Probabilistic Estimation of Losses, Injuries, and Community resilience Under Natural hazard events

What is it?

pelicun is a Python package that provides tools for assessment of damage and losses due to natural hazard events. It uses a stochastic damage and loss model that is an extension of the high-resolution PEER performance assessment methodology described in FEMA P58 (FEMA, 2012). While FEMA P58 aims to assess the seismic performance of a building, with pelicun we provide a more versatile, hazard-agnostic tool to assess the performance of several types of assets in the built environment.

Detailed documentation of the available methods and their use is available at http://nheri-simcenter.github.io/pelicun

What can I use it for?

pelicun quantifies losses from an earthquake or hurricane scenario in the form of decision variables. This functionality is typically utilized for performance-based engineering and regional risk assessment. There are several steps of performance assessment that pelicun can help with:

• Describe the joint distribution of asset response. The response of a structure or other type of asset to an earthquake or hurricane wind is typically described by so-called engineering demand parameters (EDPs). pelicun provides methods that take a finite number of EDP vectors and find a multivariate distribution that describes the joint distribution of EDP data well. You can control the type of target distribution, apply truncation limits and censor part of the data to consider detection limits in your analysis. Alternatively, you can choose to use your EDP vectors as-is without resampling from a fitted distribution.

• Define the damage and loss model of a building. The component damage and loss data from the first two editions of FEMA P58 and the HAZUS earthquake and hurricane models for buildings are provided with pelicun. This makes it easy to define building components without having to collect and provide all the data manually. The stochastic damage and loss model is designed to facilitate modeling correlations between several parameters of the damage and loss model.

• Estimate component damages. Given a damage and loss model and the joint distribution of EDPs, pelicun provides methods to estimate the amount of damaged components and the number of cases with collapse.

• Estimate consequences. Using information about collapse and component damages, the following consequences can be estimated with the loss model: reconstruction cost and time, unsafe placarding (red tag), injuries and fatalities.

Why should I use it?

1. It is free and it always will be.
2. It is open source. You can always see what is happening under the hood.
3. It is efficient. The loss assessment calculations in pelicun use numpy, scipy, and pandas libraries to efficiently propagate uncertainties and provide detailed results quickly.
4. You can trust it. Every function in pelicun is tested after every commit. See the Travis-CI and Coveralls badges at the top for more info.
5. You can extend it. If you have other methods that you consider better than the ones we already offer, we encourage you to fork the repo, and extend pelicun with your approach. You do not need to share your extended version with the community, but if you are interested in doing so, contact us and we are more than happy to merge your version with the official release.

Installation

For users

pelicun is available at the Python Package Index (PyPI). You can simply install it using pip as follows:

pip install pelicun

If you are interested in using an earlier version, you can install it with the following command:

pip install pelicun==2.6.0

Note that 2.6.0 is the last minor version before the v3.0 release. Other earlier versions can be found here.

For contributors

Developers are expected to fork and clone this repository, and install their copy in development mode. Using a virtual environment is highly recommended.

# Clone the repository:
git clone https://github.com/<user>/pelicun
cd pelicun
# Switch to the appropriate branch, if needed.
# git checkout <branch>

# Install pelicun:
# Note: don't forget to activate the corresponding environment.
python -m pip install -e .[development]

Contributions are managed with pull requests. It is required that contributed code is PEP 8 compliant, does not introduce linter warnings and includes sufficient unit tests so as to avoid reducing the current coverage level.

The following lines implement the aforementioned checks. flake8, pylint and pytest can all be configured for use within an IDE.

cd <path-to>/pelicun
export PYTHONPATH=$PYTHONPATH:$(pwd)

# Linting with flake8:
flake8 pelicun

# Linting with pylint:
pylint pelicun

# Type checking with mypy:
mypy pelicun --no-namespace-packages

# Running the tests:
python -m pytest pelicun/tests --cov=pelicun --cov-report html
# Open `htmlcov/index.html`in a browser to see coverage results.

Feel free to open an issue if you encounter problems setting up the provided development environment.

Changelog

Changes in v3.3

• Changes affecting backwards compatibility

  • Remove "bldg" from repair consequence output filenames: The increasing scope of Pelicun now covers simulations for transportation and water networks. Hence, labeling repair consequence outputs as if they were limited to buildings no longer seems appropriate. The bldg label was dropped from the following files: DV_bldg_repair_sample,DV_bldg_repair_stats,DV_bldg_repair_grp, DV_bldg_repair_grp_stats, DV_bldg_repair_agg, DV_bldg_repair_agg_stats.

• Deprecation warnings

  • Remove Bldg from repair settings label in DL configuration file: Following the changes above, we dropped Bldg from BldgRepair when defining settings for repair consequence simulation in a configuration file. The previous version (i.e., BldgRepair) will keep working until the next major release, but we encourage everyone to adopt the new approach and simply use the Repair keyword there.

