Thursday, 1 February 2024
Hall E (The Baltimore Convention Center)
The skill of NWP precipitation forecasts deteriorates with lead time. Hence, there is a need for correcting these model forecast errors towards improving the quality of products provided by climate services. We describe the calibration methods used to correct biases in the GEFS Week-1 to week-4 and week 3-4 precipitation forecasts over Africa. They include Canonical Correlation Analysis (CCA), Extended Logistic Regression (ELR) and Extended Probabilistic Output Extreme Leaning Machine (EPOELM). We use Generalized Relative Operating Characteristic Score (GROCS) and reliability diagrams to evaluate the performance of the GEFS based on selected cases over the 2000 to 2019 hindcast period. We also compare the calibrated forecasts with the GEFS raw forecasts for selected recent extreme wet and dry events. RPSS and Heidke Skill score are used to verify forecasts during these events.

