Welcome to the EM-DAT Documentation

• 1: Introduction
• 2: Data Accessibility
• 3: Data Structure and Content Description
• 3.1: General Definitions and Concepts
• 3.2: Core Structure of the Database
• 3.3: EM-DAT Public Table
• 3.4: Disaster Classification System
• 3.5: Classification Glossary
• 3.5.1: Biological Hazards
• 3.5.2: Climatological Hazards
• 3.5.3: Extra-terrestrial Hazards
• 3.5.4: Geophysical Hazards
• 3.5.5: Hydrological Hazards
• 3.5.6: Meteorological Hazards
• 3.5.7: Technological Hazards
• 3.6: Hazard and Disaster Magnitude Units
• 3.7: Impact Variables
• 3.7.1: Human Impact Variables
• 3.7.2: Economic Impact Variables
• 3.8: Spatial Information and Geocoding
• 3.9: EM-DAT Sources of Information
• 4: Protocols
• 4.1: Entry Criteria
• 4.2: Encoding, Quality Control, and Validation Procedure
• 4.3: Economic Adjustment
• 5: Known Issues and Limitations
• 5.1: General Issues
• 5.2: Specific Biases
• 6: Additional Resources and Tutorials
• 6.1: Tutorials
• 6.1.1: Python Tutorial 1: Basic Operations and Plotting
• 6.1.2: Python Tutorial 2: Making Maps
• 6.2: Additional Resources
• 6.3: FAQ
• 7: About the EM-DAT Project
• 7.1: EM-DAT and CRED
• 7.2: EM-DAT Partners
• 7.3: Ways to Contribute
• 7.4: Contact
• 8: Legal Texts
• 8.1: Use Of EM-DAT Database Data and Derived Products
• 8.2: Database License Agreement (Commercial Use)
• 8.3: EM-DAT Attribution and Citation Policy

1 - Introduction

Why Document EM-DAT?

Since its creation in 1988, the Emergency Events Database (EM-DAT) has undergone many changes. Tracing these changes over time was a difficult task that required combing through the archives to reconstruct the history of the database structure, concepts, and definitions. This versioned documentation gives you a rigorous tool which you can use to track its evolution and access up-to-date information.

EM-DAT Documentation

EM-DAT documentation 2023.09 was the first versioned release. As the version number indicates, it was released in September 2023. The current version is 2026.09. The 2023.09 release introduced breaking changes in the public table format; a backup of the legacy public table is available under the same conditions.

Overview of EM-DAT

In 1988, the Centre for Research on the Epidemiology of Disasters (CRED), at the University of Louvain (UCLouvain), launched the Emergency Events Database (EM-DAT). EM-DAT was created with the initial support of the World Health Organization (WHO) and the Belgian government. From 1999 to the beginning of 2025, EM-DAT was supported by the Bureau for Humanitarian Assistance (BHA, previously the Office of US Foreign Disaster Assistance, or OFDA) within the United States Agency for International Development (USAID).

The initial objective of the database was to serve the purposes of humanitarian action at the national and international levels. Today, EM-DAT is also used to rationalize disaster preparedness and decision-making while providing an objective basis for vulnerability and risk assessment.

The EM-DAT database records mass disasters as well as their human and economic impacts at the country level (see Data Structure). The database contains core data on the occurrence and effects of disasters worldwide, with over 27,000 disaster records from 1900 to the present. The database is compiled from various sources of information, including UN agencies, non-governmental organizations, insurance companies, research institutes, and press agencies (see Sources).

How to Cite?

To acknowledge EM-DAT in documents, presentations, figures, or scientific publications, please follow the EM-DAT Attribution and Citation Policy. The policy defines the required elements of a proper citation and provides examples for each type of use, including how to cite the EM-DAT Archive and the EM-DAT descriptor article in scientific publications.

Highlighted Publications

Check our latest research papers:

Speybroeck, N., Noy, I., Kettner, A., Below, R., Messori, G., Thiery, W., Wathelet, V., Jézéquel, A., Delforge, D., & Van Niekerk, D. (2026). The Emergency Events Database, EM-DAT, must be preserved as a global public good. International Journal of Public Health, 71, Article 1609990. https://doi.org/10.3389/ijph.2026.1609990

Ronco, M., Tilloy, A., Corbane, C., Delforge, D., Feyen, L., Jäger, W. S., Matanó, A., Paprotny, D., Sibilia, A., Tiggeloven, T., & Ward, P. J. (2026). Compounding hazards increase flood economic losses across Europe. Nature Communications, 17(1), Article 6614. https://doi.org/10.1038/s41467-026-73248-0

Ronco, M., Bandelli, L., Bertolini, L., Consoli, S., Delforge, D., Spadaro, A., Verile, M., & Corbane, C. (2026). Disaster storylines and knowledge graphs from global news with large language models and retrieval-augmented generation. Scientific Data, 13(1), Article 689. https://doi.org/10.1038/s41597-026-07036-2

All CRED publications related to EM-DAT are accessible from our online repository.

Further Resources

For EM-DAT-derived datasets, other disaster occurrence and loss databases, software, and further reading, see the Additional Resources page.

2 - Data Accessibility

What are the Different EM-DAT Products?

• The living EM-DAT Public Data is the main data product, providing disaster impact details by event and country, updated weekly.
• The EM-DAT Archive is a validated backup of the EM-DAT Public Data shared to support open and reproducible scientific research.
• The EM-DAT HDX Country Profiles summarize yearly disaster data by type and country.

How to Download the EM-DAT Public Data?

The public EM-DAT records are accessible from our online data portal public.emdat.be. Access for non-commercial use is free after registration and subject to the Terms of Use. Access for Commercial Use is possible on the basis of an annual paid subscription (see Commercial License).

After registration and login, the data can be downloaded using the EM-DAT “Access Data” Tab or Toolbox (see the figure below) as a flat table in Microsoft Excel format (.xlsx). The EM-DAT database is described in the Data Structure and Content Description section. In particular, the Column Description section contains the description of each column of the public table found in the Excel file.

How to Download the EM-DAT Archive Data?

The EM-DAT Archive is available from the University of Louvain (UCLouvain) Dataverse. Hosted on a dataset repository under a CC BY-NC-ND 4.0 license, it adheres to FAIR principles. Therefore, it should be the preferred choice for conducting research with EM-DAT. The archived versions and the references to cite for each of them are listed in Additional Resources.

The Humanitarian Data Exchange (HDX) EM-DAT Country Profiles

In addition, annual data summaries per country (country profiles) are available for download on the Humanitarian Data Exchange platform (see the HDX documentation). These consist of aggregated figures, updated weekly, in the following format:

3 - Data Structure and Content Description

3.1 - General Definitions and Concepts

EM-DAT was designed in 1988 based on an anthropocentric vision of disasters and emergencies¹. It considers disasters to be events involving an unexpected and overwhelming harmful impact on human beings. Formally, EM-DAT’s definition of a disaster is:

EM-DAT inventories only disasters that fit its Entry Criteria. EM-DAT records both disasters triggered by natural hazards and technological disasters. The latter are unintentional accidents, and not situations of conflict, violence, or terrorism. For more details, we refer to the Disaster Classification System.

3.2 - Core Structure of the Database

The EM-DAT data model records disasters and maps their impacts at the country level. Accordingly, the EM-DAT database references disasters with a unique identifier (see DisNo. in Column Description). A disaster may affect more than one country. In EM-DAT, the country-specific information about the disaster is recorded as an “Impact.” The Impact is the level at which EM-DAT reports entries in the EM-DAT Public Table. To document each Impact, we cross-validate and select information from the sources that are available (see EM-DAT Sources and Protocols).

The figure below provides a visual representation of the foundational structure of the EM-DAT database’s relational model. On the left is the core structure, and on the right, a hypothetical case serves as an illustrative example. To enhance clarity, the relations have been streamlined, omitting certain tables such as dictionaries, focusing solely on the main connections and the underlying logic of the database.

