On February 24, 2022, the Russian invasion of Ukraine began. On the first day of the invasion, battles between Russian and Ukrainian forces occurred in the vicinity of the Chernobyl power plant [1,2]. Following the invasion of Chernobyl on February 25, 2022, the Ukrainian government reported increased levels of radiation in the air [3]. Unverified reports at the time raised concerns of increased radiation levels in the area, potentially due to the disruption of the soil in highly contaminated areas around the power plant due to the fighting and military vehicles moving over the exclusion zone [1,2]. Armed conflict continued throughout March 2022, with intense fighting in Slavutych, a town nearby that houses workers at the power plant. On March 22, 2022, forest fires broke out within the Chernobyl exclusion zone with the potential for generation of contaminated smoke [4]. On March 31, there were reports of confirmed radiation exposure of Russian soldiers, most likely due to soldiers digging trenches in the soil within the Red Forest area to the west of the nuclear power plant [5]. This highlighted the possibility of contamination of previously uncontaminated areas and the potential for subsequent radiological impacts on human and environmental health.

Epidemic open-source intelligence (OSINT) systems provide new approaches to public health surveillance and are increasingly used for epidemic early warning [6]. Early warning OSINT systems can complement and improve the performance of formal surveillance systems by enabling early detection of serious events or fill a gap when routine surveillance systems fail or are absent. Indicator-based surveillance systems largely require clinicians to link cardinal clinical features of specific diseases with key historical, geographic, and social data, thus recognizing the potential occurrence of disease either in an individual or in populations. This process can easily be undermined by a lack of clinical experience, biological variability of presentations in populations, and most importantly, a delay in the recognition of potential disease due to the time it takes for cardinal features to manifest in patients. By contrast, OSINT systems can provide earlier warning through the analysis of large volumes of unstructured digital data and communications. Such data do not rely on clinical experience or acumen, official health system reporting, or the results of laboratory testing. Through the use of specialized processes and algorithms, early warning of potential outbreaks of diseases in populations can be flagged from unstructured sources such as new articles and social media [7]. EPIWATCH and Epitweetr are examples of such systems [6,8]. While OSINT lacks verification, an early warning can be followed by a formal investigation to verify a signal.

Early warning for radiation effects is time critical, as exposures may result in severe outcomes and affect large populations. Following radiation and radioisotope exposure, acute radiation syndrome (ARS) can manifest as early as hours after exposure, and certain therapies require immediate delivery. During the Ukraine conflict, public health surveillance and health protection programs relevant to radiological exposures have been limited or completely ceased. Therefore, in the context of conflict and degraded public health systems, the use of OSINT to rapidly identify locations where a radiological event may have occurred is important and enables the most efficient and timely allocation of limited health resources to limit the spread and impact of contamination. A key distinction in the Ukrainian conflict, as compared broadly to other conflicts, has been the widespread and continued access to high quality open-source data communications, including social media and news sources, across the broad Ukrainian geography and continued penetration of access within the Ukrainian population. Unstructured data from local news reports, social media, and various open-source channels from the Ukrainian population and occupying forces can be used by systems such as EPIWATCH and Epitweetr to detect signals for health-related events of importance.

This study aims to demonstrate the value of OSINT in Ukraine to identify signals of potential radiological events of health significance during the Ukrainian conflict.

We have shown that under conditions of war, when routine reporting and monitoring may be disrupted or absent, open-source intelligence from news reports or social media can be used for early warning of potential radiation events. While signals were detected on February 24 with the beginning of the invasion of the Chernobyl plant, both systems detected further signals in March, which could be linked to the rise in radiobiological events, such as radiation exposure during the armed conflict within the exclusion zone. The Russians occupied the Chernobyl plant from February 24 to March 31, 2022 with acute radiation syndrome reported in Russian soldiers on March 31 and one death reported [5,16]. Potential exposure could have been throughout the occupation of Chernobyl and surrounding areas by the Russian soldiers. Additionally, there was global concern about the disruption of Chernobyl and other nuclear sites during the invasion. Geolocation analysis of radiobiological events for Epitweetr found 2 terms within the Chernobyl region: RAD and Cherenkov. For acute radiation poisoning syndromic analysis, vomiting and headaches were identified within regions surrounding the Chernobyl exclusion zone in the month of March. Clustering of signals in our syndromic analysis for radiation sickness appeared in or around Kyiv. The results from this study show the usefulness of immediate, timely information, particularly in a war zone where access for investigations might be minimal. This information, obtained rapidly, can complement the formal intelligence systems already in place.

