Fig. 2: Dust event forecasting performance over the validation set for increasing lead time.
From: Deep multi-task learning for early warnings of dust events implemented for the Middle East
Presented are the recall and precision of \({\Phi}_{loc}\) over all months of the year (solid blue and orange lines) and between December to April only (dashed blue and orange lines); the precision–recall average of a Shallow classifier composed of an extreme gradient boosting algorithm (XGBoost) operated on the entire flattened input data (dash-dotted red line); the recall and precision (which are almost identical) of a Naive classifier (persistence) which essentially returns the last observed PM10 level (dash-dotted green line); and the event rate (dotted black line), represents the precision of a classifier that constantly forecasts a dust event.