Cardinality, or the nature of the relationship between entities, is denoted by numbers or the letter “N” on each link. For instance, “1, N” signifies a one-to-many relationship. A clear example of this is the link between the “Disaster” and “Impact” tables. This implies that a single disaster entry might correspond to multiple impacts across various countries. Consider a flood affecting multiple nations, such as Belgium and Germany; the connection between these entries is facilitated by the shared DisNo. attribute.

The primary classification of a disaster, such as its main type, is housed in the “Disaster” table. Consequently, all impacts linked by the same DisNo. will have a consistent primary disaster type. However, EM-DAT introduces a nuance with associated disasters linked to the “Impact” table. Unlike the main type, these associated disasters can vary by country. They represent secondary disasters that either result from or occur simultaneously with the main event. An example would be a landslide triggered by a primary flood event, which is considered to be an associated disaster. It’s important to note that associated disasters are optional, leading to their zero-to-many relationship with the “Impact” table.

Each source can provide data for any given variable within the Impact category. During the validation process (as detailed in Encoding, Quality Control, and Validation Procedure), one value is chosen for each variable, which is independent of other variables. Some figures presented in the EM-DAT Public Table are aggregates of various variables and might originate from multiple distinct sources. For example, figures for deaths and missing persons might be reported at different times after an event. After evaluating and selecting from the available data, the Total Deaths column might consolidate information from various sources.

3.3 - EM-DAT Public Table

Overview

The EM-DAT Public Table is a flat representation of EM-DAT data in a single downloadable table. Most impact variables are part of the public table (see Impact Variables). The public table provides a flat view of the general structure in which each record (row) corresponds to a disaster impacting a country.

Column Description

3.4 - Disaster Classification System

A Brief History of the EM-DAT Classification System

EM-DAT’s classification system originally started with a simple 20-type list¹. However, in 1992, CRED and other international stakeholders proposed a hierarchical classification system² that distinguishes natural and man-made disasters (described as technological disasters in EM-DAT). A further distinction was based on the timing of disasters: slow vs. rapid onset disasters.

In the 2000s, CRED collaborated with Munich Re and other stakeholders on a common classification system³. Since then, EM-DAT’s main classification system has followed the logic of referring to the hazard or event triggering the disaster. Consequently, some disasters were reclassified, and some types were removed from the primary classification system. This was the case for famines reclassified as drought for the most part⁴. However, “famine” offers more information than “drought.” Therefore, in order not to lose this added value, EM-DAT has adopted the secondary classification system of Associated Disasters. This describes disasters that coincide with or result from the primary type.

In 2014, the Integrated Research on Disaster Risk (IRDR) working group, which included CRED, established a new reference called the Peril Classification and Hazard Glossary⁵. This document is currently the primary reference for classifying natural hazards in EM-DAT, which divides them into six main groups: Geophysical, Hydrological, Meteorological, Climatological, Biological, and Extra-terrestrial. EM-DAT also includes more detailed subtypes.

Main Classification Tree

The main classification tree has four levels of depth, so disasters are divided into groups, subgroups, types, and subtypes, as presented in the EM-DAT Public Table columns. The two EM-DAT disaster groups are ‘Natural’ and ‘Technological’. The table below shows the complete tree for the ‘Natural’ and ‘Technological’ groups, with the occurrence for each subtype. Their corresponding definitions are available in the Classification Glossary.

Associated Disasters

In addition to the main classification system, EM-DAT makes it possible to refer to “associated disasters” to describe disaster events in more detail (see Associated Types in EM-DAT Public Table). They represent subsequent or co-occurring hazards that may have contributed to the disaster impact. These associated disasters may not fit into the main classification system and do not have a hierarchical structure. This additional tagging system allows for a better description of disaster events, particularly multi-hazard ones. The figure below shows the main associations found in the database.

About 15% of disaster entries in EM-DAT have an associated disaster type, and only 3.5% mention two or more associated types. The most common associations are floods with landslides (23% of associations), storms with floods (22%), and storms with landslides (8%). Earthquakes are sometimes associated with landslides (4%) and tsunamis (3%) when their damage is deemed negligible compared to ground movement damage.

3.5 - Classification Glossary

EM-DAT’s definitions related to the group of natural hazards mainly refer to the IRDR Peril Classification and Hazard Glossary¹ (see A Brief History of the EM-DAT Classification System). These are reported in the following sections by disaster subgroups.

EM-DAT definitions related to the technological group are listed separately in the Technological Hazards section. These are legacy definitions from the EM-DAT project and do not refer to a particular glossary serving as an international standard.

3.5.1 - Biological Hazards

3.5.2 - Climatological Hazards

3.5.3 - Extra-terrestrial Hazards

3.5.4 - Geophysical Hazards

3.5.5 - Hydrological Hazards

3.5.6 - Meteorological Hazards

3.5.7 - Technological Hazards

3.6 - Hazard and Disaster Magnitude Units

Some disaster types may have a reported magnitude in the Magnitude and Magnitude Scale columns of the EM-DAT Public Table. EM-DAT disaster magnitude scales vary depending on the disaster type. The table below specifies the magnitude property, unit, and exact value found in the Magnitude Scale column for the related disaster types. Note that the Magnitude Scale column is filled for all disasters of these types, even when no magnitude value is reported.

3.7 - Impact Variables

3.7.1 - Human Impact Variables

Five variables describe the human impact of disasters in the EM-DAT Public Table:

The reported total number of deaths (column Total Deaths) includes confirmed fatalities directly attributed to the disaster plus missing people whose whereabouts since the disaster are unknown and so they are presumed dead based on official figures.

Aside from fatalities, the number of injured people (column No. Injured) is entered when the term “injured” is written in the source. Any related word like “hospitalized” is considered as injured. If no precise number is given, such as “hundreds of injured,” 200 injured will be entered (although this figure is probably an underestimate).

The number of affected people (column No. Affected) is often mentioned and is widely used by different actors to convey the extent, impact, or severity of a disaster in non-spatial terms. If only the number of families affected or houses damaged are reported, the figure is multiplied by the average family size for the affected area (×5 for developing countries, ×3 for industrialized countries, according to the UNDP country classification).

Similarly, the indicator No. Homeless is mentioned whenever it is found in reports. If only the number of families that are homeless or houses that are destroyed are reported, the figure is multiplied by the average family size for the affected area (×5 for developing countries, ×3 for industrialized countries, according to the UNDP country classification).

Finally, the indicator Total Affected is the total of injured, affected, and homeless people. For each disaster and its corresponding sources, the population referred to in these statistics and the apportionment between injured, affected, homeless, and the total is carefully checked by CRED staff members.

Usually, at least the field Total Deaths or Total Affected is found in EM-DAT records as these numbers are involved as entry criteria. However, records often contain incomplete impact statistics (see Accounting Biases).

3.7.2 - Economic Impact Variables

Six variables describe the economic impact of disasters in the EM-DAT Public Table:

These six statistics are the same three statistics (Reconstruction Costs ('000 US$), Insured Damage ('000 US$), and Total Damage ('000 US$)) repeated to also provide an amount corrected for inflation, i.e., “Adjusted” (see Economic Adjustment). Damage and costs are converted and expressed in thousands of US dollars (‘000 US$).

Reconstruction costs are different from total damage as they must consider the current construction or purchase costs of goods, as well as the additional cost of prevention and mitigation measures designed to reduce damage from future disasters. Hence, when reconstruction costs are specified, they are usually greater than the total damage.

Insured damage is usually reported by reinsurance companies that publish figures about disaster losses, e.g., Munich Re, Swiss Re, or Aon. When insured damage is reported, the total damage is generally reported from the same source for consistency.

