MIDSTREAM applications

The MIDSTREAM model is composed of three main elements:

• MIDSTREAM – Load data: A module for data preparation (which reads the input data and generates the datasets used for the successive two modules).

• MIDSTREAM – Model training: A module for the inference of model parameters (learning);

• MIDSTREAM – Testing and forecasting: a module for testing the model's performance and forecasting the seismicity rates as a function of fluid injection activity (testing & forecasting).

MIDSTREAM: Load data application

Use this module to load the data for the analysis. The required data is composed by: i) a seismic catalog containing at least event times and magnitudes, and ii) injection data (usually the injection rate), that indicatively should be sampled at times with a comparable order of magnitude as the average inter-event times in the catalog.
In the following paragraphs we show an example of how to prepare the data for the analysis with the MIDSTREAM service:
1.1 Load the MIDSTREAM – Load Data application in the user's workspace.
This can be done by accessing the EPISODES platform, selecting the application "MIDSTREAM: load data", and then "add to workspace" (Figure 1.1). Once the application is loaded in the workspace, it should look as shown in Figure 1.2; note that all the controls are disactivated because no data has been selected yet.

Figure 1.1 List of applications in the EPISODES platform. Select the Load data module.


Figure 1.2 MIDSTREAM: load data tool loaded in the workspace.

1.2 Load the data for the analysis
The data can be loaded by the user (using the platform's tools for importing data -injection rate and the seismic catalog- from files sin csv format) or directly from the episodes available in the platform.
1.2.1 Loading data from an available Episode
The simplest way to test this tool is to use data from an episode already available in the platform. Since the model has been developed mostly to analyze data from pressurized fluid injections, in this example we illustrate the analysis of seismicity associated with an episode related with geothermal energy development in Cooper Basin. Select this episode in the platform as shown in Figure 1.3a and 1.3b, and select the seismic data and the relative time series of the industrial forcing (Figure 1.4) to be added to the load data module. When you select the data to load, a pop-up window appears for selecting the point in the workspace where to load the data (Figure 1.5).

(a)

(b)

Figure 1.3 Selecting an episode (Cooper Basin) from the dataset available in the platform: the Catalog contains the information related with the seismic events, and "injection rate" the data of the industrial forcing used to test the eventual correlation.

(a)

(b)

Figure 1.4 Selecting the data for the analysis: (a) seismic catalog and (b) relative injection rate data.

Figure 1.5 Select the directory where the files will be loaded (in this example "Midstream – LoadData (example).

1.2.2 Loading data CSV from local computer
CSV (comma separated values) ascii files containing the earthquake catalog and the injection rate time series can be loaded and imported by using dedicated applications available in the EPISODES platform. The suer is invited to follow the instructions for using these applications:

• CSV to Catalog converter:

Tool for conversion of a CSV file into a Catalog in EPISODES Platform format. The CSV file should use commas (',') as delimiters and use English locale for numbers. The text fields can be enclosed in double quotes. The file has to include header line with names of the columns. See details in the link: https://episodesplatform.eu/?lang=en#app:CsvToCatalog

• CSV to GDF converter:

Tool for conversion of a CSV file into a Matlab format - GDF used by EPISODES Platform (see more in the EPISODES Platform Documentation). The CSV file should use commas (',') as delimiters and use English locale for numbers. The text fields can be enclosed in double quotes. The file has to include header line with names of the columns.
The current version of the application converts only CSV files compatible with GDF with time-correlated parameters (GDF categories: "Data containing one parameter and date" and "Data containing parameters and dates"), future versions might be compatible also with other GDF types. See details in the link:
https://episodesplatform.eu/?lang=en#app:CsvToGdf

Once the seismic catalog and the injection rate data have been imported, assign these files to the MIDSTREAM – load data application in the workspace and continue with the instrucitons in the following paragraphs.
1.3 Select data and settings for data partition (training and testing)
Once the data (seismic and injection rate) have been loaded in the MIDSTREAM – Load Data application in the user's workspace (Figure 1.6a), select the files in the respective fields ('Catalog' and 'Injection rate'); once this is done, all the fields in the application's interface appear active and you will be able to set the partition of the data between training' and 'testing datasets (Figure 1.6b). The data partition can be done inserting directly the date and time in the respective fields (start, split and end time) or by using the chart pressing the "Select from Chart" button (Figure 1.6c).

(a)

(b)

(c)

Figure 1.6. (a) MIDSTREAM – load data application with the a seismic catalog and an injection rate time series correctly loaded in the work space; (b) after selecting the these files, the application becomes active and the user choices can be performed. (c) Select start, split, and end time from chart.

