Developing innovative software for mineral prospectivity mapping – EIS project at closure
Gispo has been part of the EU’s EIS (Exploration Information System) Horizon project for the past couple of years. The project’s aim is to aid the EU’s efforts in the green transition by securing critical raw materials. The project consists of several work packages ranging from conducting research to raising awareness of the importance of minerals in the green transition. We at Gispo have been most involved in a work package that aims to develop innovative software for mineral prospectivity mapping (MPM). Aside from software development, a large part of the work package has been planning, comparing existing methodology, arranging workshops, and holding interviews. We have discussed the project in two of our previous blog posts: you can check them here and here. Now, as the project is approaching its end, it is time to look back at what we have achieved.
During the project, Gispo has been developing two software applications : EIS Toolkit and EIS QGIS Plugin, with our partners Geological Survey of Finland (GTK), Beak consultants GmbH, University of Turku, and French Geological Survey (BRGM). EIS Toolkit is a standalone Python library which brings together and implements relevant tools for mineral prospectivity mapping. It is the first free open source software of its kind. To make using EIS Toolkit as easy and efficient as possible, EIS QGIS Plugin was created to provide a graphical interface for each individual tool and to implement EIS Wizard that provides guided workflows for MPM. GTK has been the main responsible party for developing EIS Toolkit and EIS QGIS Plugin development has been led by Gispo.
EIS Toolkit
Let’s have a look at what the two software applications contain. EIS Toolkit has a number of tools used in MPM. The tools currently implemented in the toolkit are divided into nine categories: conversions, evaluation, exploratory analysis, prediction, raster processing, training data tools, transformations, utilities and vector processing. Most categories contain several tools, but let’s go through just a few here. The full list of implemented tools can be found here.
Conversion and transformation tools are quite straightforward. Conversion tools can be used, for example, for taking raster pixel values and putting them in a Pandas dataframe. With transformation tools, you can e.g. binarize or normalize raster data, or perform transformations on compositional data (CoDa transformations).
Exploratory analysis tools are for exploring data using various plots, statistics and data transformation methods. These tools include algorithms such as K-means clustering, principal component analysis (PCA), and Local Morans’ I. All basic distribution and relational plots are included along with some more specialized plots like parallel coordinates plot.
Raster processing tools are for manipulating and preprocessing raster data. Using them, one can for example clip raster extents using polygons, snap rasters to grids and create distance rasters. With vector processing tools similar preprocessing tasks can be performed such as calculating distances from raster cells to the nearest geometries, performing IDW or kriging interpolation, and rasterizing vector data.
An important part of MPM is predicting the mineral deposit locations. The modeling module of EIS Toolkit focuses on ML models. The supervised methods include gradient boosting, logistic regression, random forest and multilayer perceptron. In addition to these tools, EIS Toolkit has fuzzy overlay and weights of evidence. For evaluating the performance of the ML models, EIS Toolkit contains tools for calculating different metrics, such as precision and recall, and visualization such as for plotting confusion matrix or prediction area curve.
The tools mentioned above form only a small part of all the tools included in EIS Toolkit. The toolkit is built to be modular and can be extended with new tools in the future.
EIS Toolkit is a Python package and as such, it can be used in Python scripts as a library. Since the beginning of development, convenience of use has been a priority. All of the tools are categorized, documented with docstrings and have a thought-through number of parameters, which should make the toolkit approachable. To get started with EIS Toolkit, we suggest creating an empty Python virtual environment with a compatible Python version and installing EIS Toolkit there.
Since EIS Toolkit has a large number of dependencies, conflicts with the dependencies of other Python software are very likely to occur. This is why EIS Toolkit has an additional CLI API that is primarily intended for integrations with other software (such as QGIS). Using the CLI API, each tool can be accessed in a well-defined way by other programs that execute EIS Toolkit as an isolated Python process. The CLI API can be used directly in a traditional geoprocessing fashion too, but so far the focus has not been to make it a convenient interface for people.
EIS QGIS Plugin
Next, let’s look at the EIS QGIS Plugin. It consists of two parts: EIS QGIS processing algorithms and EIS Wizard. The processing algorithms can be accessed from QGIS Processing Toolbox.
A more guided way to use the plugin is via EIS Wizard. It is a GUI ( graphical user interface) designed to perform MPM workflows in a structured way. EIS wizard consists of pages that are designed to be walked through from top to bottom in general, but in a non-restrictive way that allows switching pages whenever the user wishes so. The pages of EIS Wizard are Mineral system proxies, EDA, Modeling, Evaluation, History, Settings and About
On the Mineral system proxies page users can process their measurement data so that it can be fed into the prediction models. Users can use the provided default IOCG (iron oxide copper gold) mineral system proxies or create their own mineral systems with custom proxies. These custom mineral systems can be exported and imported as JSON files, making it possible to build a shared mineral system library in the future. Each proxy is associated with a workflow with one or more processing steps. The available processing steps are:
- Distance to features
- Distance to anomaly
- Interpolate
- Binarize
- Proximity to anomaly
- Proximity to features.
The output of a workflow is a proxy raster that contains some information that is believed to correlate with mineral deposit locations. The raw measurement data is processed to proxy form to filter out irrelevant pieces/noise, normalize data ranges and to facilitate a mineral system -driven approach for the modeling process among some other reasons.
At any time, users can inspect their data in the EDA page: EIS Wizard enables creating plots for vector and raster data. Users can also compute different statistics from their data or use the exploratory analysis tools.
In the Modeling page, users can apply ML and data-driven models on their data. The page consists of a model selection dropdown menu and tabs for different steps in the modeling process. For supervised ML models the tabs are data preparation, training, testing and application. In the data preparation tab, users can prepare their data with several transformation tools before moving on to training the selected model with the data. After training and testing the model, it can be applied to classification/regression tasks on new, unseen data to make the actual predictions. Fuzzy overlay and weights of evidence don’t follow the same train-test-predict workflow and therefore have their own sets of tabs.
In the Evaluation page, users can evaluate their model by computing several metrics such as accuracy, precision, recall and F1 score. Users can also plot, for example, a confusion matrix or a ROC curve to evaluate the prediction results.
On the History page, users can view information about previously trained models whereas on the Settings page, EIS Toolkit configuration is set and various customizations to the user experience can be made. For a more thorough overview of EIS QGIS Plugin, please visit its wiki page.
Retrospective
The project has been largely successful in terms of the software work package. We implemented a wide range of tools, with EIS Toolkit reaching its full release in May 2024 and EIS QGIS Plugin last November. Since then, both software applications have received new features and undergone minor improvements.
However, not all planned features were implemented. The originally intended support for mineral systems beyond IOCG is missing, mainly due to dependencies on other work packages. To address this, we have provided users with the option to define custom mineral systems.
Overall, we are pleased with the project’s results and the potential impact of these tools. EIS Toolkit brings together relevant tools for MPM like no other free software before, while EIS QGIS Plugin offers a user-friendly interface for navigating each step of the MPM workflow in QGIS. As open source software, they are freely accessible without licensing costs. We look forward to seeing them adopted by geologists and mining professionals. If you feel that some tool that is important for your workflow is missing, or you need help with using either software, don’t hesitate to reach out.