3.8 - Spatial Information and Geocoding

Country Codes (ISO-3)

In the EM-DAT Public Table, the ISO column indicates a 3-letter (alpha-3) code representing a specific country, e.g., “BEL” for Belgium. This code is presented according to the international standard ISO 3166 determined by the International Organization for Standardization (ISO). Due to historical changes in the countries’ denomination or boundaries, you may find country codes that are not found in the current ISO 3166 alpha-3 country codes. These extensions are listed in the table below.

Country codes are particularly useful to link the EM-DAT tabular data to a spatial layer using a Geographic Information System (GIS), spatial database, or geoprocessing programming library.

United Nations M49 Standard Country or Area Codes

The UN M49, also known as the Standard Country or Area Codes for Statistical Use (Series M, No. 49), is a set of area codes formulated by the United Nations for statistical use. This standard is curated and upheld by the United Nations Statistics Division. The UN M49 alpha-3 codes largely overlap with the ISO 3166 alpha-3 codes.

Since September 2023, the EM-DAT Public Table refers to the Country, Region, and Subregion names as found in the UN M49 standard. Codes used for the EM-DAT extensions to the ISO alpha-3 codes and UN M49 are listed in the table below.

ISO / UN M49 Extensions

GAUL 2015 Administrative Units

Between 2014 and 2025, EM-DAT relied on the Global Administrative Unit Layers (GAUL 2015) implemented by the Food and Agriculture Organization (FAO) to indicate in the Admin Units column of the EM-DAT Public Table which region is affected by the disaster, up to the Admin-2 level (e.g., districts). The mapping of EM-DAT disaster events at this higher level of geographical precision has only been completed for non-biological natural-hazard data since 2000.

The human and economic impact variables at the country level (Admin-0) are not disaggregated between regions at the Admin-1 or Admin-2 level. Hence, only the occurrence is available at a more precise administrative level, and the impact variables remain representative of the country level.

The geometries for GAUL 2015 are available as a Geopackage file.

GADM 4.1 Administrative Units

The GADM Database of Global Administrative Areas was selected to replace GAUL 2015 geocoded administrative units, from 2026 onwards. As with GAUL 2015 Administrative Units, geocoding remains limited to non-biological natural-hazard data since 2000. The primary geocoding workflow remains manual. GADM 4.1 geocoded units are assigned manually by database managers, using existing geocoding services, based on the Location column (see Public Table). However, as of January 2026, most GADM administrative units in EM-DAT for the 2000–2025 period are the result of a migration from the GAUL 2015 Administrative Units, and this migration is hereby documented to reflect potential precision losses inherent in the process.

GAUL 2015 to GADM 4.1 Migration

The complete procedure followed two steps:

1. The generation of a translation layer from GAUL to GADM. Based on geoprocessing, GAUL and GADM geometries were compared using geometric and topological indicators. Several strategies and outcomes are considered in the translation layer.
2. The optimal translation selection workflow. Among candidate GAUL-to-GADM translation options, the workflow follows a decision tree based on indicators associated with the translation layer, ultimately selecting GADM units.

Translation Methods and Strategies

Candidate translation options are identified using four methods, listed below in order of priority. The companion table provides an additional summary, including the labels that could be found in the GADM Admin Units column (see Public Table) to identify the matching method.

1. One-to-one matching using overlap indices, i.e., the Jaccard Index (JI). Labels jaccard1 and jaccard2 refer to units that have been matched using this method, either at the Admin-1 or Admin-2 level.

2. One-to-many matching using a selection strategy to reconstruct a composite GADM administrative footprint matching the original GAUL one with a list of GADM units. The quality of the composite “puzzle” footprint is also evaluated using the Jaccard Index (JI). The approach can be applied at the same administrative level (puzzle1_1 or puzzle2_2) or between GAUL Admin-1 and GADM Admin-2 (puzzle1_2). *

3. One-to-one matching using the discrete Hausdorff distance as an alternative indicator, primarily focused on the spatial proximity of points defined in nearby geometries. A buffer distance is used to identify nearby geometries. In contrast, JI is focused on overlap ratios. The routine is either applied at the Admin-1 (bufhaus1) or Admin-2 level (bufhaus2).

4. An upward fallback strategy, which consists of identifying a GADM unit containing the GAUL original one. The parent2_1 workflow selects a jaccard1 match for the GAUL parent unit when applicable. The within2_1 workflow identifies the upward GADM Admin-1 unit based on the area ratio of the contained GAUL Admin-2 unit.

* While we do not provide a jaccard1_2 label, a puzzle1_2 method resulting in one admin unit being identified is conceptually and methodologically equivalent.

As of January 2026, around 90% of the GAUL 2015 footprints have been matched to corresponding GADM 4.1 footprints, satisfying quality thresholds. The remaining 10% will be geocoded progressively, using the manual workflow.

See also: Additional Resources/Augmented EM-DAT Data

3.9 - EM-DAT Sources of Information

The EM-DAT database is compiled from various sources, including the United Nations, governmental and non-governmental agencies, insurance companies, research institutes, and the press. The table below lists the most common sources as of September 2023. For further inquiries on the data collection and selection process, see EM-DAT Protocols.

Sources are not displayed in the public records for legal reasons, to preserve the integrity of the services EM-DAT depends on, and to uphold neutrality, especially for politically sensitive data. For a quantitative overview of the sources used between 2000 and 2023, including the top 20 sources, their documentation routines, and how often each contributes to Total Deaths, Total Affected, and Total Damage, see Delforge et al. (2025)¹. Note that the set of sources evolves over time: some systematic sources have been discontinued, which may introduce biases (see Known Issues and Limitations).

4 - Protocols

4.1 - Entry Criteria

The EM-DAT definition of a disaster considers unintended hazards with a substantial impact unforeseen by a community (see General Definitions and Concepts). For management and operational purposes, EM-DAT has a set of entry criteria that specify what substantial impact means. EM-DAT disaster records related to natural and technological hazards meet at least one of the following entry criteria:

There are, however, secondary criteria, especially for past events where quantitative data were not available (e.g., “the worst disaster in a country or region” or “an event that resulted in considerable damage”).

4.2 - Encoding, Quality Control, and Validation Procedure

Source identification and collection can be facilitated and partially automated thanks to online services offered by specific sources (e.g., email alert systems, news feeds, or APIs). However, data collection and encoding are always supervised manually by the database manager. The database manager controls the sources that are selected, the classification of the event, its spatiotemporal delimitation, and the identification of the impact figures. Data encoding and validation in EM-DAT is a three-step process:

In addition, Automated Procedures and Constraints prevent or check for abnormal values or data formats in the database.

Daily Encoding

The database manager checks daily information using the preferred source list (see EM-DAT Sources). Whenever a new disaster is identified based on a source, it is added to the database. In the first stage, the event is not made public. It becomes so when an entry criterion is met and confirmed by at least two sources. The figures remain subject to changes. Any publication or modification made on a public disaster entry will be visible to the user after the weekly update routine. This routine is usually executed on Mondays but may be triggered by the manager if deemed necessary, e.g., in the case of faulty figures or typos.

The published impact variables could be selected from one or more sources. An event can therefore be validated from several sources of information. For example, the human impact can come from an OCHA report and the economic data from a reinsurance report, depending on specific expertise. If the figures differ between the sources, the database manager decides which ones to attribute to the disaster. The choice depends on several elements: the figure itself and the area and period to which it refers, the sources’ chronology, and their degree of reliability. Because this task is complex and case-dependent, there is no predetermined rule for selecting figures, and the database manager makes the final choice. Some examples of general but not systematic decision rules are illustrated in the box below.

The rules here are only informal and the database manager may disregard them. For example, suppose a report mentions 200 deaths and a news article says, “regional authorities estimate the death toll now stands at 243 dead and missing”. Although this is a newspaper article, the precision of the statement suggests that this figure is more reliable than the one in the official report.

