Institute of Geography

News

WINTER - Winter clImate predictionN wiTh machinE leaRning

Despite best efforts, reconstructions of past winter climate variability lack either in spatial coverage, temporal coverage or incorporate high uncertainties. Machine learning assisted seasonal prediction could help to solve these issues but has only been applied for recent climate studies. Combining paleo climate, seasonal prediction and machine learning is unprecedented and would allow to close substantially knowledge gaps in climate science. 

The goal of WINTER is to use open access reanalysis datasets together with independent proxy data to predict winter climate anomalies on a monthly to seasonal basis for the past 400-500 years. These climate anomalies will be assimilated into upcoming paleo reanalyses efforts and decrease the uncertainty margin about past winter climate substantially. To achieve this goal WINTER will utilize open source, open access machine learning software on Python basis.