layers within the [CompMap – Outputs] folder.
Step 4 – Examine the Compositional Mapping Results – Relative Abundance.
The objective for this step is to assess and validate the values and distribution of the relative abundance data.
There is a relative abundance layer and a compositional layer generated for each mineral/mineral group. Examine each result individually.
How do the results compare with what you know about the presence and abundance of the mineral / mineral group across your AOI to-date?
Turn on the legend for each layer [click on the three dots at the end of the layer name], and select [toggle legend]. The default relative abundance is typically between a high of 0.2 and a low near 0, and a ‘jet’ color ramp is applied to the data, such that pixels with higher relative abundance are red, and those with lower relative abundance are blue.
Does this pattern match what you saw from the MS data?
For example, white mica should be prominent in the MS product [AlOH_b6 index].
Are the patterns and distribution of the mineral group consistent with / match this product well?
Note that this CompMap result will include contributions from all white mica species (i.e. paragonite, muscovite, phengite, illite), and may also have contributions from other Al-OH species such as nontronite, illite-smectite, etc. You can validate the distribution by opening the settings window for the layer, and verify that the relative abundance values are consistent with the data by selecting the magic wand tool (linear 2% stretch) for the layer and look at the distribution of the relative abundance values.
Do the default values given match the distribution of the data? Perhaps they should be adjusted.
For example, an examination of the data may show many lower values, and a better fit for the upper range may also be lower at 0.14, as shown. You can adjust the values used for the legend by adjusting the min and max values for the histogram in the settings menu.
Don’t worry, you can’t break it! And you can always go back to the original values, or re-compute.
In general, what you want to show with the color bar is where the pixels with the highest values are. These pixels should dominantly and consistently contain the highest amount of the mineral predicted by the tool [in this case white mica], whereas the pixels with the lowest relative abundance will likely only contain a small amount of the predicted mineral. It is your choice to decide how you want to present your results.
Do you want the full distribution of abundance of the mineral displayed for your targeting, or only the pixels with the highest values should be mapped significantly. This is the same kind of decision you have to make when generating results for MS data, and any for any other RS layer.
Step 5 – Examine the Compositional Mapping Results- Composition.
The objective for this step is to assess the composition of the pixels and assess whether they match what is predicted by the Compositional Mapping results. What you are mapping in each case may be quite different depending on the mineral you selected.
In the case of [white mica], the primary compositions we are trying to differentiate are paragonite, muscovite and phengite. They are differentiated spectrally by the presence of their primary absorption (abs) feature. This abs feature is typically between 2185 and ~2195/2000 nm for Paragonite, 2195/2000 nm to 2210 nm for Muscovite, and 2210 – 2225 nm for Phengite. The best way to investigate the results of this mapping is to spectrally verify the composition of a selection of the pixels across the project area. Is the prominent alteration in the area white mica? This is a view of the original results produced by the tool. Is the distribution of values a good fit for the distribution of values, and resulting plot used by the color table. This is showing the original legend and color table distribution.
The second image below shows a better distributed application of the color table to fit the distribution of values. This can be investigated in more detail using the spectral query tool (Option under the + by the spectral plot).
- Does the identification of the various species as derived from the data match the predicted results?
- How good are the predictions in areas where the relative abundance is high, or low?
- Are you observing other minerals being mapped as white mica? This is possible, particularly in areas where the relative abundance of white mica might be lower, due to the presence of mineral mixtures.
- If this is the case, does the predicted result match what you are observing compositionally?
- Finally, is there a positional relationship between the position of the paragonitic vs phengitic white mica to mineralization such as Au (if known), or with structural features? In the above example there are 3 prominent abs positions 2199 nm, 2208 nm and 2216 nm, representing paragonite, muscovite and phengite. The paragonitic mica is associated with the intrusions/alteration centres (in blue).