Quality Control and Annual Validation

Quality control and annual validation are systematic checks of all the entries starting in a specific year. They typically take place at the beginning of the following year. During this validation, all the disasters that took place in the previous year are reviewed to consolidate the data, identify possible additional sources, and modify the published figures accordingly. In addition, the georeferencing, i.e., the more precise attribution of the disaster to GADM 4.1 Admin-1 or Admin-2 units (GAUL 2015 before 2026), is also finalized during the annual validation period (see GADM 4.1 Administrative Units).

Thematic Reviews

CRED periodically conducts thematic reviews of disasters to mitigate the database’s weaknesses (see Known Issues). This task involves systematically checking entries for a type of disaster over a given period or region. The revision can be a data analysis for further quality control, a systematic review of the scientific literature, or a comparison with other existing databases.

From September 2023 onward, these substantial database content updates are planned to be notified on the EM-DAT website and in the documentation release notes for tracking purposes (see Introduction). The Entry Date and Last Update columns have also been introduced in the EM-DAT Public Table.

Automated Procedures and Constraints

In 2023, the constraints of the EM-DAT database were strengthened. These constraints define the domain of values that can be encoded and, if correctly set, can prevent encoding errors in the value or its format. In addition, automated procedures check for anomalies that are likely to be an error, i.e., those without enough certainty to be a constraint but with sufficient likelihood to be notified to the database manager for verification.

Currently, constraints and automated routines check for the consistency of date and time fields, latitude and longitude values, and hazard magnitude values. These procedures are implemented incrementally each time an error that could be prevented is discovered. Hence, by detecting and reporting issues, users may contribute to developing these routines and improving EM-DAT data quality.

Issue Reporting

CRED encourages any external initiative that can help us improve our data quality. Any problem with the data content (such as errors or missing data) can be reported to CRED by email using the contact address (see Send Us an Email). For missing entries, you should be aware that CRED only publishes disasters for which reliable figures are available and when these are corroborated by at least two sources (see Daily Encoding). CRED reserves the right to modify the data according to subsequent notifications. For issues concerning specific disaster events, the problem should be explained and notified with a list including the related DisNo. values (see Column Description).

For less specific issues, users who have analyzed the quality of EM-DAT are encouraged to share their reports or scientific studies with the EM-DAT team to improve the database (see Contributing).

4.3 - Economic Adjustment

All economic damage in EM-DAT is expressed in thousands of US$ (see Column Description and Economic Impact Variables). For each disaster, the registered figure corresponds to the value of the damage at the moment of the event. If a source reports damage in another currency, it is first converted to US$ before it is entered in EM-DAT using a currency converter (e.g., the OANDA Currency Converter), at the exchange rate when the damage occurred.

The adjusted economic losses in EM-DAT provide a monetary value in US$ that has been adjusted for inflation (see Column Description). The adjustment is linearly proportional to the OECD Consumer Price Index (CPI) provided by the Organisation for Economic Co-operation and Development (OECD). The CPI reflects the change in prices of a basket of goods and services that are typically purchased by specific groups of households. EM-DAT relies on the total CPI, i.e., including both food and energy products in the basket, defined for the USA.

In practice, OECD provides the US CPI with 2015 as the base year (\(CPI_{2015}=100\), see Consumer prices for G20). In EM-DAT, the CPI is rescaled to use the last year as a reference, e.g., \(CPI_{2021}=100\). The rescaling is performed when the OECD CPI value becomes available. The economic losses given for the current year are therefore not adjusted and not reported in the column Total Damage, Adjusted ('000 US$).

5 - Known Issues and Limitations

EM-DAT is the only comprehensive, free-access disaster loss database with effective global coverage¹. However, it has limitations due to the limited number of sources and to how effectively disasters are reported worldwide. This can lead to biases in the data over which CRED may have limited control and which may be overlooked in the literature². No current impact database is completely accurate. Nevertheless, EM-DAT remains a key resource for understanding disaster events and impacts. The United Nations emphasizes the importance of global improvements in documenting disasters in global agendas such as the Sendai Framework for Disaster Risk Reduction.

Understanding the limitations of a dataset such as EM-DAT is of paramount importance for those who wish to adequately use the data and mitigate its weaknesses for the following purposes: disaster risk management, emergency planning, scientific research, and raising public awareness. The General Issues page describes broad data quality concerns, and the Specific Biases page details the time, hazard-related, threshold, accounting, and geographic biases in EM-DAT.

5.1 - General Issues

Three types of data quality issues can be considered:

A cross-comparison of EM-DAT with a local database and/or a disaster-specific database can help identify Issue 1 (e.g., Koç & Thieken, 2018¹; Lin et al., 2021²). For an account of missing values for existing events, we refer to Jones et al. 2022³ and the section on Accounting Biases. Issue 3 is partially related to the data collection sources, protocols, or reporting systems generally used by different databases.

Data quality issues within EM-DAT are related to the data collection protocols from dedicated sources. EM-DAT’s completeness reflects the coverage of its sources. Since source reporting has improved over the years, EM-DAT data coverage has improved significantly over the last 30 to 40 years. Nevertheless, gaps and quality issues remain. EM-DAT protocols are meant to guide the way information is monitored and collected from sources. However, no universally applied protocol ensures that different sources report disaster impact and losses using the same guidelines to define, for instance:

• the beginning and end of disaster events.
• the geographical footprint of a disaster.
• impact variables such as deaths (in particular, when computed based on excess mortality), affected people, or economic costs.
• the disaster type selected by the sources.

Some references illustrate the issues and challenges related to collecting and maintaining a disaster database, e.g., Guha-Sapir & Misson 1992⁴, Kron et al. 2012⁵, and Wirtz et al. 2014⁶.

To some extent, EM-DAT owes its popularity to its simplicity. It reports disaster events as rows in an Excel table. However, this simplicity comes at the cost of conceptual limitations in dealing with complex and compound events and situations. In such cases, as exemplified in the box below, EM-DAT will probably report the disaster in the same way as the source which presented it. The EM-DAT database manager can only choose between the figures reported by different sources (see Daily Encoding). However, no model is involved in correcting differences in reporting protocols because this task goes beyond the information monitoring conducted at CRED by the EM-DAT team.

Such biases that result from differences in the impact reporting systems were generally referred to by Gall et al. 2009⁷ as systemic biases. Some studies point to systemic biases by highlighting that EM-DAT does not correlate well with other databases (e.g., Moriyama et al., 2018⁸; Panwar & Sen, 2020⁹). In their article, Gall et al. 2009⁷ cover five other types of biases: time, hazard-related, threshold, accounting, and geographic biases. These are illustrated on the Specific Biases page.

5.2 - Specific Biases

Time Bias

Time biases result from unequal reporting quality and coverage over time¹. The figure below shows the occurrence of disasters in EM-DAT. The figure shows a significant increase that starts in the 1960s. This increase coincides with the creation of OFDA. In 1973, OFDA started compiling disaster data, and CRED was created². In 1988, CRED took over the disaster database and created EM-DAT. In the meantime, communication technologies have improved, with the first personal computers and satellites appearing in the 1970s and the advent of the World Wide Web in the 1990s (see also History of EM-DAT).

Technologies and initiatives can be considered responsible for the dominant trend observed. Therefore, it is challenging to draw conclusions about the actual drivers of disasters such as climate change, population growth, or disaster risk management³. Accordingly, excluding pre-2000 data from trend analyses based on EM-DAT is strongly recommended. From September 2023 onward, CRED refers to pre-2000 data as Historic data in the EM-DAT Public Table.

Hazard-related Biases

Hazard-related biases result from unequal reporting quality and coverage for different hazard types¹. For example, in EM-DAT, data related to biological hazards (e.g., epidemics) and extreme temperature hazards (e.g., heat waves) are less covered, and their coverage is of lower quality. Some hazard-related biases are illustrated in the Accounting Biases and Geographic Biases sections.

