Forecasts of ENSO evolution using spatial–temporal projection model. (23rd April 2020)

• Record Type: Journal Article
• Authors: Pan, Xiao; Zhu, Zhiwei; Li, Tim
• Is Part Of: International journal of climatology. Volume 40:Number 15(2020)
• Page Start: 6301
• Page End: 6314
• Publication Date: 2020-04-23
• Subjects: ENSO evolution -- ENSO prediction -- spatial–temporal projection model; Climatology -- Periodicals; Climat -- Périodiques; Climatologie -- Périodiques; 551.605
• DOI: 10.1002/joc.6581
• Languages: English
• ISSNs: 0899-8418

Abstract

Abstract: El Niño–Southern Oscillation (ENSO) prediction is one of the most debated and challenging tasks, whilst its real‐time operational prediction skill still has room for improvement. In this study, spatial–temporal projection model is applied to predict Niño3.4 index at lead time of one to six months. By regressing variable fields onto Niño3.4 index month by month, physical‐based predictability sources, that is, the mixed‐layer oceanic temperature, sea surface temperature, thermocline depth, and accumulated westerly‐wind‐events index over the specific regions are detected as the predictors. Based on the temporal evolution of coupled modes of predictors and Niño3.4 index, the Niño3.4 index from JAS (July, August, and September) to DJF (December, January, and February) can be predicted once a year. The model could nicely reproduce the evolution of Niño3.4 index from JAS to DJF. It also achieved high prediction skills for the year‐to‐year DJF Niño3.4 index, with a root‐mean‐square error of 0.46 in the training period (1950–2000) and 0.52 in the independent‐forecast period (2001–2016). Further investigation shows that the forecast is more reliable when the forecasted ENSO amplitude is much larger. Abstract : The spatial–temporal projection model (STPM) has been applied to predict Niño3.4 index from one to six months ahead. The model can nicely reproduce the evolution of Niño3.4 index from JAS to DJF. It also achieved high prediction skills for the year‐to‐year DJF Niño3.4 index with a root‐mean‐square error of 0.46 in the training period (1950–2000) and 0.52 in the independent forecast period (2001–2016), outperforming persistent forecasts.