Human losses due to climate-related disasters: an urgent call for quality control


 Missing data in climate impact databases are a source of major bias in disaster impact analyses and can mislead policy making. EMDAT, an international reference data source on disaster impacts, is often used without much quality control or consideration of data gaps. Ignoring these gaps undermines both the evaluation of disaster risk management activities and the evidence base for global policies for early warnings and disaster risk reduction. EMDAT, considered a key data source on climate extremes and their human impacts, records casualty information for only one in three climatological disasters. In this study, we provide insights into the missing data, highlighting the importance of data quality control to account for bias. Understanding where and why data is missing provides insights into whether trends can be attributed to true progress in disaster risk reduction or are merely statistical artifacts. We also underline the pressing need for harmonization across other databases that are relevant for measuring the human impact of climate disasters and suggest new technologies to fill gaps. As major resource investments were requested during COP 28 (2023) to attenuate the impact of future disasters, more complete and convincing data on climate disasters and human impacts will be needed to better identify the efficiency of countermeasures.


Introduction
The UN Conference of the Parties (COP) 28 ended on 13 December 2023.It concluded with new targets for the Global Goal on Adaptation (GGA) and its framework, as well as high-income countries pledging a combined total of around US$ 700 million to a loss and damage fund.The UN Office for Disaster Risk Reduction and the UN Office for Project Services will jointly host the Santiago Network secretariat to 'avert, minimize and address loss and damage' 4 in low-income countries that are particularly vulnerable to the adverse effects of climate change.Credible evidence on direct human and economic impacts will drive resource priority-setting and allocation.
Data on climate impacts have increasingly become central to global and national policy.The international disasters database EMDAT 5 , which is publicly accessible, is widely used to guide key national and international policy and scientific analyses.Reports from influential institutions such as the Intergovernmental Panel on Climate Change's Managing the Risks of Extreme Events and Disasters to Advance Climate Change Adaptation (SREX) (Intergovernmental Panel on Climate Change 2012) report or the World Disasters Report (International Federation of Red Cross and Red Crescent Societies 2023) have used data from EMDAT to underpin their conclusions.Many others rely on EMDAT to calculate 4 www.lossanddamagecollaboration.org/pages/did-cop-28-getus-closer-to-the-world-we-want-assessing-the-outcome-on-lossand-damage. 5 EMDAT is maintained by the Université Catholique de Louvain, receives long term funding by the United States Agency for International Development (US AID), offers freely accessible data that can be filtered by time, country, and disaster type.
indicators reflecting risk or resilience to climaterelated disasters.The establishment of the loss and damage fund and the emphasis on measuring impacts at COP28 highlight the critical need for accurate, unbiased impact data.However, most public reports discussing trends in disaster-related damages fail to mention critical limitations of EMDAT, such as data gaps and a temporal bias caused by improved reporting mechanisms.
For example, reports related to the Early Warnings for All Initiative6 of the UN World Meteorological Organization credited the success of early warning systems (EWS) and disaster risk reduction (DRR) with 'slashing the human casualty toll over the past half of a century'7 based on EMDAT death reports.Indeed, the number of disaster-related casualties for various climate-related emergencies dropped from 666.877 in the 1980s to 328.672 in the 1990s.We argue, however, that a more nuanced perspective is needed to avoid the communication of potentially misleading messages.One example is that approximately 550.000 casualties are related to three drought events in Ethiopia, Sudan, and Mozambique in the 1980s.Simultaneously, the >600.000casualties related to famine in North Korea in the 1990s (that had been exacerbated by droughts and floods) are exclusively associated with a non-climate-related category ('complex disasters') in EMDAT.If just this one event had been categorized in EMDAT as either drought or flood-associated, an increase in total deaths of 40% from the 1980s to the 1990s would have been recorded.
Globally, the annual average of climate-related disasters documented by EMDAT has increased by a factor of 7.5 from 67 events per year between 1970-1979-572 during the latest complete decade (2010-2019).Deaths per event due to climaterelated disasters in EMDAT have globally been on a steep decline since the 1970s, until the 1990s, after which the rate of decrease is slower.This trend, however, is not uniform throughout all disaster categories.In the case of climatological events (primarily droughts and wildfires), for instance, there has been an increase in average deaths per event from 17 in the 1990s to 70 in the 2010s (figure 1).We highlight two factors-one methodological and one data-related-that can introduce substantial distortion for setting priorities for resource allocation.