Threshold Biases

Threshold biases result from unequal reporting quality and coverage for disasters of different magnitudes¹. High-impact disasters attract more attention, resulting in better media coverage and reporting. This could lead to threshold biases in EM-DAT. The EM-DAT entry criteria introduce a kind of threshold bias, as shown in the figure below regarding disaster mortality, while some studies have shown locally that small disasters may have a high cumulative impact, e.g.⁴. Regarding disasters that fit EM-DAT’s entry criteria, it is fair to assume that disasters close to the entry criteria are more likely to be missing than high-impact disasters. However, as shown in the figure below, the cumulative mortality associated with low-mortality events exceeds the cumulative impact of higher-mortality events.

Accounting Biases

Accounting biases result from unequal reporting quality and coverage for different impact variables¹. For instance, in EM-DAT, the economic losses are, on average, less frequently reported than the human impact variables, which may also depend on the hazard type (see the figure below). Furthermore, insured damages are naturally more reported than uninsured damages, which creates a geographic bias where there is a lack of insurance coverage, as in Africa. For droughts, EM-DAT fails to capture the associated mortality because it is overlooked as an indirect impact, as evidenced by the UNDRR 2021 Global Assessment Report on Droughts.

Besides, it cannot be assumed that because an impact is reported in EM-DAT, there is no accounting bias. In general, direct impacts are often reported by EM-DAT sources, while indirect impact estimates are less available. For instance, direct deaths for a flood event correspond to the number of fatalities occurring during the event, while indirect deaths result from disease outbreaks due to deteriorated sanitary conditions. Yet, indirect mortality is sometimes higher than direct mortality, for example⁵.

Geographic Biases

Geographic biases result from unequal reporting quality and coverage across space¹. In general, EM-DAT has a relatively worse coverage for sub-Saharan Africa regarding the occurrence and the accounting of impact variables⁶. Any disaster type may be subject to geographic biases in EM-DAT as there may be discrepancies between reporting systems from one country to another (see General Issues).

This issue is particularly pronounced regarding heat waves⁷, as shown in the figure below. Heat waves are often overlooked in sub-Saharan Africa⁸. About a third (32%) of the 275 heatwave events recorded in EM-DAT between 2000 and 2024 occurred in seven countries: Japan, India, Pakistan, the USA, and Western European countries (Belgium, France, and the United Kingdom).

References

6 - Additional Resources and Tutorials

6.1 - Tutorials

6.1.1 - Python Tutorial 1: Basic Operations and Plotting

This tutorial shows basic examples on how to load, handle, and plot the EM-DAT data using the pandas Python data analysis package and the matplotlib charting library.

Note: The Jupyter Notebook version of this tutorial is available on the EM-DAT Python Tutorials GitHub Repository.

Import Modules

Let us import the necessary modules and print their versions. For this tutorial, we used pandas v.2.1.1 and matplotlib v.3.8.3. If your package versions are different, you may have to adapt this tutorial by checking the corresponding package documentation.

import pandas as pd #data analysis package
import matplotlib as mpl
import matplotlib.pyplot as plt #plotting library
for i in [pd, mpl]:
    print(i.__name__, i.__version__)

pandas 2.1.1
matplotlib 3.8.3

Load EM-DAT

To load EM-DAT:

• Download the EM-DAT data at https://public.emdat.be/ (registration is required, see the EM-DAT Documentation page on Data Accessibility);
• Use the pd.read_excel method to load and parse the data into a pd.DataFrame object;
• Check if the data has been successfully parsed with the pd.DataFrame.info method.

Notes:

1. You may need to install the openpyxl package or another engine to make it possible to read the data.
2. Another option is to export the .xlsx file into a .csv, and use the pd.read_csv method;
3. If not in the same folder as the Python code, replace the filename with the relative path or the full path, e.g., E:/MyDATa/public_emdat_2024-01-08.xlsx

#!pip install openpyxl
df = pd.read_excel('public_emdat_2024-01-08.xlsx') # <-- modify file name or path
df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 15560 entries, 0 to 15559
Data columns (total 46 columns):
 #   Column                                     Non-Null Count  Dtype  
---  ------                                     --------------  -----  
 0   DisNo.                                     15560 non-null  object 
 1   Historic                                   15560 non-null  object 
 2   Classification Key                         15560 non-null  object 
 3   Disaster Group                             15560 non-null  object 
 4   Disaster Subgroup                          15560 non-null  object 
 5   Disaster Type                              15560 non-null  object 
 6   Disaster Subtype                           15560 non-null  object 
 7   External IDs                               2371 non-null   object 
 8   Event Name                                 4904 non-null   object 
 9   ISO                                        15560 non-null  object 
 10  Country                                    15560 non-null  object 
 11  Subregion                                  15560 non-null  object 
 12  Region                                     15560 non-null  object 
 13  Location                                   14932 non-null  object 
 14  Origin                                     3864 non-null   object 
 15  Associated Types                           3192 non-null   object 
 16  OFDA Response                              15560 non-null  object 
 17  Appeal                                     15560 non-null  object 
 18  Declaration                                15560 non-null  object 
 19  AID Contribution ('000 US$)                490 non-null    float64
 20  Magnitude                                  3356 non-null   float64
 21  Magnitude Scale                            9723 non-null   object 
 22  Latitude                                   1809 non-null   float64
 23  Longitude                                  1809 non-null   float64
 24  River Basin                                1197 non-null   object 
 25  Start Year                                 15560 non-null  int64  
 26  Start Month                                15491 non-null  float64
 27  Start Day                                  14068 non-null  float64
 28  End Year                                   15560 non-null  int64  
 29  End Month                                  15401 non-null  float64
 30  End Day                                    14132 non-null  float64
 31  Total Deaths                               12485 non-null  float64
 32  No. Injured                                5694 non-null   float64
 33  No. Affected                               7046 non-null   float64
 34  No. Homeless                               1312 non-null   float64
 35  Total Affected                             11508 non-null  float64
 36  Reconstruction Costs ('000 US$)            33 non-null     float64
 37  Reconstruction Costs, Adjusted ('000 US$)  29 non-null     float64
 38  Insured Damage ('000 US$)                  691 non-null    float64
 39  Insured Damage, Adjusted ('000 US$)        683 non-null    float64
 40  Total Damage ('000 US$)                    3070 non-null   float64
 41  Total Damage, Adjusted ('000 US$)          3020 non-null   float64
 42  CPI                                        15056 non-null  float64
 43  Admin Units                                8336 non-null   object 
 44  Entry Date                                 15560 non-null  object 
 45  Last Update                                15560 non-null  object 
dtypes: float64(20), int64(2), object(24)
memory usage: 5.5+ MB

Example 1: Japan Earthquake Data

Filtering

Let us focus on the EM-DAT earthquakes in Japan from the years 2000 to 2003 and create a suitable filter utilizing the EM-DAT columns Disaster Type, ISO and Start Year.

For simplicity, let’s retain only the columns Start Year, Magnitude, and Total Deaths and display the first five entries using the pd.DataFrame.head method.

Note: For further details about the columns, we refer to the EM-DAT Documentation page EM-DAT Public Table.

eq_jpn = df[
    (df['Disaster Type'] == 'Earthquake') &
    (df['ISO'] == 'JPN') &
    (df['Start Year'] < 2024)
][['Start Year', 'Magnitude', 'Total Deaths', 'Total Affected']]
eq_jpn.head(5)

Grouping

Let us group the data to calculate the number of earthquake events by year and plot the results.

• Use the groupby method to group based on one or more columns in a DataFrame, e.g., Start Year;
• Use the size method as an aggregation method (or count).
• Plot the results using the pd.DataFrame.plot method.

Note: The count method provides the total number of non-missing values, while size gives the total number of elements (including missing values). Since the field Start Year is always defined, both methods should return the same results.

eq_jpn.groupby(['Start Year']).size().plot(kind='bar', ylabel='Count')

<Axes: xlabel='Start Year', ylabel='Count'>

Customize Chart

The pandas library relies on the matplotlib package to draw charts. To have more flexibility on the rendered chart, let us create the figure using the imported plt submodule.

