A stochastic weather generator for multisite, multivariate climate scenarios with transient climate change
This paper presents a scalable stochastic weather generator that combines k-nearest-neighbor resampling with transient climate perturbations from CMIP6 models to produce spatially coherent, multivariate climate scenarios for hydrological impact assessment under non-stationary climate change conditions.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Planning for water in a changing climate is like trying to navigate a ship through fog while the map itself is slowly shifting. Water managers need to know not just how much rain might fall in the future, but how that rain will arrive over decades, how it will interact with rising temperatures, and how those conditions will play out across different parts of a river basin simultaneously. The challenge is that the future is not a single, fixed destination; it is a vast range of possibilities. To prepare, scientists must generate thousands of plausible weather scenarios to test if dams, reservoirs, and supply networks can survive the most difficult conditions. However, creating these scenarios is difficult because weather is complex. Rain and temperature are linked, and conditions in one valley often depend on what is happening in the next. Furthermore, the climate is not staying still; it is moving, warming, and drying in ways that change over time. Traditional tools often assume the climate stays the same or apply simple, static adjustments that fail to capture these evolving, interconnected risks.
To solve this, a team of researchers from Stanford University and the Pontificia Universidad Católica de Chile has developed a new method for creating realistic, long-term weather forecasts for water planning. They focused on the Maipo River Basin in central Chile, a critical water source for the city of Santiago that has already been suffering through a severe, multi-decade drought. The researchers built a system that combines two powerful ideas. First, they used a technique that looks at decades of historical weather records to find patterns. Instead of inventing new weather from scratch, the system picks and stitches together real weeks of rain and temperature from the past that look statistically similar to the current situation. This ensures that the generated weather keeps the natural relationships between different locations and variables, such as the fact that it is often colder and wetter at higher elevations. Second, they added a module that slowly shifts these historical patterns to match the warming and drying trends predicted by global climate models. This allows the system to create weather sequences that feel real and connected, but which gradually drift toward the hotter, drier future that scientists expect.
The researchers tested their approach by generating massive sets of weekly weather data for 257 different locations across the Maipo Basin. They created 108 distinct future scenarios, each representing a different combination of climate change pathways and natural weather variations. When they fed these scenarios into a detailed computer model of the river system, the results showed that their method successfully preserved the complex web of connections between different parts of the basin. The generated weather kept the correct seasonal rhythms and the right relationships between temperature and rain, even as the overall climate warmed. Crucially, the method captured the full range of uncertainty, producing a wider spread of possible outcomes than the raw climate models alone, which helps planners see the full scope of potential risks.
The study revealed a stark difference between looking at the future as a static snapshot versus watching it unfold as a moving story. When the researchers compared their new, evolving scenarios against older methods that assumed a fixed future climate, the difference in drought risk was dramatic. Under the new, transient scenarios, the river's flow did not just drop; it declined progressively over the century. By the end of the 2100s, the simulated droughts were far more severe than what the static models predicted. This happened because the rising temperatures and falling rain compounded each other over time, drying out the soil and reducing the snowpack that feeds the river in the spring. The study suggests that relying on static assumptions could lead water managers to underestimate the severity of future droughts and the timing of when they will become critical. By providing a tool that generates large, realistic, and evolving weather ensembles, this research offers a clearer, more urgent picture of the challenges ahead, helping communities prepare for a future where the climate is not just different, but constantly changing.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.