Floods already affect roughly 250 million people each year, but scientists cannot reliably predict how river floods will change as the climate warms. Current flood projections are plagued by a “cascade of uncertainties”—flawed climate models, oversimplified river networks, and patchy data on human interventions like dams and levees. This leaves governments and planners unable to decide how much flood defence, land-use restriction, or infrastructure is actually needed to protect communities. DRIFT-2 aims to close that gap by building the first global “digital twin” of flood hydrology, a consistent computer framework that links streamflow observations with climate, topography, groundwater, satellite data, and human impacts across more than 60,000 river catchments. If successful, the project will produce reliable, long-term projections of flood size, frequency, duration, and extent from 1950 to 2100. That would give engineers, insurers, and local authorities concrete numbers to work with when designing flood defences, zoning floodplains, or setting building codes—decisions that currently rely on guesswork. The work is fundamentally about improving the models that underpin those decisions, not about immediate on-the-ground interventions.
View original technical description
Floods impact approximately 250 million people annually and are feared to keep rising as the climate changes. However, there are critical gaps in our understanding of what drives changes in the size, occurrence, and characteristics of river floods. Further, our projections of future floods remain hindered by a severe ‘cascade of uncertainties’ arising from the biases in climate model outputs and hydrological model structures, uncertain emissions scenarios, oversimplified river networks, and limited global information on human impacts. These gaps affect stakeholders' ability to make planning decisions about appropriate levels of flood risk management, land use, or infrastructure required to protect populations. The Dynamic Drivers of Flood Risk (DRIFT) project was designed to address challenges in understanding past and future changes in flood characteristics to improve decision-making support. DRIFT’s overarching aim is to develop a unified understanding of flood nonstationarity from the past into the future (1950-2100), transitioning seamlessly from short to long timescales. Recognizing the urgency of supporting societies exposed to flooding, DRIFT-2 (2025-2028) builds on DRIFT-1 (2021-2025) in seeking to build the first global models targeted at unravelling flood-generating mechanisms and generating more reliable future projections. Large-sample (multi-catchment) causal and explainable machine learning models will be developed describing the influence of changing climate, engineering structures, and land cover on flood characteristics, such as flood peaks, return periods, probabilities, durations and extent. Past trends in flood characteristics will be seamlessly linked with future climate predictions and projections. These efforts will lean on the Hydro-Climate Extremes Research Group’s recent advances, including the largest existing dataset of hydroclimatic observations in over 60,000 river catchments (OHDB), the first physically-realistic global river network (GRIT), and developments in explainable AI for large-sample flood hydrology. DRIFT-2 will employ GRIT as the backbone of the first ‘Digital Twin’ designed specifically for global flood hydrology, connecting streamflow observations with climate and atmosphere, landscape topography, groundwater, satellite information, and human impacts within a consistent framework, allowing for the exploration of climate impacts using the latest advances in causal and explainable AI (XAI) methods. The long-term aim is to develop robust decision-making support on the evolution of flood characteristics, providing direct benefits for societies globally. Overall, DRIFT-2 seeks to deliver a step change in the use of large-sample machine learning to generate process understanding of flooding, with the aim of supporting better-informed flood risk globally.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know