# Group earthquake data by 'Start Year' and count occurrences
eq_cnt = eq_jpn.groupby(['Start Year']).size()

# Initialize plot with specified figure size
fig, ax = plt.subplots(figsize=(7, 2))

# Plot number of earthquakes per year
ax.bar(eq_cnt.index, eq_cnt)

# Set axis labels and title
ax.set_xlabel('Year')
ax.set_ylabel('N° of Earthquakes')
ax.set_yticks([0, 1, 2, 3])  # Define y-axis tick marks
ax.set_title('EM-DAT Earthquakes in Japan (2000-2023)')

Text(0.5, 1.0, 'EM-DAT Earthquake in Japan (2000-2023)')

Example 2: Comparing Regions

Let us compare earthquake death toll by continents. As before, we filter the original dataframe df according to our specific needs, including the Region column.

eq_all = df[
    (df['Disaster Type'] == 'Earthquake') &
    (df['Start Year'] < 2024)
][['Start Year', 'Magnitude', 'Region', 'Total Deaths', 'Total Affected']]
eq_all.head(5)

In this case,

• Use the groupby method to group based on the Region column;
• Use the sum method for the Total Deaths field as aggregation method;
• Plot the results easilly using the pd.DataFrame.plot method.

eq_sum = eq_all.groupby(['Region'])['Total Deaths'].sum()
eq_sum

Region
Africa        5863.0
Americas    229069.0
Asia        548766.0
Europe         783.0
Oceania        641.0
Name: Total Deaths, dtype: float64

Finally, let us make an horizontal bar chart of it using matplotlib. In particular,

• use the ax.ticklabel_format method to set the x axis label as scientific (in thousands of deaths);
• use the ax.invert_yaxis to display the regions in alphabetical order from top to bottom.

fig, ax = plt.subplots(figsize=(4,3))
ax.barh(eq_sum.index, eq_sum)
ax.set_xlabel('Total Earthquake Deaths')
ax.ticklabel_format(style='sci',scilimits=(3,3),axis='x')
ax.invert_yaxis()
ax.set_title('EM-DAT Earthquake Deaths by Regions')

Text(0.5, 1.0, 'EM-DAT Earthquake Deaths by Regions')

Example 3: Multiple Grouping

At last, let us report the earthquake time series by continents. To avoid the creation of a ['Region', 'Start Year'] multiindex for future processing, we set the argument as_index to False. As such, Region and Start Year remain columns.

eq_reg_ts = eq_all.groupby(
    ['Region', 'Start Year'], as_index=False
)['Total Deaths'].sum()
eq_reg_ts

97 rows × 3 columns

Next, we apply the pivot method to restructure the table in a way it could be plot easilly.

eq_pivot_ts = eq_reg_ts.pivot(
    index='Start Year', columns='Region', values='Total Deaths'
)
eq_pivot_ts.head()

ax = eq_pivot_ts.plot(kind='bar', width=1, figsize=(6,3))
ax.set_ylabel('Total Deaths')
ax.set_title('EM-DAT Earthquake Deaths by Regions')

Text(0.5, 1.0, 'EM-DAT Earthquake Deaths by Regions')

In order to be able to visualize the data in more details, let us make a subplot instead by setting the subplot argument to True within the plot method.

ax = eq_pivot_ts.plot(kind='bar', subplots=True, legend=False, figsize=(6,6))
plt.tight_layout() # <-- adjust plot layout

We have just covered the most common manipulations applied to a pandas DataFrame containing the EM-DAT data. To delve further into your analyses, we encourage you to continue your learning of pandas and matplotlib with the many resources available online, starting with the official documentation.

If you are interested in learning the basics of making maps based on EM-DAT data, you can also follow the second EM-DAT Python Tutorial.

6.1.2 - Python Tutorial 2: Making Maps

If you have followed the first EM-DAT Python Tutorial 1 or are already familiar with pandas and matplotlib, this second tutorial will show you basic examples on how to make maps with the EM-DAT data using the geopandas Python package.

Note: The Jupyter Notebook version of this tutorial is available on the EM-DAT Python Tutorials GitHub Repository.

Import Modules

Let us import the necessary modules and print their versions. For this tutorial, we used pandas v.2.1.1, geopandas v.0.14.3, and matplotlib v.3.8.3. If your package versions are different, you may have to adapt this tutorial by checking the corresponding package documentation.

import pandas as pd #data analysis package
import geopandas as gpd
import matplotlib as mpl
import matplotlib.pyplot as plt #plotting library
for i in [pd, gpd, mpl]:
    print(i.__name__, i.__version__)

pandas 2.2.1
geopandas 0.14.3
matplotlib 3.8.3

Creating a World Map

To create a world map, we need the EM-DAT data and a shapefile containing the country geometries.

EM-DAT: We download and load the EM-DAT data using pandas.

Country Shapefile: We download a country shapefile from Natural Earth Data. For a world map, we download the low resolution 1:110m Adimin 0 - Countries (last accessed: March 10, 2024) and unzip it.

Load and Filter EM-DAT

Let us load EM-DAT and filter it to make a global map of Earthquake disasters between 2000 and 2023. We calculate the number of unique identifiers (DisNo.) per country (ISO) We refer to the standard ISO column instead of the Country column of the EM-DAT Public Table to be able to make a join with the country shapefile.

df = pd.read_excel('public_emdat_2024-01-08.xlsx')
earthquake_counts = df[
    (df['Disaster Type'] == 'Earthquake') &
    (df['Start Year'] < 2024)
].groupby('ISO')["DisNo."].count().reset_index(name='EarthquakeCount')
earthquake_counts

91 rows × 2 columns

Load the Country Shapefile

We use the gpd.read_file method to load the country shapefile and parse it into a geodataframe. A geodataframe is similar to a pandas dataframe, extept that has a geometry column.

We provide the filename argument, which is either a file name if located in the same directory than the running script, or a relative or absolute path, if not. In our case the shapefile with the .shp extension is located in the ne_110m_admin_0_countries folder.

Since the geodataframe contains 169 columns, we only keep the two column that we are interrested in, i.e., ISO_A3 and geometry.

gdf = gpd.read_file ('ne_110m_admin_0_countries/ne_110m_admin_0_countries.shp') # <-- Change path if necessary
gdf = gdf[['ISO_A3', 'geometry']]
gdf

Cannot find header.dxf (GDAL_DATA is not defined)

177 rows × 2 columns

Important Notice: Above, some geometries do not have a ISO code, such as the one at row 174. Below, you will see that some ISO in EM-DAT are not matched with a geometries. Beyond this basic tutorial, we advice to carefully evaluate these correspondance and non-correspondance between ISO codes and to read the EM-DAT Documentation about ISO codes.

Join the Two Datasets

Let us merge the two dataset with an outer join, using the merge method. We prefer an outer join to keep the geometries of countries for which EM-DAT has no records.

earthquake_counts_with_geom = gdf.merge(
    earthquake_counts, left_on='ISO_A3', right_on='ISO', how='outer')
earthquake_counts_with_geom

190 rows × 4 columns

Make the Map

To make the map, we use the geopandas built-in API, through the plot method built on the top of matplotlib. Below, we show an hybrid plotting approach and first create an empty figure fig and ax object with matplotlib before passing the ax object as an argument within the plot method. This approach gives more control to users familiar with matplotlib to further customize the chart.

fig, ax = plt.subplots(figsize=(8,3))
earthquake_counts_with_geom.plot(
    column='EarthquakeCount',
    ax=ax,
    cmap='Reds',
    vmin=1,
    legend=True,
    legend_kwds={"label": "Nb of EM-DAT Earthquake"},
    missing_kwds= dict(color = "lightgrey",)
)
_ = ax.set_xlabel('Lon.')
_ = ax.set_ylabel('Lat.')

