How digital media can fill global disaster information gaps

One of the study’s most significant implications is that online media can function as a real-time sensor network, particularly in contexts where national datasets are slow, incomplete, or aggregated at a high level. The researchers suggest that disaster impact data from local digital news could be mined using NLP to detect named entities, track geospatial mentions, and analyze sentiment—all of which could feed into databases like EM-DAT and DesInventar.

In the face of worsening climate-related disasters, new research underscores how local digital journalism in data-poor regions could play a critical role in disaster risk reduction. The study, titled “Framing of Disaster Impact in Online News Media: A Case Study from Malawi on Flood Risk Management”, was published in Frontiers in Communication. It finds that online news coverage, especially from newspapers with social media presence, can offer granular, human-centered data to complement traditional impact databases that often overlook localized damage in countries like Malawi.

The researchers applied natural language processing (NLP) to 132 online news articles from leading Malawian newspapers, analyzing how flood events in 2019 were framed. Their findings revealed that media coverage in Malawi disproportionately emphasized political actors, relied on a neutral tone, and provided detailed geographic references. This hybrid of narrative and precision suggests that online news could serve as a valuable input to global emergency data systems, particularly where official data is limited, delayed, or insufficient.

How do local news outlets frame flood disasters in the absence of complete data?

The study focuses on Malawi, a country highly vulnerable to floods yet characterized by significant gaps in official disaster reporting. Government datasets such as EM-DAT and DoDMA provide useful national-level summaries but often lack localized indicators, especially around economic, agricultural, and human impact. To address this, the researchers examined the framing and sentiment of flood coverage across four Malawian digital newspapers: Nyasa Times, Malawi 24, MW Times, and MW Nation.

Instead of purely technical reporting, the news articles tended to prioritize human-focused narratives, particularly stories involving political leaders and affected communities. Named entity recognition showed that location names, especially flood-stricken southern districts like Chikwawa and Nsanje, were the most frequently mentioned entities, followed closely by politicians and disaster relief organizations. However, cities like Lilongwe were overrepresented despite not being the most affected, revealing a potential bias toward politically or economically significant locations.

This finding challenges the notion that local media simply repackage state data. Instead, journalists construct their own frames by choosing which locations and actors to highlight, and how to portray their roles. These decisions carry weight in shaping public understanding, influencing both citizen behavior and institutional responses. For example, the frequent mention of DoDMA as a key information source reflects how state agencies dominate the narrative, potentially crowding out community voices or alternative expertise.

What does the tone of reporting reveal about the credibility and function of digital media in disaster scenarios?

Contrary to assumptions that disaster news is dominated by sensationalism, the majority of analyzed articles conveyed a neutral tone, focusing on factual summaries rather than emotional appeals. The average polarity score hovered near zero, while subjectivity remained low. This pattern suggests that Malawian newspapers, despite resource constraints, often strive for objectivity in their disaster coverage. However, tone analysis also revealed subtle variation by flood type. Flash floods were reported with a slightly more negative tone than river floods, possibly due to their abrupt onset and dramatic impacts.

Importantly, tone neutrality does not imply lack of influence. By maintaining an even tone, newspapers position themselves as reliable sources of official information. Yet, this same neutrality may result in missed opportunities to spotlight underreported suffering or expose systemic gaps in flood preparedness. Moreover, the framing of politicians as central actors—often depicted as donors or responders—introduces a risk of politicizing disaster narratives at the expense of long-term structural analysis.

The use of emotionally neutral language may be partially attributable to media norms and technical reporting constraints. Malawi’s press faces challenges in accessing remote regions, verifying field-level impacts, and translating citizen experiences into quantifiable metrics. However, sentiment analysis also revealed that even when disaster victims were mentioned, their stories were often framed passively—as victims rather than agents. This underrepresentation limits the ability of media to advocate for community-led resilience or to provide early signals of systemic vulnerability.

How can digital media reporting enrich institutional disaster databases and inform future risk management?

One of the study’s most significant implications is that online media can function as a real-time sensor network, particularly in contexts where national datasets are slow, incomplete, or aggregated at a high level. The researchers suggest that disaster impact data from local digital news could be mined using NLP to detect named entities, track geospatial mentions, and analyze sentiment—all of which could feed into databases like EM-DAT and DesInventar.

The granular nature of local reporting allows for impact measurement at the district or even village level—an advantage over top-down systems that prioritize only large-scale events. For instance, the study found multiple references to specific damages to homes, crops, and livestock, which are often absent from official datasets. These insights are especially valuable in agricultural economies like Malawi’s, where flood-related losses affect food security and inflation in ways that are not always visible in aggregated statistics.

However, the study also acknowledges a few limitations. The NLP models used were not fine-tuned for Malawian dialects, and the dataset excluded articles in Chichewa, potentially omitting hyper-local insights. Additionally, reliance on Twitter as a primary gateway to articles may skew the dataset toward more digitally active media outlets, leaving out print or community-level publications that may use different frames or tones.