DIGITAL.CSIC. Repositorio Institucional del CSIC
Research data
Set de datos (Dataset)
Dataset, code, and figure outputs for “What controls maximum injection-induced earthquake magnitude?”
[Description of methods used for collection or generation of data] Selected variables from the open-access GEoREST database were characterized, and rock names were standardized into 12 categories. Missing values were filled using literature values, deterministic calculations, and random-forest regression. Moment_Max was neither imputed nor used for imputation.
Two LightGBM models were compared: an injection-only model and a full 12-feature model. Data were divided into 90% training and 10% testing subsets. Predictors were normalized from 0 to 1 using training-set ranges only., [Description of data processing and preparation]
1. Select and filter the GEoREST variables.
2. Fill supported literature values.
3. Calculate deterministic elastic and stress properties.
4. Impute remaining eligible values using random forests.
5. Merge the completed data and train the LightGBM models.
Excel colours show provenance: white = original, yellow = literature, orange = random forest, and green = deterministic. Full processing details are provided in the scripts., [Codebooks or data dictionaries] This README documents the final variables. The original GEoREST workbook and Kivi et al. (2023) provide the source-variable definitions., Fluid injection into the subsurface for energy and resource development has led to increased seismic activity in recent years. Understanding the key factors governing maximum induced earthquake magnitude remains critical for their improved prediction and risk assessment. In this study, we apply a machine-learning model using the LightGBM algorithm to a global induced seismicity database spanning multiple geoenergy applications and settings to identify major factors associated with maximum injection-induced moment magnitude. We find that incorporating operational parameters, site characteristics, and host rock properties as input variables, compared to only operational parameters, significantly increases the predictive model accuracy. Cumulative injected fluid volume and injection rate impose first-order controls on seismic magnitude, while rock porosity, permeability, pore pressure, and in-situ stress field are found to play important secondary roles. These results highlight the importance of physical site characteristics beyond injection parameters in enriching magnitude-scaling relations and developing more reliable, physics-based seismicity forecasting and risk mitigation strategies., - The authors acknowledge support for data collection from the European Research Council (ERC) under the European Union’s Horizon 2020 Research and Innovation Program through the starting grant GEoREST under grant agreement no. 801809.
- S.S. and V.V. acknowledge funding from the European Research Council through the Horizon Europe Proof-of-Concept project NOSHAKE, Grant Agreement No. 101248689.
- I.R.K. acknowledges funding from MICIU/AEI/10.13039/501100011033 and the European Social Fund Plus (FSE+) through the Ramón y Cajal fellowship RYC2024-048711-I.
- IMEDEA is a María de Maeztu Excellence Unit supported by Grant CEX2021-001198, funded by MICIU/AEI/10.13039/501100011033., With funding from the Spanish government through the "Maria de Maeztu Centre of Excellence" accreditation (CEX2021-001198)., File list Main files - README.md: workflow guide. - README_DATASETS.txt: dataset metadata. - requirements.txt: Python dependencies. - rf_script.py: random-forest imputation. - principal_stress_rf_script.py: principal-stress imputation. - final_dataset_script.py: final data merge and deterministic recalculation. - lightgbm_base_script.py: baseline LightGBM workflow. - lightgbm_normalization_script.py: normalized LightGBM workflow and figures. 01_Data_Curation - GEoREST_induced_seismicity_database_v20.11.2022.xlsx/.csv: source dataset. - DataCuration.xlsx: selected, standardized, literature-filled, and deterministic data. - Literature/RockProperties_4.png to RockProperties_8.png: literature tables used for filling rock properties. 02_RF_Imputation - RF_Datacurated.xlsx: random-forest-imputed data. - Audit_Outputs/principal_stress_values.csv: imputed principal stresses. - Audit_Outputs/principal_stress_validation.csv: stress-imputation validation. 03_Model_Training - Final_Imputed_Dataset.xlsx/.csv: final model input. - Previous_Baseline/Mean_RockSiteInjection.xlsx: earlier comparison dataset. - Results/model_comparison.csv: model-performance summary. - Results/new_v2_all_row_predictions.csv: model predictions. - Results/normalization_ranges.csv: training-set normalization ranges. - Results/normalized_labelled_model_inputs.csv: normalized model inputs. - Results/comparison_report.json: model settings and validation report. - Results/LightGBM_V2_TestV1_model.txt: trained LightGBM model. - Figures/F10.png, F21.jpg, F31.png, and F32.png: manuscript figures., Peer reviewed