Declining deaths from climate-related disasters: causality vs. association
It is important to avoid confusing a chance association with a causal relationship between EWS and DRR interventions and their impacts on the decline of death tolls.DRR policies must be built on sound statistical evidence that convincingly links successful DRR activities to declines in disaster-related deaths.We must be able to show that a specific intervention has had the desired effect on mortality in the affected community against a counterfactual or through a valid comparative design.Persuading national governments or international development partners to invest in and scale up DRR or locally-led humanitarian instruments like anticipatory action will be much easier if there is robust evidence showing that human and/or economic losses could be reduced.

Missing data on climate-related deaths
Data quality considerations, especially accounting for data gaps, is a basic requirement for quality statistical analyses.Admittedly, few global data sets are as complete and accurate as we would like.Most scientifically sound analyses of global data that attribute causality-e.g.linking the incidence of a specific type of cancer to food consumption patterns, and in this case, between EWS or DRR and decline in deathswould need to account for data gaps causing biases in the data.
Two rare studies (Jones et al 2022, 2023) investigated how well the literature accounted for missing data on economic losses in the EMDAT database.They found that missing data are rarely acknowledged or, if they are, are constrained to a brief mention.Few incorporate any sensitivity or bias correction to reduce the influence of missing data on the conclusions.Unfortunately, there has been little progress in reducing the share of missing data on deaths from climate disasters (figure 2).
Over 12 000 climate-related disasters were recorded between January 1970 and December 2022, and about a third of those were missing data on deaths.This proportion showed little improvement over the last four decades (figure 2).Most records of missing deaths occur in the category of 'climatological' disasters.During the 1970s, no deaths were registered in 91% of all climatological events despite an increase from 68 annual average events in the 1970s to 266 between 2010-2019.Droughts are admittedly a data challenge regarding deaths as no appropriate methodology to attribute deaths to drought exists and reported drought deaths are pitifully small compared to the real tolls (Enenkel et al 2020).Excess mortality methods could be a better method to reflect reality.Today, only one in three climatological disasters registers any information on casualties in EMDAT.
Finally, as EMDAT data on deaths are so widely used to establish trends and patterns, we also need to address the implications of blank entries for deaths in a record.Currently, these cannot be distinguished between 'data not available' or 'no deaths' .Understanding the exact nature of these data gaps could have made previous research, such as the studies of Formetta and Feyen (2019), Tin et al (2024), even more impactful.Articles published by influential think tanks, such as the World Economic Forum8 , suffer from the same limitation.

