What Floods Leave Behind: Dev Patel on Disaster, Adaptation, and the Economics of Resilience

My own research asks what happens to children's education when disaster strikes. So when Dev Patel walked into the seminar room at Tufts in October 2025 with satellite imagery, twenty years of flood data, and a question I had been circling in my own research, I was paying close attention. What he presented was one of those rare papers that solves two problems at once: a measurement problem that has quietly undermined an entire literature, and a substantive question that matters enormously for how we think about human capital in vulnerable places.

The question is deceptively simple. How do floods affect economic development, and how do people adapt to recurring exposure? The answer turns out to require building, almost from scratch, the data infrastructure to answer it properly.

Patel develops a globally applicable flood detection system that combines geophysics and machine learning to analyze satellite radar and optical data, enabling the detection of inundation anywhere on Earth, daily, for the past two decades. This corrects longstanding weaknesses in flood databases: inconsistent definitions, sparse sensors, and chronic underreporting in poorer regions. The result is a standardized, unbiased measure of flooding across space and time that researchers can actually trust.

He applies this framework to Bangladesh, one of the world's most flood-prone countries, constructing a daily inundation dataset for all of Bangladesh's unions, small geographic administrative units, spanning twenty years. Validation against government hydrological records, news archives, and in situ data (measurements taken directly in the field, rather than remotely) shows that his algorithm explains about 90 percent of the variation in surface water observed from gold-standard satellites.

Using this newly built dataset, Patel estimates the causal effects of floods on economic activity through a stacked difference-in-differences event-study design. The identification strategy compares areas with similar flood risk but differing exposure in a given year, using strata constructed with machine learning to ensure comparable pre-trends. The findings are striking. Floods cause sharp and persistent declines in economic activity, visible even from space: nighttime luminosity, a proxy for output and electricity use, drops significantly and remains depressed for at least seven years. Complementary satellite data confirm a reduction in building footprints, indicating destruction of physical capital. Administrative microdata covering nearly 19 million people reveals a pattern of structural change following floods: employment shifts from agriculture toward non-farm work, migration rises, and parents invest more in their children's schooling.

In the immediate aftermath of a flood, when income has collapsed and physical capital has been destroyed, parents invest more in education. That is not an obvious response to catastrophe. It suggests something important about how households update their beliefs about the future and where they perceive returns to be, a point Patel formalizes in his model.

He further shows that prior flood exposure mitigates future damage. Areas that have experienced floods before suffer smaller subsequent losses, consistent with experience-driven adaptation. Repeated exposure raises households' perceived returns to adaptation investments while reducing the fixed costs of coping behaviors like temporary migration. Empirical evidence from farmer surveys supports both mechanisms: households with past flood experience expect greater future risk and invest more in mitigation, while those with migration experience face lower costs of relocating.

Perhaps the most elegant piece of the paper involves colonial transportation networks. Using routes digitized from an 1877 publication by the British Military, Patel shows that better historical market access and mobility reduce contemporary flood vulnerability. Infrastructure built for extraction over a century ago still shapes who can adapt and who cannot. That is a human capital and institutional persistence argument dressed in climate economics clothing, and it is exactly the kind of finding that makes this paper worth reading beyond its primary contribution.

The strengths are considerable: methodological originality, extensive validation, and rigorous causal inference using rich, high-frequency data. But there are questions worth pressing.

The Bangladesh focus is a genuine external validity concern. Does the adaptation story travel? Bangladesh has a specific geography, a dense population with long experience of flood management, and particular historical and institutional characteristics. The mechanisms Patel identifies may not operate the same way in, say, sub-Saharan Africa, where flood exposure is growing but institutional capacity looks very different.

The schooling result also deserves deeper scrutiny. Is the increase in educational investment a resources story, a beliefs story, or a relative returns story? When agricultural income collapses and non-farm work expands, the returns to education relative to unskilled labor may simply rise mechanically. That would be an important finding, but a different one from households genuinely updating their beliefs about the future. The paper gestures at mechanisms without fully resolving which one is doing the work.

Finally, the colonial transportation instrument is elegant but not without its own complications. Colonial infrastructure was not placed randomly. It followed existing population centers, resource extraction routes, and political calculations. The places the British connected were already different in ways that may independently predict resilience. Patel is aware of this, but it is the kind of identification concern that deserves more than reassurance.

None of this diminishes what the paper achieves. Patel has built something genuinely useful: a measurement framework that can be applied globally, a causal identification strategy that holds up, and a set of findings that complicate easy narratives about disaster and recovery. Floods do not simply destroy. They also, in their aftermath, reshape how households invest in themselves and their children. That is a human capital story, and it is not the one most people tell about natural disasters.

Dev Patel presented this work at Tufts University on October 21, 2025. The paper is available at dev-patel.com.