OSINT systems have already been used in Ukraine to aid in the detection of potential war crimes and military movements [17]. We showed that OSINT can detect potential radiation events and can be used in real time for early warning. While not a replacement for validated data, such as radiation measurements, open-source data can provide early intelligence when formal reporting is absent and can provide a trigger for an early investigation or emergency response.

A potential limitation to using open-source data is the possibility of manipulation or interference by third parties through the injection of tweets or news sources to boost sentiment. This can be mitigated through multi-source data fusion, triangulation of data, and correlation within and between NLP and machine learning (ML) identified data and other sources of data. In this study, the use of news-based OSINT allowed the validation of Twitter-based OSINT. EPIWATCH, an AI system, applies a model using contemporary NLP and named entity recognition (NER) algorithms in order to detect unusual spikes or signals in particular topics. In addition to system filtration, human moderation is implemented to verify the authenticity of reports. Epitweetr, unlike EPIWATCH, does not individually filter tweets but instead uses a modified EARS, which is a well-established model developed by the US Centers for Disease Control as a baseline for signal detection. OSINT can result in lexical bias that can lead to overreporting of signals and cause issues establishing signal validity. The bias can be allayed by using specific terminology to decrease irrelevant outside noise. This mitigation was confirmed by the detection of distinct spikes in specific terminology not used in regular vernacular.

A further limitation of this study is the dependence on the quality of the data inputs. The data obtained from EPIWATCH could have delayed reporting time or contain biases. There are also language biases, with a predominant amount of news reports being in English for EPIWATCH, despite searching in Ukrainian, which could indicate events outside the scope of the Ukrainian invasion. However, we did perform an analysis on the total reports from EPIWATCH in addition to solely reports in Ukrainian to detect varying signals, if any, from the 2 languages. Additionally, searching in Ukrainian only began in February 2022, whereas searching in Russian was part of EPIWATCH since 2019. For Epitweetr, the tweets are aggregated and rely on built-in signal detection algorithms to distinguish actual events from “white noise.” In addition, the symptoms of radiation poisoning can be indicative of other diseases rather than radiation poisoning. We, however, attempted to mitigate by clustering and geolocating symptoms, in which all symptoms in our syndromic analysis appeared in or around Kyiv. Lastly, the signal detected using open-source syndromic analysis may not reflect radiation exposure and may be a false positive. However, the purpose of OSINT is to monitor the baseline, detect early warning signals above the baseline, and then formally investigate for confirmation.

Using OSINT systems such as EPIWATCH and Epitweetr, signal detection from war zones can be used in the absence of formal detection methods to help rapidly discover and control public health risks. Several studies have identified social media, particularly Twitter, that can be used to identify particular syndromes [18-20]. The value of these open-source data systems, like Epitweetr and EPIWATCH, is the rapid detection of outbreaks and public health events when surveillance systems are not as robust or have been weakened, such as with the invasion of Ukraine [21,22]. An estimated 50% of the stakeholders in epidemic response report lacking access to timely surveillance data, yet 90% do not use available open-source systems, highlighting the potential to improve the use of OSINT [9].

Both systems identified potential radiobiological events throughout Ukraine, particularly on March 4 in Kyiv, Bucha, and 16 reports within Chernobyl. The risk of a nuclear accident will remain a pressing matter as the conflict continues in Ukraine. While Chernobyl has been returned to the Ukrainian government, the Zaporizhzhia plant, where spent fuel assemblies can be damaged, is still under the control of Russia [23,24]. An accident involving spent fuel assemblies could be equivalent in magnitude to the initial Chernobyl event in 1986 and requires the site to undergo constant preventive activities and monitoring. Additionally, normal preventive activities and radiation monitoring sensors have been nonfunctional during parts of the occupation, specifically in Chernobyl, and do not allow for real-time data to be received at this time [25]. OSINT reports can support governmental classified intelligence sources, gather information where formal surveillance might not be as robust or be hindered during the conflict, and provide this information in real time, which can inform timely government responses to the data presented. The significance of OSINT during the invasion, where formal information is scarce, will be to supplement more formal data sources, provide essential early warning of radiobiological events, and ensure timely emergency and public health responses. Both Epitweetr and EPIWATCH can be rapidly adapted to evolving biosecurity or other acute threats. In addition, EPIWATCH continues with the search terminology presented in this study, which routinely monitors potential radiobiological events. These systems can be used as collaborative tools with many stakeholders as a means of surveillance, both in peacetime and in active war zones.