We now discuss possible areas of development of the scientific capability underpinning this climate service (i.e., seasonal TC landfall risk forecasts for East China) ahead of the 2020 typhoon season. These developments aim to provide an improved service of greater benefit to CMA, both in terms of the increased lead time as well as through improvements in the forecast skill.
Following the early evaluation and feedback of the 1 May 2019 forecast in June 2019, it was found that a key benefit to CMA would be a skillful forecast released at a longer lead time to allow more time to prepare and to enable input to the existing institutional procedures (Hewitt et al., 2020). For the TC landfall risk forecast specifically, it is preferable that the initial predictions are released during March to coincide with seasonal forecast meetings held by CMA prior to the typhoon season. This would enable the forecast to be used alongside predictions from the Beijing Climate Center and the Shanghai Typhoon Institute (Hewitt et al., 2020). To examine whether GloSea5 is able to provide forecasts at this longer lead time, we analyze the skill of GloSea5 hindcasts of the WPSH index for JJA 1993–2016 from both a 2- (1 April) and a 3-month (1 March) lead time. Time series of the GloSea5 ensemble-mean WPSH index versus ERA-Interim are shown for both 1 March and 1 April for hindcasts covering JJA 1993–2016, as well as forecasts for JJA 2017–2019, in Fig. 5. The corresponding linear correlations between GloSea5 and ERA-Inte-rim are provided in Table 1 (for brevity this table also includes the skill of forecasts issued on 1 May).
Forecast date 1 March 1 April 1 May Linear correlation 0.72 0.74 0.80
Table 1. The linear correlation between the GloSea5 ensemble-mean predicted WPSH index and ERA-Interim reanalysis for JJA 1993–2016 for forecasts initialized on 1 March, 1 April, and 1 May 2019. For all the three start dates, the skill is significant at the 95% confidence level
GloSea5 shows significant skill for predictions of the JJA WPSH index from both 1 March (linear correlation r = 0.72) and 1 April (0.74), with the greatest skill from 1 May (0.80; all correlations are significant at the 95% confidence level). Nevertheless, it is worth noting that the forecast performance for JJA 2017–2019 from 1 March and 1 April was mixed. In 2018, GloSea5 correctly predicted the below-average WPSH index from 1 March and 1 April; however, in 2017 (2019), GloSea5 underestimated (overestimated) the size of the WPSH index with respect to the 1993–2016 climatology (Fig. 5). Thus, in these two years, the best guidance was provided by forecasts issued on 1 May (Fig. 2). This suggests that even though forecasts from 1 March and 1 April exhibit certain skill and would provide useful information for decision-makers, a further forecast update around 1 May would likely be beneficial to providing the most skillful and up-to-date guidance to users ahead of the forthcoming season.
To address improvements in the skill of seasonal forecasts, we explore two key areas as part of ongoing research for CSSP China: the East China forecast region and underlying statistical model used to forecast TC landfall risk. Firstly, the East China region currently used to provide an indication of the landfall risk (Fig. 1) is large, extending towards both northern coastal China and the Taiwan Island. The eastward boundary of the box is also far offshore, allowing TCs that pass through the region but do not make a direct landfall in East China, to be included in the landfall forecast (e.g., Tropical Storm Danas in July 2019), which is not desirable. Research is currently underway to examine whether adjustments to the East China region (to focus more on the coast of East China specifically) will improve the skill of TC landfall forecasts and make them more relevant for CMA. Finally, attempts are also being made to improve the underlying statistical model used to create the forecast. The statistical model used to predict the TC landfall risk in this study assumes that the historical observed WPSH index and TC landfall frequency in East China are Gau-ssian distributed. However, this assumption does not hold for the TC landfall frequency, which is positive definite and is better approximated by a Poisson process. We have identified that allowing the TC landfall frequency to be modeled as a Poisson process can improve the skill of landfall forecasts, in agreement with previous studies (e.g., Zhang et al., 2017; Zhang and Villarini, 2019). Forecasts of the landfall risk for the 2020 typhoon season will therefore use this improved statistical method to provide a better estimate of the landfall risk for CMA.
Finally, although not considered in detail as part of CSSP China at present, the underlying predictors used to forecast the TC landfall risk for East China could also be re-examined. To date, we have based the seasonal forecast of the TC landfall risk solely on predictions of the WPSH index (i.e., 850 hPa GPH), following the studies of Wang et al. (2013) and Camp et al. (2019). However, recent studies have shown that SSTs in both the North Atlantic (Gao et al., 2018, Zhang and Villarini, 2019) and Pacific Ocean (Zhang and Villarini, 2019) could also provide skillful indicators of the TC landfall risk for East Asia. Tian and Fan (2019) also demonstrated that SSTs in Southwest Indonesia, as well as summer atmospheric predictors (such as divergence and vorticity), were important for skillful forecasts of the TC landfall risk specifically in China. Further consideration of these predictors (either individually or in combination) could be beneficial, as this may further improve the forecast skill and predictability at longer lead times. Examining the role of El Niño–Southern Oscillation (ENSO) on the TC landfall frequency, such as in studies of Wu et al. (2004) and Zhang et al. (2012), could also extend predictability of the TC landfall risk in East China from JJA into autumn (September–November) in the future.