Comparison of Statistical Downscaling Methods for Seasonal Precipitation Prediction: An Application Toward a Fire and Haze Early Warning System for Southeast Asia

Smoke haze from forest fires is among Southeast Asia’s most serious environmental problems and there is a clear need for a long-lead fire and haze early warning system (EWS) for the regions. The seasonal forecast supplied by the APEC Climate Center (APCC) is one of available information can be used to predict drought conditions triggering forest fires in the region. The objective of this study is to assess the skill of the current and downscaled products of APCC’s seasonal forecast of 6-month lead-time for predicting ASO precipitation over the fire-prone regions. First, seasonal forecast skill by six individual models (MSC_CANCM3, MSC_CANCM4, NASA, NCEP, PNU, POAMA) and simple composite model (SCM) ensemble was assessed by considering available each ensemble members. Second, three different statistical downscaling methods including simple bias-correction (SBC), moving window regression (MWReg), and climate index regression (CIReg) were applied and the forecast sill were compared. Both current and downscaled seasonal forecast showed higher predictability over Sumatra regions compared to the Kalimantan regions. Statistical downscaling of forecasts showed the skill improvement over the Kalimantan region where current APCC’s forecast shows low predictability. Study also shows that temporal correlation coefficient (TCC) between observed and forecasted ASO precipitation increases as lead-time decrease.