• New features

  • Location-specific damage processes: This new feature is useful when you want damage to a component type to induce damage in another component type at the same location only. For example, damaged water pipes on a specific story can trigger damage in floor covering only on that specific story. Location-matching is performed automatically without you having to define component pairs for every location using the following syntax: '1_CMP.A-LOC', {'DS1': 'CMP.B_DS1'} , where DS1 of CMP.A at each location triggers DS1 of CMP.B at the same location.

  • New custom_model_dir argument for DL_calculation: This argument allows users to prepare custom damage and loss model files in a folder and pass the path to that folder to an auto-population script through DL_calculation. Within the auto-population script, they can reference only the name of the files in that folder. This provides portability for simulations that use custom models and auto population, such as some of the advanced regional simualtions in SimCenter's R2D Tool.

  • Extend Hazus EQ auto population sripts to include water networks: Automatically recognize water network assets and map them to archetypes from the Hazus Earthquake technical manual.

  • Introduce convert_units function: Provide streamlined unit conversion using the pre-defined library of units in Pelicun. Allows you to convert a variable from one unit to another using a single line of simple code, such as converted_height = pelicun.base.convert_units(raw_height, unit='m', to_unit='ft') While not as powerful as some of the Python packages dedicated to unit conversion (e.g., Pint), we believe the convenience this function provides for commonly used units justifies its use in several cases.

• Architectural and code updates

  • Split model.py into subcomponents: The model.py file was too large and its contents were easy to refactor into separate modules. Each model type has its own python file now and they are stored under the model folder.

  • Split the RandomVariable class into specific classes: It seems more straightforward to grow the list of supported random variables by having a specific class for each kind of RV. We split the existing large RandomVariable class in uq.py leveraging inheritance to minimize redundant code.

  • Automatic code formatting: Further improve consistency in coding style by using black to review and format the code when needed.

  • Remove bldg from variable and class names: Following the changes mentioned earlier, we dropped bldg from lables where the functionality is no longer limited to buildings.

  • Introduce calibrated attribute for demand model: This new attribute will allow users to check if a model has already been calibrated to the provided empirical data.

  • Several other minor improvements; see commit messages for details.

• Dependencies

  • Ceiling raised for pandas, supporting version 2.0 and above up until 3.0.

Changes in v3.2

• Changes that might affect backwards compatibility:

  • Unit information is included in every output file. If you parse Pelicun outputs and did not anticipate a Unit entry, your parser might need an update.

  • Decision variable types in the repair consequence outputs are named using CamelCase rather than all capitals to be consistent with other parts of the codebase. For example, we use "Cost" instead of "COST". This might affect post-processing scripts.

  • For clarity, "ea" units were replaced with "unitless" where appropriate. There should be no practical difference between the calculations due to this change. Interstory drift ratio demand types are one example.

  • Weighted component block assignment is no longer supported. We recommend using more versatile multiple component definitions (see new feature below) to achieve the same effect.

  • Damage functions (i.e., assign quantity of damage as a function of demand) are no longer supported. We recommend using the new multilinear CDF feature to develop theoretically equivalent, but more efficient models.

• New multilinear CDF Random Variable allows using the multilinear approximation of any CDF in the tool.

• Capacity adjustment allows adjusting (scaling or shifting) default capacities (i.e., fragility curves) with factors specific to each Performance Group.

• Support for multiple definitions of the same component at the same location-direction. This feature facilitates adding components with different block sizes to the same floor or defining multiple tenants on the same floor, each with their own set of components.

• Support for cloning demands, that is, taking a provided demand dataset, creating a copy and considering it as another demand. For example, you can provide results of seismic response in the X direction and automatically prepare a copy of them to represent results in the Y direction.

• Added a comprehensive suite of more than 140 unit tests that cover more than 93% of the codebase. Tests are automatically executed after every commit using GitHub Actions and coverage is monitored through Codecov.io. Badges at the top of the Readme show the status of tests and coverage. We hope this continuous integration facilitates editing and extending the existing codebase for interested members of the community.

• Completed a review of the entire codebase using flake8 and pylint to ensure PEP8 compliance. The corresponding changes yielded code that is easier to read and use. See guidance in Readme on linting and how to ensure newly added code is compliant.

• Models for estimating Environmental Impact (i.e., embodied carbon and energy) of earthquake damage as per FEMA P-58 are included in the DL Model Library and available in this release.

• "ListAllDamageStates" option allows you to print a comprehensive list of all possible damage states for all components in the columns of the DMG output file. This can make parsing the output easier but increases file size. By default, this option is turned off and only damage states that affect at least one block are printed.

• Damage and Loss Model Library

  • A collection of parameters and metadata for damage and loss models for performance based engineering. The library is available and updated regularly in the DB_DamageAndLoss GitHub Repository.

  • This and future releases of Pelicun have the latest version of the library at the time of their release bundled with them.

• DL_calculation tool

  • Support for combination of built-in and user-defined databases for damage and loss models.

