One of our projects, Seasonal Forecasting Engine, recently published their June forecast based on five numerical weather prediction models. The June forecast was remarkable in that it predicted almost no temperature anomalies in Northern Europe. When we sum up in July, we will probably find that some regions experienced either warm or cold anomalies, but there was no consistent signal across the models this time.
In Climate Futures, we will make use of advanced statistics and machine learning to create algorithms that improve the skill of forecasts going from 10 days to 10 years into the future. The key will be to learn how different climate variables interact and then to use that knowledge to make better use of the physical models. Erik Kolstad, the PI of the SFE project, was the lead author of a 2017 article on the different roles that snow cover, soil moisture and soil temperature play in dictating the persistence of temperature from month to month.
A key figure in that paper showed which of these variable were the most important mediator of persistence in winter and summer:
At midlatitudes, snow depth plays an important role. This means that when there is a lot of snow on the ground, the temperature will probably stay anomalously cold from January to February. Or if there is thin snow or no snow at all, the temperature will be anomalously warm in both months. The specific role of snow was examined in a follow-up study. In summer, soil variables dominate completely.
Our long-term goal is to assign a higher weight to the models that represent these physical mechanisms well while assigning a lower weight to less realistic models. To be able to plough through the huge data amounts available from models, observations and satellite imagery, we need to let computers do the heavy lifting. Another important reason not to do this manually is that humans are prone to making mistakes, often in the form of being too subjective. Machines, for better or worse, do not have this bias, which we believe makes them ideally suited for this work.
Stay tuned for our July forecast and other developments!
Yesterday, we submitted an outline for a proposal for a new Centre for Research-based Innovation. With 16 user partners and 6 research partners, this was the first step towards a full proposal to be submitted in September 2019. Here’s the summary:
Climate Futures will create usable and reliable climate forecasts that do not exist today. Short-range weather forecasts are already invaluable tools for planning ahead, but there is a clear need for climate information beyond the next 10 days and up to decades into the future – the subseasonal-to-decadal (‘S2D’ hereafter) time horizon. For instance, hydropower companies make crucial decisions based on assumptions about future rainfall, snow accumulation and heating demand. Insurance companies would save large sums if they could prepare for cold spells, floods, tropical cyclones and droughts. Farmers need to know when the growing season starts, how much it will rain and how warm or cold it will be, and when to harvest.
Climate risks are manifold and escalating, but there is a critical shortage of tools for managing them. We will capitalize on recent scientific advances in climate prediction and artificial intelligence (AI) and develop new products to radically expand the use of S2D forecasts in Norway. The products must be easy to use, relevant, dependable and scalable. This requires cooperation across disciplines and sectors and innovation on a grand scale, which our consortium is uniquely positioned to accomplish. By bringing together some of the largest energy and insurance companies in Norway, leading agricultural actors, a global provider of weather intelligence, public authorities, world-leading climate scientists and outstanding expertise in weather forecasting, economics, statistics, AI and co-production, we will make climate forecasts available for everyone, to benefit companies, individuals, organizations and policy-makers.