Improving human impact data: building robust foundations for policy-making
Data on human impacts of climate disasters are key to protecting communities on the front lines of climate-related emergencies.Accurate and reliable data are also the bedrock for effective and targeted adaption choices, especially as various types of financial resources run thin9 .There are many options to improve the quality and accuracy of climate disaster impact data.We offer two pragmatic paths that are realistic and affordable.
First, resources for DRR and recovery are often distributed based on reported death tolls (or number of people affected) and displacements-the most influential indicators of severity.However, data collected manually at subnational levels can be contradictory and inconsistent, especially if they pertain to losses from areas that are physically inaccessible.We can close some of these data gaps by harnessing state-of-the-art technology.Novel data-driven insights, such as those gathered via satellite earth observation already enable more rigorous empirical analyses of disaster determinants and impacts.Rosvold and Buhaug (2021) highlighted the relevance of geospatial information to connect EMDAT to other subnational data sources by developing a Geocoded Disasters dataset.In addition, satellitederived information can already be used to estimate the number of people affected by different natural hazards, such as floods, based on building footprints (Portalés-Julià et al 2023).However, complementing or validating statistics on disaster-related casualties will likely remain challenging.
The targeting of humanitarian aid can be improved by applying machine learning models to data collected via mobile phones (Aiken et al 2022).
Similarly, the rapid acquisition of spatial data via mobile technology, such as interactive voice response systems or platforms like Kobo Toolbox (Klabbers et al 2023), ideally combined with automated data mining, can increase the quality of data-driven insights.This enhancement in data acquisition and analysis can also streamline or refine the decisionmaking process and support closing gaps regarding 'small events' , characterized by fewer than 30 deaths or fewer than 5000 houses destroyed.Between 1990 and 2013, 99.7% of all disasters fell under this category and thousands of small events are unreported every year (UN Office for Disaster Risk Reduction 2022).However, as highlighted by UN Office for Disaster Risk Reduction (2023), only front-line responders can gather individual-related data, such as demographics, injuries, casualties, or disabilities.Combining these data with satellite-or drone-based observations, quality control, and advanced analytics provides an ideal basis for the creation of situation reports, trend analyses, or efforts to fill statistical gaps.
Second, large differences between the damage estimates reported in EMDAT and other databases, such as DESINVENTAR, have been documented (Panwar and Sen 2019), not least because there is no common methodology to capture the human impacts of climate disasters especially droughts, heatwaves, and wildfires.
Disaster databases, such as EMDAT, are not unimpeachable.EMDAT quantified the human impacts of disasters in the 1980s when they were considered unfortunate and uncontrollable events and no global data existed.It was done by manually checking selective websites and information received from sources such as the Red Cross and Red Crescent Movement.Today, there are multiple excellent data sources, such as databases maintained by governments or insurance companies.However, many of these data sources, such as SwissRe's CatNet database, are not publicly accessible and/or proprietary.They range from ecosystem-mediated disease impacts, deaths, and displacement to socio-economic characteristics of the affected communities.A step forward would be to build a climate impact data cooperative on a common platform.This could harmonize high-resolution information from different disciplines currently held by established global sources and would offer a richer landscape for data scientists and policymakers.

Discussion
Historical and projected impacts of climate change are the basis on which the allocation of resources is justified.Investments made today to build resilience and adapt to imminent disasters will substantially impact the lives and livelihoods of current and future generations.This is especially crucial as the impacts of extreme events could jeopardize hard-earned development gains.Immediate and strategic actions are essential to protect and enhance these achievements.However, the financial resources needed to protect the lives of poorer communities are nowhere close to what is needed, as was made clear in COP 28.According to the Loss and Damage Coalition (2023), low-income countries will need at least US$ 400 billion per year (excluding non-economic impacts) to address loss and damage.The US$ 700 million allocated to the loss and damage fund at COP28 amounts to less than 0.2% of this amount.
Most low-income countries do not have the resources to wait for decades until today's investments in climate change adaptation might materialize.Pragmatic, evidence-based policies are essential to facilitate the collaborative development of concrete coping mechanisms and national structures that enhance resilience.One cornerstone for successful interventions is cooperatively recorded impact data that allow us to analyze whether, how well, and under what conditions our actions have succeeded or failed.Statistics on disaster-related deaths are particularly sensitive, as the recorded number of victims can sway the allocation of aid or absolve authorities from blame, depending on whether these numbers are inflated or minimized.This scrutiny will challenge existing norms and structures, but it is better to face this inconvenience now than to confront it after spending limited funding with little impact on the world's most vulnerable and disaster-prone countries.

Figure 2 .
Figure 2. Missing data (%) for deaths due to climate-related disasters by period and broad hazard category in EMDAT, 1970-2019 (analysis based on data downloaded on 5 November 2023).