Fig. 1.
SST (°C) bias at the forecast lead time (a) 7th day, (b) 14th day, (c) 21st day and (d) 28th day. (e) The SST bias at the 21st forecast day initialised from 1st February. (f), (g) and (h) are the same as (e) except for May, August and November, respectively.
Fig. 2.
Left panels: model’s predicted SSTA correlation at forecast time (a) 7th day, (b) 14th day, (c) 21st day and (d) 28th day starting from the 1st of February, May, August and November over the period 1990–2012. Middle panels (e, f, g and h) are the same as the left panels but for persistence skill. Right panels (i, j, k and l) are the differences between correlation of the model and persistence.
Fig. 3.
(a) Model correlation skill for predicted SSTA at 14th forecast day initialised from 1st February. (b), (c) and (d) are the difference between February and May, August and November, respectively.
Fig. 4.
Left panels: model’s predicted SSTA (°C) root-mean-squared errors (RMSEs) at the forecast lead time (a) 7th day, (b) 14th day, (c) 21st day and (d) 28th day over the hindcast period 1990–2012. Right panels are the same as the left panel but for normalized RMSE using observed SSTA standard deviations.
Fig. 5.
Different domains used for calculating SSTA indices.
Fig. 6.
Model predicted correlations (red solid) and persistence correlations (black dash) for SSTA indices in (a) east Pacific, (b) west Pacific, (c) north Pacific, (d) south Pacific, (e) north Indian Ocean, (f) south Indian Ocean, (g) Leeuwin Current and (h) East Australian Current.
Fig. 7.
Same as Fig. 6 but for RMSE and ensemble spread (green) (unit: °C).
Fig. 8.
Left panels: model’s predicted SSHA correlation at forecast time (a) 7th day, (b) 14th day, (c) 21st day and (d) 28th day starting from the 1st of February, May, August and November over the period 1990–2012. Middle panels (e, f, g and h) are the same as the left panels but for persistence skill. Right panels (i, j, k and l) are the differences between correlation of the model and persistence.
Fig. 9.
Left panels: model’s predicted SSHA (°C) root-mean-squared errors (RMSEs) at the forecast lead time (a) 7th day, (b) 14th day, (c) 21st day and (d) 28th day over the hindcast period 1990–2012. Right panels are the same as the left panel but for normalized RMSE using observed SSHA standard deviations.
Fig. 10.
Left two panels (a) are model’s predicted SSHA in the Tasman Sea for 1st to 28th November 2009 with a 3-day interval, issued 1 November 2009. Right two panels (b) are the corresponding observed SSHA (unit: m).
Fig. 11.
Same as Fig. 10 but for mode forecasts starting 1 August 2003.
Fig. 12.
Model individual SSHA forecasts (unit: m) and sea surface wind stress (unit: N/m2) in the Tasman Sea on 28 August 2003 (28 day lead), issued 1 August 2003.
Fig. 13.
(a) The spatial correlation between the model prediction and observed SSHA in the Tasman Sea (the domain as per Figs. 10–12) at forecast day 1 (black), 7 (red) and 14 (green). (b) The model averaged spatial correlation (red line) at forecast time from day 1 to 28 over the period 1993–2013. Dots indicate forecast correlations for model start dates. There is a total of 80 cases at the same lead time. (c) Same as (b) but for RMSE (unit: m).