Monthly heatwave prediction in Sweden based on Machine Learning techniques with remote sensing data

Abstract: Heatwave events as a kind of extreme climate event, have plagued the human race for the past few years. It severely influences people’s life quality, sometimes even leads to some serious diseases. In order to alleviate the possible damages heatwave events can do, some targeted actions are necessary and forecasting heatwaves is one of them. This study focuses on predicting potential heatwave events in Sweden, replying on the correlations between multiple meteorological and surface-related features, with the help of machine learning techniques. The related remote sensing data of 21 features are extracted and implemented with features selection using a correlation heatmap and 16 of them are finally determined to be used for prediction. Five types of classifiers LR, Gaussian NB, KNN, RF and XGBoost are utilized on the training and validation datasets with hyperparameter tuning and threshold tuning methods to choose the model that has the best performance to predict heatwaves using the test dataset. The results show that RF and XGBoost both perform well on the validation set, but XGBoost is more suitable applying on the test set since XGBoost possesses a higher generality.