One flood, seven ways of measuring it: Lessons from Nigeria

How many people were exposed to a major flood? Surprisingly, the answer may vary significantly depending on how flooding is measured. Using Nigeria’s devastating 2012 floods as a case study, we compare seven flood measures based on household and community survey reports, media and administrative records, and satellite imagery. While these sources exhibit some overlap, overall they identify very different places and populations as exposed. They also lead to different conclusions about household agricultural losses related to flooding. The analysis illustrates that different ways of observing the same disaster can produce very different empirical pictures, raising important questions about both the consequences of floods and how researchers measure them.

Why measuring a flood is difficult

Floods are among the most damaging natural hazards worldwide, and their consequences are especially severe where households have limited resources to protect themselves or recover. Yet even after a major event, researchers and policymakers may struggle to answer a basic question: where exactly did the flooding occur? This is particularly true in low-income settings where capacity for on-the-ground flood monitoring and reporting is limited.

Flooding can vary sharply over short distances. One neighborhood may be inundated while another nearby remains largely untouched. Even within an affected community, differences in elevation, drainage, housing, and proximity to rivers can produce very different experiences, though exposure to nearby flooding can also affect households and businesses that are not inundated.

There are several ways to observe where a flood has occurred. Official declarations and media reports are based on contact with representatives in flooded areas. Household and community surveys ask people whether they experienced flooding or suffered damage. Satellite products attempt to identify water on the ground from space, and algorithms can determine when the presence of water is likely to represent flooding. Flood models estimate where water is likely to flow based on rainfall, rivers, and terrain.

Each approach has advantages, but none is perfect. Records based on media and administrative reports often provide limited spatial detail on affected areas, and are less likely to include areas where few media or government contacts are present. Surveys provide detailed local information but can be influenced by memory, interpretation, and the specific questions asked. Satellite products provide consistent spatial information, but clouds, tree cover, buildings, the timing of satellite overpasses, and differences between detection algorithms can all affect what is observed.

These sources may also capture different aspects of a flood. A satellite identifies physical inundation. A household reporting that a flood caused a harvest failure is describing an experienced economic consequence. A media or administrative report typically records a flood event with human impacts considered significant enough to be reported, often without precisely identifying which locations were inundated. Alternative ways of measuring a flood are therefore not necessarily interchangeable inputs for analysis of a flood’s effects.

Seven views of Nigeria’s 2012 floods

Nigeria’s 2012 floods were among the most destructive disasters the country has experienced. Flooding affected much of the country, displaced more than one million people, and caused extensive damage to infrastructure, housing, and agriculture.

We use this event as a case study and measure flood exposure in seven different ways. Three measures come from Nigeria’s General Household Survey-Panel, which includes a nationally representative sample of households surveyed in the months following the 2012 floods. We consider households reporting a flood that caused harvest failure, households reporting crops lost because of flooding, and communities where local knowledgeable informants reported a damaging flood. Two spatial measures come from the Dartmouth Flood Observatory (DFO). The DFO archive compiles information on reported flood events from multiple sources and estimates affected areas for each event, and the Global Flood Database refines the affected areas for these events using satellite imagery. Another two spatial measures, a map associated with Nigeria’s Post-Disaster Needs Assessment and the VIIRS satellite flood-detection product, use satellite imagery to more broadly map flooding across Nigeria in 2012. We combine these spatial measures with population data and geolocated survey households to compare estimates of population and household exposure.

The differences are striking. Depending on the spatial flood product and population dataset used, estimates of the share of Nigeria’s population exposed range from less than 0.1 percent to 13 percent. Among households in the national survey, estimates of exposure range from around 2 percent to 38 percent.

The measures also show limited agreement about which households and communities were exposed. Even satellite-derived products based on broadly similar principles identify different flooded areas. Survey measures also disagree with one another because their questions capture different experiences: a community may report a damaging flood even if a particular household did not lose crops, while a household may report flood-related crop damage in an area where a satellite product does not detect inundation.

We next consider whether these differences in flood exposure measures have implications for estimating how these floods affected Nigerian households. We combine the alternative flood measures with household survey data collected before and after the floods to examine agricultural production during the affected growing season. This analysis is deliberately descriptive rather than causal, particularly because some survey measures define flood exposure partly through reported agricultural damage, but the comparison remains interesting. 

Satellite-based measures generally show little evidence of agricultural losses on average for households in communities identified as exposed to floods. In contrast, and as might be expected given how the survey measures are defined, household survey based measures show large declines in agricultural production relative to the season before the floods for households reporting direct flood-related crop or harvest losses.

Combining information from different sources also helps distinguish experiences within broadly affected areas. Households that both report direct flood-related crop losses and live in communities identified as flooded by a satellite-based source experience particularly large agricultural declines, several times larger than those estimated for households reporting direct losses outside such communities. In contrast, households living in areas where flooding is detected but not reporting direct crop losses show no consistent changes in agricultural production.

The broader lesson from the case study is not that one of these seven measures is universally “correct.” Rather, seemingly reasonable ways of observing the same disaster can identify very different populations and can lead an empirical analysis in different directions.

A starting point for a broader research agenda

This project raised several questions that we have pursued in subsequent work. One question concerns the economic consequences of the 2012 floods. Administrative reports identify these floods as one of the worst disasters in Nigeria’s history. Our correlational analysis indicates that estimated effects may depend strongly on the flood measure used. 

The analysis in this paper focuses on short-run agricultural outcomes and is not designed to identify the causal or longer-run effects of flood exposure. In subsequent work (available as a STEG working paper), we take that question further using a longer household panel and an empirical design aimed at identifying causal long-run effects of flood exposure. We study how exposure to the 2012 floods affected household consumption, agricultural production, migration, wage work, and livelihood strategies over the following seven years. We find persistent declines in consumption and farm production, but little evidence that households broadly diversify away from agriculture after the shock. Importantly, conclusions about some livelihood responses depend on whether flood exposure is measured using survey reports or satellite data.

A second question is more fundamental: why do different flood measures disagree, and what should researchers make of those differences? The seven-measure comparison in this paper provides one illustration of that problem. In ongoing work, we are studying flood measurement more systematically across other events and settings, using richer data to understand when and why different approaches identify different places as flooded.

This analysis of Nigeria’s 2012 floods serves as a useful starting point for this broader research program. It shows that flood measurement is not simply a technical detail in the background of an empirical analysis. In this case, considering multiple measures reveals substantial uncertainty about who was exposed and how exposure relates to economic losses. Different views of the same event – whether from the ground or from the sky – can produce different pictures of exposure and motivate very different questions about what happens next. 

Published article reference

Bangalore, M., McDermott, T.K.J. & Biscaye, P. Measuring Exposure and Vulnerability to Nigeria’s 2012 Floods: Views from the Ground and Views from the Sky. Spatial Demography Vol.14, N°16 (2026).