top of page

Today's Extreme Event is Tomorrow's 'Normal': Long-Term Forecasting Under Climate Change

Today's Extreme Event is Tomorrow's Normal

The effects of climate change are not limited to rising global temperatures. Rather, climate change will generally result in extreme weather phenomenon occurring more often. For example, we're already familiar with the rise in heat waves and polar vortices in recent years. Other climatological variables - precipitation, pressure, and wind speed, to name a few - will also see distributional shifts. Put simply: events that seemed like they happened very rarely - say, 1% of the time - will occur more often. Today's extreme event will be tomorrow's normal weather.

We've already released some posts highlighting our work in sub-seasonal forecasting, where we aim to predict climatological variables 2 weeks to 2 months in advance. However, our forecasting models are not limited to this time frame. The question of how climate change impacts the occurrence of extreme events, for example, is a long-term forecasting problem, in which we have to predict patterns on the timescale of 10 or more years. At Vā, we're using physics and AI to tackle long-term forecasting in the presence of climate change.

What's Happening with Wind Speed?

Wind speed is an often overlooked climatological variable. How often do we check the day's weather for wind speed? Given that wind power is an increasingly sizable renewable energy source, it's prudent to develop good forecasting models for wind speed in both sub-seasonal and long-term windows.

Globally, wind speeds have started to increase after a several decade-long period of slowing down. As we feel the effects of climate change more strongly, wind speeds may exhibit increasingly complicated patterns. How can we go about forecasting wind speed on a long-term scale?

One approach to long-term forecasting is a General Circulation Model (GCM). GCMs capture the general circulation of the atmosphere with physics-based models in order to simultaneously forecast multiple climatological variables. There are a variety of GCMs available, at different resolutions, incorporating different physical processes, and accounting for climate change in various ways. Each has its strengths and weaknesses.

Improving General Circulation Models with AI

While GCMs are very powerful, they can be prone to errors. They are highly dependent on their initial conditions and are prone to biases, especially on longer timescales. However, we can use machine learning methods to both 1) correct for these biases and make their predictions more accurate and 2) leverage the strengths of multiple GCMs in an ensembling procedure to supercharge prediction. This approach is in line with our philosophy at Vā taking the best of both physics-based and AI approaches to improve forecasting.

Let's examine wind speed predictions as an example. We took a look at how extreme wind speeds have shaped up in recent years relative to the prior few decades. Specifically, we examined the wind speed distribution from 1980-2010 and calculated the 99th percentile or the wind speed you'd need to be in the top 1% of extreme events during that time period. If wind speed doesn't change over time, that wind speed would occur 1% of the time in the ensuing years, from 2016 to 2020. In reality, we know this is not true, because global wind speed is changing. There are many regions in the continental United States that experience wind speeds of more than 1% - often up to 4% of the year (Figure 1, left: "ERA5" is the observed truth). That's the extreme wind speed becoming tomorrow's normal!

Wind Speed ChangesExtremes Skill Analysis

Figure 1: How is wind speed changing over the continental United States? Left: Observed number of 'extreme wind speed' days, defined as days reaching the 99th percentile from 1980-2010. Right: Predicted number of 'extreme wind speed' days, forecasted using Vā's models.

Could we have accurately forecasted this using GCMs? We took some GCMs, applied our AI methods to their outputs in order to improve their predictive performance, and looked at how often the forecast achieved extreme wind speeds. The distribution, shown in Figure 1, right, looks very similar to the ground truth. Vā's model is able to accurately forecast the long-term changes in extreme wind speed!

Since GCMs are global models, we can perform this same experiment in a different region, such as Europe (Figure 2). We find that our models are, once again, able to capture the increasingly extreme wind speeds. Today's extreme weather might be tomorrow's 'normal' - but that doesn't mean we can't accurately predict and prepare for the new normal.

Wind Speed Changes

Figure 1: How is wind speed changing over the continental United States? Left: Observed number of 'extreme wind speed' days, defined as days reaching the 99th percentile from 1980-2010. Right: Predicted number of 'extreme wind speed' days, forecasted using Vā's models.

How Can Our Long-Term Forecasts Support Your Business?

Your business will undoubtedly feel the effects of climate change. Our long-term forecasts can help you better manage your risk in an increasingly uncertain future.

Reach out to us at if you'd like to hear more about our data products!

bottom of page