Flooding is a major hazard that is responsible for substantial damage and risks to human health worldwide. The 2018 flood event in Kerala, India, killed 433 people and displaced more than 1 million people from their homes. Accurate and timely information can help mitigate the impacts of flooding through better preparedness (e.g. forecasting of flood impacts) and situational awareness (e.g. more effective civil response and relief). However, good information on flood impacts is difficult to source; governmental records are often slow and costly to produce, while insurance claim data is commercially sensitive and does not exist for many vulnerable populations. Here we explore “social sensing” – the systematic collection and analysis of social media data to observe real-world events – as a method to locate and characterise the impacts (social, economic and other) of the 2018 Kerala Floods. Data is collected from two social media platforms, Telegram and Twitter, as well as a citizen-produced relief coordination web application, Kerala Rescue, and a government flood damage database, Rebuild Kerala. After careful filtering to retain only flood-related social media posts, content is analysed to map the extent of flood impacts and to identify different kinds of impact (e.g. requests for help, reports of medical or other issues). Maps of flood impacts derived from Telegram and Twitter both show substantial agreement with Kerala Rescue and the damage reports from Rebuild Kerala. Social media content also detects similar kinds of impact to those reported through the more structured Kerala Rescue application. Overall, the results suggest that social sensing can be an effective source of flood impact information that produces outputs in broad agreement with government sources. Furthermore, social sensing information can be produced in near real-time, whereas government records take several months to produce. This suggests that social sensing may be a useful data source to guide decisions around flood relief and emergency response.