Creating a Map at Admin Level 1

We can create a more detailed map using the Admin Units column in the EM-DAT Public Table. This column contains the identifiers of administrative units of level 1or 2 as defined by the Global Administrative Unit Layer (GAUL) for country impacted by non-biological natural hazards.

Similarly to the country map, we need to download a file containing GAUL geometries. The file corresponds to the last version of GAUL published in 2015. In this tutorial, we will focus on Japanese earthquake occurrence in EM-DAT.

Note: the file size is above 1.3Go and requires a performant computer to process in Python. Using a Geographical Information Software (GIS) for the preprocessing is another option.

Load the Admin Units Geopackage

The file is a geopackage .gpkg that contains multiple layers. Let us first describe these layers with the fiona package, which is a geopandas dependency.

import fiona
print(fiona.__name__, fiona.__version__)
for layername in fiona.listlayers('gaul2014_2015.gpkg'):
    with fiona.open('gaul2014_2015.gpkg', layer=layername) as src:
        print(layername, len(src))

fiona 1.9.5
level2 38258
level1 3422
level0 277

• The level0 layer contains the country geometries defined in GAUL.
• Here, we make a map at the level1.
• Still, we need to load the administrative level2 because the Admin Units column may refer to Admin 2 levels without mentionning the corresponding Admin 1 level.
• Given the high size the admin 2 layer, we filter the data about Japan and overwrite our geodataframe variable to save memory.

gaul_adm2 = gpd.read_file ('gaul2014_2015.gpkg', layer='level2') 
gaul_adm2 = gaul_adm2[gaul_adm2['ADM0_NAME'] == 'Japan']
gaul_adm2.info()

<class 'geopandas.geodataframe.GeoDataFrame'>
Index: 3348 entries, 23205 to 26552
Data columns (total 13 columns):
 #   Column      Non-Null Count  Dtype   
---  ------      --------------  -----   
 0   ADM2_CODE   3348 non-null   int64   
 1   ADM2_NAME   3348 non-null   object  
 2   STR2_YEAR   3348 non-null   int64   
 3   EXP2_YEAR   3348 non-null   int64   
 4   ADM1_CODE   3348 non-null   int64   
 5   ADM1_NAME   3348 non-null   object  
 6   STATUS      3348 non-null   object  
 7   DISP_AREA   3348 non-null   object  
 8   ADM0_CODE   3348 non-null   int64   
 9   ADM0_NAME   3348 non-null   object  
 10  Shape_Leng  3348 non-null   float64 
 11  Shape_Area  3348 non-null   float64 
 12  geometry    3348 non-null   geometry
dtypes: float64(2), geometry(1), int64(5), object(5)
memory usage: 366.2+ KB

The Admin 2 geodataframe has 12 columns describing the 3348 level 2 administrative units in Japan.

Filter EM-DAT Data

df_jpn = df[
    (df['ISO'] == 'JPN') &
    (df['Disaster Type'] == 'Earthquake') &
    (df['Start Year'] < 2024)
][['DisNo.', 'Admin Units']]
df_jpn

Note: The last two events were not geolocated at a higher administrative levels.

Convert Admin 2 units to Admin 1 units

We create a python function, json_to_amdmin1, to extract the administrative level 1 codes from the Admin Units column of EM-DAT, based on the ADM1_CODE and ADM2_CODE of the Japan geodataframe.

import json

def json_to_admin1(json_str, gdf):
    """
    Convert a JSON string to a set of administrative level 1 codes.

    Parameters
    ----------
    json_str
        A JSON string representing administrative areas, or None.
    gdf
        A GeoDataFrame containing administrative codes and their corresponding 
        levels.

    Returns
    -------
    A set of administrative level 1 (ADM1) codes extracted from the input JSON.

    Raises
    ------
    ValueError
        If the administrative code is missing from the input data or ADM2_CODE 
        not found in the provided GeoDataFrame.

    """
    adm_list = json.loads(json_str) if isinstance(json_str, str) else None
    adm1_list = []
    if adm_list is not None:
        for entry in adm_list:
            if 'adm1_code' in entry.keys():
                adm1_code = entry['adm1_code']
            elif 'adm2_code' in entry.keys():
                gdf_sel = gdf[gdf['ADM2_CODE'] == entry['adm2_code']]
                if not gdf_sel.empty:
                    adm1_code = gdf_sel.iloc[0]['ADM1_CODE']
                else:
                    raise ValueError(
                        'ADM2_CODE not found in the provided GeoDataFrame.'
                    )
            else:
                raise ValueError(
                    'Administrative code is missing from the provided data.'
                )
            adm1_list.append(adm1_code)
    return set(adm1_list)

We apply the function to all elements of the Admin Units column.

df_jpn.loc[:, 'Admin_1'] = df_jpn['Admin Units'].apply(
    lambda x: json_to_admin1(x, gaul_adm2))

df_jpn[['Admin Units', 'Admin_1']] 

Count Earthquakes per Admin 1 Units

The can be done applying the explode method on the new Admin_1 column. The method will add rows based on the number of Admin 1 we have in each set inside the Admin_1 column. Then the counting can be performed using the former groupby approach.

count_per_adm1 = df_jpn.explode('Admin_1').groupby(
    'Admin_1')['DisNo.'].count().rename('EQ Count')
count_per_adm1.head()

Admin_1
1651    1
1652    4
1653    4
1654    1
1655    1
Name: EQ Count, dtype: int64

Recreate the Admin 1 Layer

Since the Japan geodataframe contains the admin2 geometries, we could load the Admin 1 layer or simply dissolve the geometries based on the ADM1_CODE column. The geopandas package is equipped with the dissolve method.

gdf_jpn_adm1 = gaul_adm2.dissolve(by='ADM1_CODE') 
gdf_jpn_adm1.plot()

<Axes: >

Join the Two Datasets

Again, we can use the merge method to join the datasets together, here, based on their index.

gdf_jpn_adm1_merged = gdf_jpn_adm1.merge(count_per_adm1, 
                                         left_index=True, 
                                         right_index=True, 
                                         how='outer')

Make the Map

fig, ax = plt.subplots()

gdf_jpn_adm1_merged.plot(
    column='EQ Count', cmap='YlOrRd',
    linewidth=0.8, ax=ax,edgecolor='0.8',
    legend=True, 
    legend_kwds={'label': "Earthquake Count in EM-DAT"}
)
_ = ax.set_xlabel('Longitude')
_ = ax.set_ylabel('Latitude')

We have just covered the basics on how to join the EM-DAT pandas DataFrame with a geopandas GeoDataFrame to make maps. To delve further into your analyses, we encourage you to continue your learning of geopandas, matplotlib, or, in particular, cartopy for more advanced map customization, with the many resources available online.

6.2 - Additional Resources

Unless stated otherwise, these resources were neither developed nor are they maintained by the EM-DAT team. Please contact the corresponding authors with any questions.

EM-DAT Archive

The EM-DAT Archive, hosted on the University of Louvain (UCLouvain) Dataverse and maintained by the EM-DAT team, preserves official versions of the EM-DAT public data to support open and reproducible research. Each version includes the EM-DAT Public Table, its column description, and a PDF copy of this documentation, and is released under a CC BY-NC-ND 4.0 license (see also Data Accessibility).

