Improvement of CGCM prediction for wet season precipitation over Maritime Continent using a bias correction method

Abstract

ABSTRACTA new model output statistics method – Ensemble Selective Simple Linear Regression (E‐SSLR) – is developed based on SLR in order to increase the seasonal prediction skill of a Coupled General Circulation Model (CGCM) over the Maritime Continent (MC), a region with large model simulation errors. E‐SSLR is applied to Pusan National University (PNU) CGCM hindcast over the MC region for the period of 1981–2010 to reduce the systematic model bias in boreal winter (DJF) seasonal mean precipitation and outgoing long‐wave radiation (OLR) anomalies. Three oceanic indices (Nino 3.4, El Nino Modoki and Indian Ocean Dipole (IOD) Mode indices) and one atmospheric index (Southern Oscillation Index, SOI) produced from PNU CGCM hindcast are used as SLR predictor. E‐SSLR consists of three steps: Selection, SLR and Ensemble. The selection and ensemble steps are added to the conventional SLR step to overcome the weakness of the linear regression method. In the selection step, the grids with a temporal correlation coefficient between predictor and predictand exceeding the threshold are selected. These grids (grid‐selected) are corrected by SLR in the second step. For the grids that are grid‐not‐selected, the original CGCM results are used without further correction. This prevents insignificant statistical correction due to the application of low correlated predictors to the SLR. The correction effect of E‐SSLR is analysed in terms of deterministic and categorical analyses. The result shows that the seasonal predictability of DJF seasonal precipitation and OLR in the MC region is increased by using E‐SSLR, and this increment is statistically significant. The correction effect is larger when indices with high predictability that are closely correlated with the predictand are used as predictors.