The Prediction Desert

Why forecasting >2 weeks ahead is difficult

At Beyond Weather , we forecast the weather and its impact from two weeks up to months into the future. Predicting weather beyond two weeks is crucial for planning infrastructure, energy security, agricultural activities, and disaster preparedness in an increasingly volatile climate.

People often assume that these types of forecasts are similar to the daily updates they receive on their phones. However, forecasting beyond two weeks is a completely different and challenging task. Climate scientists have even dubbed this forecasting challenge 'The Prediction Desert.' Here's why.

Sources of predictability

To understand the prediction desert, it's important to grasp the difference between short-range and long-range weather forecasting. First of all, we tap into different sources of predictability depending on the lead time of the forecast.

In the short range, most of the prediction power comes from the atmosphere, which is a dynamic and rapidly changing system. However, its predictive power diminishes quickly when forecasting more than two weeks ahead.

Climate variables on land and in the ocean—such as soil moisture, ice cover, sea surface temperature, and sea surface salinity—also change over time, but the underlying processes are much slower than those in the atmosphere. Therefore, when forecasting months in advance, we must rely on the predictive power of the land and oceans.

Forecast intention

Due to these different sources of predictability, forecasts on various lead times also differ in their intention. We're still forecasting the weather, but the shape of the forecasts changes.

On short lead times, forecasts typically communicate insights about very specific events, such as the timing of rainfall or detailed pressure patterns. In contrast, on very long lead times—spanning several years to decades—we enter the realm of ocean dynamics and climate projections.

Then there's the Prediction Desert, which lies between weather and climate. In meteorology, this is referred to as the sub-seasonal-to-seasonal (S2S) domain. An S2S forecast usually provides probabilities for weekly averages, large-scale atmospheric patterns, and potential deviations from normal conditions.

Enter the desert, find the oasis

Traditionally, forecasts at the S2S timescale haven’t been very reliable. You cannot depend on the predictive power of the atmosphere, and navigating the complex interplay between the decaying influence of atmospheric conditions and the rising influence of slower-moving climate processes on land and in the oceans is too difficult to solve with classical weather models. This is the essence of the prediction desert.

But things are changing with the rise of artificial intelligence and the increasing availability of high-quality climate data. In certain locations and seasons, the prediction desert proves not to be as dry and barren as previously thought. Predictability is hidden within the vast array of climate data, much like an oasis in the desert. With our models, we can identify these oases and provide reliable forecasts within this domain.

Crossing the border of meaningfulness

Traditionally - apart from some edge use cases - industries didn't pay a lot of attention to S2S forecasts. Their skill was too low to have any meaning, and thus people simply assumed climatological conditions - i.e. normal conditions - to happen.?

With the current boost in forecasting skill, we're crossing the border of meaningfulness, unlocking a whole new world of use cases in sectors like energy, agriculture, water management, and humanitarian aid. The further we push the forecasting horizon, the more time we provide stakeholders to take anticipatory action. In a world with an increasingly volatile climate, this is far from a luxury—it's a necessity.