Seasonal forecasts of Eurasian summer heat wave frequency

Abstract

Abstract Heat wave events usually cause disastrous consequences on human life, economy, environment, and ecosystem. However, current climate models usually perform poorly in forecasting heat wave events. In the current work, we identified that the leading mode of the summer (June-July-August) heat wave frequency (HWF) over the Eurasian continent (HWF_EC) is a continental-scale pattern. Two machine learning (ML) models are constructed and used to perform seasonal forecast experiments for the summer HWF_EC. The potential predictive sources for the HWF_EC are chosen from the fields related to the lower boundary conditions of the atmosphere, i.e., the sea surface temperature, snow cover, soil moisture and sea ice. The specific regions and months of these lower boundary condition fields selected to construct the potential predictors are those that are persistently and significantly correlated with the variation in the HWF_EC preceding the summer. The ML forecasting models are trained with data from the period 1980–2009 and then used to perform real seasonal forecasts for the summer HWF_EC for 2010–2019. The results show that the ML forecasting models have reasonably good skills in predicting the HWF_EC over high HWF regions. The two ML models show obviously better skill in the forecasting experiments than a traditional linear regression model, suggesting that the ML models may provide an additional and useful tool for forecasting the summer HWF_EC.