Version 1.1 contains the same data files as version 1.0. When using an archived version, please cite the corresponding reference:

• Version 2.0: Delforge, D., Below, R., Wathelet, V., Alonso, A., & Speybroeck, N. (2026). The 2024 Emergency Events Database Archive: The year in review and historical data updates [Preprint]. Zenodo. https://doi.org/10.5281/zenodo.20071134
• Versions 1.0 and 1.1: Delforge, D., Wathelet, V., Below, R., Lanfredi Sofia, C., Tonnelier, M., van Loenhout, J. A. F., & Speybroeck, N. (2025). EM-DAT: The Emergency Events Database. International Journal of Disaster Risk Reduction, 124, Article 105509. https://doi.org/10.1016/j.ijdrr.2025.105509

Software

• EM-VIEW: A Streamlit Dashboard to explore and analyze the EM-DAT Public Table. Also available as a GitHub repo for customization. Learn more from its initial announcement.
• EM-TEST: An open-source testing framework, built on the pandas and pandera Python packages, to validate EM-DAT public data files and check the compatibility between versions. Developed by the EM-DAT team and archived on Zenodo.

Augmented EM-DAT Data

• Geocoded Disasters (GDIS): an augmented dataset with geocoding, based on the GADM data, for 9,924 disasters between 1960 and 2018 (Rosvold & Buhaug, 2021).
• Flood Disasters (FLODIS): an augmented dataset for flood disasters (2000–2018), based on Tellman et al. (2021) and IDMC data, that links EM-DAT to hazard footprints, exposed and displaced population data (Mester et al., 2023).
• Geo-Disasters: geocoding of climate-related events in EM-DAT (Teber et al., 2026).
• Disaster Storylines and Knowledge Graphs: storylines and knowledge graphs for over 3,000 EM-DAT disasters (2014–2024), extracted from global news of the Europe Media Monitor using large language models and retrieval-augmented generation (Ronco et al., 2026).

Other Disaster Occurrence and Loss Datasets

• Wikimpacts 1.0: a global climate impact database built by extracting information from Wikipedia with large language models, covering 2,726 events from 1034 to 2024 at the event, national, and sub-national levels (Li et al., 2026).
• ROUGE: the Red Cross Operations Unified Global Emergency database, documenting socio-economic impacts of natural hazards at national and sub-national levels, extracted with large language models from operational reports of the International Federation of Red Cross and Red Crescent Societies (IFRC), with broad coverage of the Global South (Hasbini et al., 2026).
• HANZE v2.1: the Historical Analysis of Natural HaZards in Europe database, covering 2,521 riverine, pluvial, coastal, and compound flood events in 42 European countries from 1870 to 2020, with fatalities, affected persons, inundated areas, and economic losses (Paprotny et al., 2024).
• Risk Data Hub – Disaster Losses: a curated European dataset of historical disaster losses since 1980, harmonized from multiple sources by the Joint Research Centre (JRC) of the European Commission, with unique event identifiers at the national and European levels. The data can be explored through the Risk Data Hub dashboards and accessed through the Risk Data Hub API under a CC BY 4.0 license.

Other Useful Resources

• Worldwide disaster loss and damage databases: a systematic review comparing available disaster loss databases (Mazhin et al., 2021).
• Natural Disasters: a comprehensive global overview of natural hazards and disasters on Our World in Data (Ritchie et al., 2022).
• A note on missing values and imputations summarizing insights from Jones et al. (2022) and Jones et al. (2023).

6.3 - FAQ

Select a question to show its answer.

General

Finding and Interpreting Records

Access and Conditions of Use

Tools and Visualization

Contributing and Support

7 - About the EM-DAT Project

7.1 - EM-DAT and CRED

The Centre for Research on the Epidemiology of Disasters (CRED)

The Centre for Research on the Epidemiology of Disasters (CRED) is currently a research group that is part of the Institute of Health and Society (IRSS) based at the Université catholique de Louvain (UCLouvain) in Brussels, Belgium. CRED was established in 1973 and has since become one of the world’s leading research centers working on the epidemiology of disasters.

CRED’s main focus is on studying the impact of disasters on populations and on ways to improve the effectiveness of disaster response and risk reduction. CRED’s research includes collecting, analyzing, and disseminating data on disasters caused by natural and technological hazards, within the EM-DAT project, as well as data on conflicts and other humanitarian emergencies.

CRED’s research findings are used to inform policy and practice in disaster management in Belgium and internationally. In addition to its research activities, CRED provides training, consultancy, and technical support to governments, humanitarian organizations, and other stakeholders involved in disaster risk reduction and emergency response.

The Joint History of CRED and the EM-DAT International Disaster Database

In the 1970s, the approach to disaster management was predominantly reactive, emphasizing immediate rescue and relief operations. Recognizing the imperative for a more proactive, informed, and systematic approach to disaster management, Professor Michel Lechat (UCLouvain) established the Centre for Research on the Epidemiology of Disasters (CRED) in 1973. One of the primary challenges faced by the emerging field of disaster epidemiology was the absence of comprehensive global data. A holistic perspective on disasters was essential not only to solidify the foundations of this nascent scientific domain but also to ensure evidence-based disaster management strategies.

In 1984, Debarati Guha-Sapir joined the UCLouvain CRED team and collaborated with Prof. Lechat to initiate the development of the EM-DAT database. This endeavor built upon the foundational efforts of a prior database project spearheaded by the United States Agency for International Development/Office of Foreign Disaster Assistance (USAID/OFDA). CRED then embarked on the meticulous task of data aggregation from diverse sources, including UN agencies, non-governmental organizations, and media outlets. CRED has been managing the EM-DAT database since 1988. After Lechat, Prof. Guha-Sapir assumed the directorship of CRED in 1992, a role she held until 2021 when Prof. Niko Speybroeck took on the role of director.

As of February 2025, the USAID funding for EM-DAT came to a conclusion. We express our sincere gratitude to USAID for more than 20 years of generous, consistent, and meaningful support for the global public good that EM-DAT represents. The continuity of EM-DAT now depends on the engagement of its community of users and partners¹. If you would like to support the project, see Ways to Contribute.

7.2 - EM-DAT Partners

USAID Bureau for Humanitarian Assistance (USAID/BHA)

From 1999 to the beginning of 2025, the Bureau for Humanitarian Assistance (BHA), part of the US Agency for International Development (USAID), supported the EM-DAT project. BHA, known as the Office of US Foreign Disaster Assistance (OFDA) until 2020, coordinated and provided humanitarian assistance to people affected by disasters and crises around the world.

Advisory Bodies

Following the end of the USAID funding, the governance of EM-DAT has evolved since 2026 (see the 2026 Roadmap for EM-DAT).

• Scientific and Technical Advisory Group (STAG). From 1999 to 2019, Technical Advisory Group (TAG) meetings were held every two years on average as part of the evaluation of the USAID-BHA/CRED initiative, producing recommendations for the continuous improvement of EM-DAT. In 2023, the meeting was opened to experts from the scientific community and renamed STAG (Brussels, Belgium, 2023; Potchefstroom, South Africa, 2024). The annual STAG meeting has been paused since 2026. Reports and presentations are available on the STAG page.
• Scientific Committee (SciCom). Established in 2023, the SciCom oversees the quality of EM-DAT data and control processes and meets online twice a year. It is being revamped to strengthen the scientific quality and consistency of EM-DAT data, open science practices, and FAIR data principles (see the SciCom page).
• Advisory Board. Established in 2026 with a broader strategic mandate, the Advisory Board supports the search for sustainable funding, helps define and monitor the social and scientific objectives of the EM-DAT project, and strengthens coordination with global, regional, and thematic partners.

7.3 - Ways to Contribute

EM-DAT was supported by USAID/BHA from 1999 until early 2025. Since the end of this funding, the continuity of EM-DAT depends more than ever on its community of users and partners (see EM-DAT and CRED). You can help in various ways.

Support EM-DAT

• Donate through the UCLouvain giving page. Donations are processed by the UCLouvain Foundation. For larger contributions, or to support our activities and future developments through funding tied to project proposals, please contact us directly.
• Sign the open letter in support of preserving disaster data.
• Cite EM-DAT properly by following the Attribution and Citation Policy, which makes the impact of EM-DAT visible to funders.
• Stay informed: subscribe to the CRED Crunch newsletter and follow us on LinkedIn.