  • Results are now also provided in standard SimCenter JSON format besides the existing CSV tables. You can specify the preferred format in the configuration file under Output/Format. The default file format is still CSV.

  • Support running calculations for only a subset of available consequence types.

• Several error and warning messages added to provide more meaningful information in the log file when something goes wrong in a simulation.

• Update dependencies to more recent versions.

• The online documentation is significantly out of date. While we are working on an update, we recommend using the documentation of the DL panel in SimCenter's PBE Tool as a resource.

Changes in v3.1

• Calculation settings are now assessment-specific. This allows you to use more than one assessments in an interactive calculation and each will have its own set of options, including log files.

• The uq module was decoupled from the others to enable standalone uq calculations that work without having an active assessment.

• A completely redesigned DL_calculation.py script that provides decoupled demand, damage, and loss assessment and more flexibility when setting up each of those when pelicun is used with a configuration file in a larger workflow.

• Two new examples that use the DL_calculation.py script and a json configuration file were added to the example folder.

• A new example that demonstrates a detailed interactive calculation in a Jupyter notebook was added to the following DesignSafe project: https://www.designsafe-ci.org/data/browser/public/designsafe.storage.published/PRJ-3411v5 This project will be extended with additional examples in the future.

• Unit conversion factors moved to an external file (settings/default_units) to make it easier to add new units to the list. This also allows redefining the internal units through a complete replacement of the factors. The internal units continue to follow the SI system.

• Substantial improvements in coding style using flake8 and pylint to monitor and help enforce PEP8.

• Several performance improvements made calculations more efficient, especially for large problems, such as regional assessments or tall buildings investigated using the FEMA P-58 methodology.

• Several bugfixes and a large number of minor changes that make the engine more robust and easier to use.

• Update recommended Python version to 3.10 and other dependencies to more recent versions.

Changes in v3.0

• The architecture was redesigned to better support interactive calculation and provide a low-level integration across all supported methods. This is the first release with the new architecture. Frequent updates are planned to provide additional examples, tests, and bugfixes in the next few months.

• New assessment module introduced to replace control module:

  • Provides a high-level access to models and their methods
  • Integrates all types of assessments into a uniform approach
  • Most of the methods from the earlier control module were moved to the model module

• Decoupled demand, damage, and loss calculations:

  • Fragility functions and consequence functions are stored in separate files. Added new methods to the db module to prepare the corresponding data files and re-generated such data for FEMA P58 and Hazus earthquake assessments. Hazus hurricane data will be added in a future release.
  • Decoupling removed a large amount of redundant data from supporting databases and made the use of HDF and json files for such data unnecessary. All data are stored in easy-to-read csv files.
  • Assessment workflows can include all three steps (i.e., demand, damage, and loss) or only one or two steps. For example, damage estimates from one analysis can drive loss calculations in another one.

• Integrated damage and loss calculation across all methods and components:

  • This includes phenomena such as collapse, including various collapse modes, and irreparable damage.
  • Cascading damages and other interdependencies between various components can be introduced using a damage process file.
  • Losses can be driven by damages or demands. The former supports the conventional damage->consequence function approach, while the latter supports the use of vulnerability functions. These can be combined within the same analysis, if needed.
  • The same loss component can be driven by multiple types of damages. For example, replacement can be triggered by either collapse or irreparable damage.

• Introduced Options in the configuration file and in the base module:

  • These options handle settings that concern pelicun behavior;
  • general preferences that might affect multiple assessment models;
  • and settings that users would not want to change frequently.
  • Default settings are provided in a default_config.json file. These can be overridden by providing any of the prescribed keys with a user-defined value assigned to them in the configuration file for an analysis.

• Introduced consistent handling of units. Each csv table has a standard column to describe units of the data in it. If the standard column is missing, the table is assumed to use SI units.

• Introduced consistent handling of pandas MultiIndex objects in headers and indexes. When tabular data is stored in csv files, MultiIndex objects are converted to simple indexes by concatenating the strings at each level and separating them with a -. This facilitates post-processing csv files in pandas without impeding post-processing those files in non-Python environments.

• Updated the DL_calculation script to support the new architecture. Currently, only the config file input is used. Other arguments were kept in the script for backwards compatibility; future updates will remove some of those arguments and introduce new ones.

• The log files were redesigned to provide more legible and easy-to-read information about the assessment.

Changes in v2.6

• Support EDPs with more than 3 characters and/or a variable in their name. For example, SA_1.0 or SA_T1
• Support fitting normal distribution to raw EDP data (lognormal was already available)
• Extract key settings to base.py to make them more accessible for users.
• Minor bugfixes mostly related to hurricane storm surge assessment

License

pelicun is distributed under the BSD 3-Clause license, see LICENSE.

Acknowledgment

This material is based upon work supported by the National Science Foundation under Grants No. 1612843 2131111. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation.

Contact

Adam Zsarnóczay, NHERI SimCenter, Stanford University, [email protected]