Quasi-global, land-only, high-resolution and spatially averaged climate variables from downscaled CMIP6 models for climate impact research

To understand how the climate will change in the coming decades, scientists rely on massive computer simulations called global climate models. These programs act as digital twins of the Earth, simulating the complex dance of the atmosphere, oceans, and land to predict future weather patterns. However, these models have a significant blind spot: they are built to see the big picture, not the details. Because the Earth is so vast, these models divide the planet into large, coarse grid squares, often spanning hundreds of miles. Within each square, the model calculates an average temperature or rainfall, smoothing over the hills, valleys, and coastlines that make real-world weather so specific. For a farmer in a specific valley or a public health official planning for a disease outbreak in a particular city, this broad average is not enough. They need to know what the weather will look like on their own doorstep, not just for a continent-sized region.

This gap between global predictions and local reality is the central challenge addressed by a new dataset created by researchers at Imperial College London and the University of Leeds. The team has developed a method to take these coarse, global climate models and sharpen them into high-resolution maps that show daily weather changes at a scale of just a few miles. Their work focuses on the tropical regions of the world, where climate variability is high and where many infectious diseases thrive. By refining the data to a level of detail previously unavailable for such a wide area, they have provided a tool that allows researchers to move from general warnings about a warming world to precise forecasts for specific communities.

The researchers started with six different global climate models, which are the leading tools used by scientists to project future climate conditions under different scenarios of human activity. These models run on two distinct paths: one representing a moderate future where emissions are curbed, and another representing a high-emissions future where fossil fuel use continues to rise. The raw output from these models covers the period from 1985 to 2100, but the data is too coarse and contains systematic errors. For instance, a model might consistently predict it is slightly too hot or too wet in a specific region compared to what actually happens. To fix this, the team applied a sophisticated statistical technique called Double Bias-Corrected Constructed Analogues. This method works by comparing the model's predictions against real-world observations of weather that have already happened. It essentially asks the model to look at past days that resembled the current conditions and use those real historical patterns to correct the model's mistakes. This process not only fixes the errors but also fills in the missing details, creating a high-resolution picture of daily temperature, humidity, and rainfall.

The result is a massive collection of data covering twelve land-based regions between 60 degrees north and 60 degrees south of the equator, a zone that encompasses most of the world's tropical and subtropical areas. The data is provided at a resolution of 0.1 degrees, which translates to a grid square roughly six miles wide. This is a significant improvement over previous datasets, which often used squares of 15 AU or more. The team generated daily weather records for six key variables, including the daily maximum and minimum temperatures, humidity levels, and total rainfall. They covered both the historical period from 1985 to 2014 to ensure the method works, and the future period from 2015 to 2100 to provide projections. Crucially, they did not stop at creating these detailed maps. Recognizing that many decision-makers need data that aligns with political boundaries rather than grid squares, they also aggregated the information. They calculated average weather conditions for 110 countries and their sub-regions, weighting these averages by where people actually live. This means the data reflects the climate experienced by the population, not just the empty land, making it directly useful for planning health interventions and resource management.

To ensure the quality of their work, the researchers rigorously tested the new data against real-world observations. They compared the downscaled historical data from 1985 to 2014 with actual weather records from satellites and ground stations. The results showed a dramatic improvement over the raw models. The original models often had large errors, sometimes predicting temperatures that were off by several degrees or rainfall amounts that were wildly inaccurate. After the correction process, these errors shrank significantly. The new data matched the real-world seasonal cycles much more closely, capturing the timing of rainy seasons and the intensity of heat waves with far greater accuracy. The team also checked how well the data represented extreme events, such as days with heavy rain or dangerously high temperatures. They found that the corrected data successfully reproduced the frequency of these extreme days, a critical factor for assessing risks like flooding or heat stress.

When looking at the future projections, the researchers found that the general direction of climate change remained consistent with the raw models, but the details were much clearer. In the future scenarios, the data showed a warming trend across the region, but the new high-resolution maps revealed how this warming varies across different landscapes. For example, in mountainous areas like the Andes, the raw models struggled to represent the complex terrain, often smoothing out the unique climate patterns. The new data resolved these areas clearly, showing that the warming signal in these high-altitude regions is actually less intense than the surrounding lowlands. Similarly, for rainfall, the new data showed that while the broad trends of wetter or drier conditions were preserved, the local patterns were far more complex than the coarse models suggested. In some coastal and mountainous areas, the new data even showed a different direction of change compared to the raw models, highlighting how local geography can alter the impact of global climate shifts.

The researchers are careful to note that while this dataset is a major step forward, it is not a perfect crystal ball. The method relies on the assumption that the relationship between large-scale weather patterns and local conditions will remain similar in the future, an assumption that may not hold if the climate changes drastically. Furthermore, the accuracy of the new data depends heavily on the quality of the real-world observations used to correct the models. In some remote areas where weather stations are sparse, there is still uncertainty. However, the team has demonstrated that their approach successfully removes the systematic errors found in the original models and provides a much more realistic view of daily weather patterns. By making this data freely available in formats that are easy to use, they have removed a major barrier for scientists and policymakers. This resource allows researchers to build more accurate models of how diseases spread, how crops will fare, and how water resources will change, turning broad climate warnings into actionable, local knowledge.