LEAS Cuts 5‑Week Temperature Forecast Errors by 10 %
Get the Health newsletter
Daily health & science — research, biotech, public health, the studies worth knowing. Free.
- Institute of Industrial Science (collaborating with George Mason University) developed a new post‑processing technique that improves air‑temperature forecasts up to five weeks ahead without extra model runs.
- LEAS (lagged ensemble analog sub‑selection) works by retaining only past ensemble members that showed high predictive skill, allowing existing S2S forecast systems to be upgraded without increasing computational cost.
- North America saw forecast error reductions of roughly 10 % and better extreme‑heat prediction when LEAS was applied to hindcasts from four operational S2S models across lead times of one to five weeks.
- Paul Dirmeyer emphasized that atmospheric and land‑surface memory makes “previous forecasts not outdated,” and that grouping the best‑performing members can boost skill without rerunning the model.
- Dr. Daisuke Tokuda noted that the simple selection strategy performed consistently across all four independent forecast systems, surprising the team with its robustness.
Why it matters: Forecasting agencies gain a 10% error reduction in North American temperature forecasts, enabling more reliable extreme‑heat warnings and more accurate medium‑range planning, all while avoiding additional supercomputing costs that would strain limited budgets. This efficiency helps keep forecast development costs stable as computational demand is already high.
