Spatio-Temporal Disaggregation of CMIP6 Rainfall Datasets and Generation of Intensity Duration Frequency (IDF) Curves for Hyderabad Metropolitan City, India

Cities are not just collections of buildings; they are complex, breathing systems where the weather behaves differently than in the open countryside. When heavy rain falls, the ground in a city cannot absorb it the way a forest floor can. Instead, water rushes over concrete and asphalt, filling drains and streets with terrifying speed. To design the pipes and channels that keep a city safe, engineers rely on a specific kind of map called an intensity-duration-frequency curve. This map does not show geography, but rather the statistical relationship between how hard it rains, how long the storm lasts, and how often such storms occur. It tells planners, for example, how much water a drain must handle if a storm drops a certain amount of rain in one hour versus six hours. However, these maps are becoming outdated. As the global climate warms, the patterns of rain are shifting, becoming more erratic and intense. The old maps, drawn from decades of past weather, no longer reflect the reality of a changing atmosphere.

The challenge for scientists is that the tools used to predict future weather, known as global climate models, are too coarse to see the details of a single city. These models divide the world into large grids, often hundreds of miles across, which is too big to capture the specific, violent bursts of rain that flood urban areas. Furthermore, these models usually predict rainfall in daily chunks, but city drainage systems need to know what happens hour by hour. To bridge this gap, researchers must take the broad, daily predictions from global models and break them down into the fine, hourly details that engineers need. This process involves two steps: first, correcting the models so they match the local weather history, and second, splitting the daily totals into realistic hourly patterns.

A team of researchers led by Naresh Aadhi and colleagues at Osmania University in India tackled this problem for Hyderabad, a massive metropolitan region on the Deccan Plateau. Hyderabad faces a unique climatic mix, sitting at the intersection of tropical wet conditions and semi-arid dryness. The city relies heavily on the southwest monsoon for most of its rain, but recent years have seen a disturbing trend: fewer rainy days overall, but those that do occur are delivering water with dangerous intensity. In 2020, the city recorded over 250 millimeters of rain in a single day, causing severe flooding. To prepare for what comes next, the team needed to understand how these extreme short-duration storms might change by the end of the century. They turned to the latest generation of global climate models, specifically the Coupled Model Intercomparison Project Phase 6, which offers the most advanced simulations of Earth's future climate.

The researchers began by selecting eight different global climate models to see which ones could best replicate the rainfall history of Hyderabad. They used 45 years of actual rainfall records from a single weather station in the city, stretching from 1969 to 2013, as their benchmark. They tested the models against four key characteristics of the rain: the average amount, the variability or how much the rain fluctuates from day to day, the pattern of wet and dry hours, and how the rain on one hour relates to the next. The results were revealing. No single model was perfect at everything. One model, CNRM-ESM2-1, was the best at getting the average rainfall right. Another, BCC-CSM2-MR, was the best at matching the natural ups and downs of the rain. When it came to the timing of wet and dry spells, the models struggled, often predicting too many dry hours. This finding underscored a difficult truth for climate science: there is no single "best" model. The choice of which model to trust depends entirely on which specific aspect of the rain is most important for the task at hand.

Once the models were corrected to match the past, the team had to solve the second part of the puzzle: turning daily rainfall totals into hourly data. They tested three different mathematical approaches to do this. The first, known as the Bartlett-Lewis model, treats rain as a series of pulses that cluster together. The second, the Microcanonical Cascade, breaks the day down like a branching tree, splitting the total rain into smaller and smaller chunks while ensuring the total amount remains the same. The third method, the K-Nearest Neighbor, works by looking at past days with similar weather patterns and borrowing their hourly breakdowns to create new, realistic scenarios. By running these three methods against the corrected global models, the team generated thousands of possible future hourly rainfall sequences.

With these high-resolution sequences in hand, the researchers constructed new intensity-duration-frequency curves for Hyderabad. They looked at two different future worlds: one where greenhouse gas emissions are stabilized at a medium level, and another where they continue to rise at a high rate. They examined two timeframes: the near future, from 2015 to 2056, and the far future, from 2057 to 2100. The results painted a clear and concerning picture for the city's infrastructure. The simulations suggested that the risk of extremely short, intense storms is increasing significantly. Under the medium emissions scenario, the models projected that one-hour storms could reach intensities of approximately 250 millimeters per hour in the near future. This is a staggering amount of water to fall in such a short time. The three-hour storms were also projected to be severe, reaching about 85 millimeters per hour.

Interestingly, the severity of these short bursts did not necessarily increase as the century progressed. The intensity of one-hour and three-hour storms in the far future (2057–2100) remained broadly similar to the near-future projections, suggesting that the city faces a heightened risk of these specific types of extreme events for the rest of the century, regardless of whether emissions are high or medium. However, the story was different for longer storms. The six-hour and twelve-hour rainfall intensities, which are more typical of the monsoon season, stayed within the ranges the city has historically experienced. This indicates that the primary threat to Hyderabad is not a change in the length of the rainy season, but a dramatic increase in the violence of the shortest, most sudden downpours.

The study also highlighted which tools worked best for this specific job. The researchers found that the K-Nearest Neighbor method and the Microcanonical Cascade method were particularly effective at reproducing the patterns of the most intense storms, especially when paired with the MIROC6 global climate model. While other models tended to overestimate the peak intensity of one-hour storms, the MIROC6 model tracked the observed historical trends closely. This gives engineers a specific, reliable combination of tools to use when designing future drainage systems. The findings serve as a stark warning: the old rules for building city infrastructure are no longer sufficient. As the climate shifts, the city of Hyderabad, and indeed many urban centers, must prepare for a future where the most dangerous rain falls in the shortest amount of time, requiring a fundamental rethinking of how water is managed in the concrete jungle.