Other parameters needing to be set before you execute the application are:

• Delay time (in seconds): an estimated delay time between the start of the pressurized injection and the seismic response of the reservoir.
• Magnitude column: select the correct magnitude data to be used in the analysis.
• Completeness magnitude (Mc): set the completenss magnitude for the seismic catalog (Note: if Mc is unknown, you can use the application for completeness magnitude determination from the application list).

Once all the parameters are set, press the RUN button and wait until the results are produced. As output, two files (and the respective plots, as shown in Figure 1.7): the training dataset, and the testing dataset. Such two files are the are input data for the following two modules of this service:

Figure 1.7. Plot of the two output data files: the training dataset, which is the input file to the "MIDSTREAM – training module", and the testing dataset which is one of the inputs to the MIDSTREAM – Testing module.

MIDSTREAM: Model Training

The code implementing the model by Garcia-Aristizabal (2018) for testing the correlation between seismicity rates (inter-event times, IET) and fluid injection rates. The code allows the user to test and compare 3 competing models, namely:

• Stationary (i.e., seismicity rate is independent of injection rate)
• Power-law relationship between mean IET and injection rate.
• Exponential relationship between mean IET and injection rate.

Details of the model can be found in the reference paper (Garcia-Aristizabal 2018). The input data needs to be formatted and created using the MIDSTREAM – Load data module, as the one shown in Figure 2.1 (see details in Section 1 of this guide).

Figure 2.1 Example of the training dataset created by the MIDSTREAM – Load data module.

To use this module, you can either load the MIDSTREAM – Model training in the workspace (in a similar way as shown in Figure 1.1 and 1.2 for the load data module), and then select the training dataset created by the load data module, or you can directly select the relevant application by selecting the file and, clicking in the three vertical points near the training dataset file, select 'use in application', as shown in Figure 2.2.

Figure 2.2 Direct association between the training dataset and the MIDSTREAM – training model application.

Once the application is in the workspace, and the input file has been assigned, you can set the analysis configuration by setting the different parameters as follows (Figure 2.3):

• Model (select among stationary, power law, power function). This is the deterministic model to test the relationship between inter-event times and injection rate.
• Uncertainty bounds: low and high percentiles for defining uncertainty bounds in the results
• Units of the injection rate data (units of volume / units of time, as e.g.: m³/s)
• Injection rate threshold (minimum value to consider in the analysis). Default value is zero.
• Units of inter-event times (e.g., seconds, or days)
• Inter-event time (IET) min threshold: minimum IET to consider in the analysis (default value: zero).

Figure 2.3 Setting the basic model parameters for executing MIDSTREAM – Model training.


Figure 2.4 Advanced options, mostly related to the setting of the Markov chain Monte Carlo approach for model parameter estimation. In most of the uses of this code, such values can remain set according to the default values.

The outputs of the processing are stored in a number of files, as shown in Figure 2.5. The files are named according to the following structure:
root_file_name = midstream_training_dataset_analysis_Model_XXX_
where XXX corresponds to the tested model (i.e., Stationary, Power law, Exponential)

Figure 2.5. List of outputs after running Midstream model training.

The output files include:

• Data files:
  • root_file_name _MODEL_TRAINING.mat: a file with all the environment variables of the analysis (Octave/Matlab format)
  • root_file_name _summary_parameters.txt: ascii file (Fig. 2.5a):

Figure 2.5a. with a summary of the results of the model fit.

• • root_file_name _parameter_samples.txt: ascii file (Figure 2.5b)

Figure 2.5b. with a vector containing model parameter samples extracted from the posterior distribution (in the domain of model parameters).

• • root_file_name _ModelFit_data.txt: ascii file with 3 columns of data: Injection rate (ir) in units of volume/units of time, median of mu(ir) in days, 16th perc of mu(ir) in days, 84th perc of mu(ir) in days (percentiles as set in when running the model training)

Figure 2.5c. Model fit data.

• Figures:
  • root_file_name _Param_samples.jpg (Fig. 2.5d): plot of the markov chain data (after removing the burn-in and resampling the chains).

Figure 2.5d. Samples of the Markov chain (after removing the burn-in and resampling the chains); plot of the data in the file shown in Fig. 2.5a.

• • root_file_name _Parameter_correlation.jpg (Fig. 2.5e): For models with a number of model aprameters > 1, it calculates and plots the correlation between the model aprameetr values.

Figure 2.5e. Plot showing the correlation between model parameters.

References
Alexander Garcia-Aristizabal, Modelling fluid-induced seismicity rates associated with fluid injections: examples related to fracture stimulations in geothermal areas, Geophysical Journal International, Volume 215, Issue 1, October 2018, Pages 471–493, doi: 10.1093/gji/ggy284.

Acknowledgements

The development of this tool have been performed with the financial support of the Joint Research unit (JRU) of EPOS Italia (https://www.epos-italia.it/)

Back to top

Related Documents