4.1. Control experiment of Mindulle
The CE experiment of Mindulle (2004), which starts at 0000 UTC 25 June 2004 and spans for 9 days using NCEP FNL data as IC/BCs (Table 1), is first presented and verified against the observation in this section. In general, the simulated track agrees with the Japan Meteorological Agency (JMA) best track, with a maximum track error of about 150 km near 0000 UTC 26 June (Fig. 3a). However, the model TC makes landfall in Taiwan slightly too south. As for TC intensity, the model storm develops from 1000 to about 970 hPa in its central SLP and from 13 to nearly 40 m s-1 in maximum surface wind speed on 29 June (Figs. 3b–c). Thus, the model storm is not as intense compared to the JMA best-track estimates of 940 hPa and 50 m s-1. Part of the reason, presumably, is that the TC in the initial field is weaker than the observation, by about 20 hPa and 17 m s-1, whereas the model storm also does not intensify as rapidly as in the observation on 27 June. Nonetheless, the tendency that Mindulle intensified up until it approached Luzon and weakened afterwards is captured by the model. Since the focus of our study is on how past climate change affects typhoon rainfall rather than its intensity, whereas the intensity errors reduce after 30 June when Taiwan receives heavy rainfall, such a discrepancy in the simulated intensity is considered acceptable.
Figure 4 shows the observed cloud structure of TY Mindulle (2004) by the Aqua satellite at four selected times with better coverage (left column), and its comparison with model simulated hourly rainfall and other relevant fields in CE when the TC moves to a similar location during days 5–8 (right column). Although the two quantities are not identical, many features of Mindulle during its lifetime can be reasonably well captured by the model. These include an asymmetric cloud and rainfall pattern mainly to the south and southwest of the TC, as well as an elongated eye when Mindulle was near Luzon and south of Taiwan on 29 June (Figs. 4a and 4e). On 1 July when Mindulle was close to landfall in Taiwan, the cloud/rainfall pattern became even more asymmetric, with cloud/rain bands extending from its east to south, and then to southwest (Figs. 4b and 4f). Later on 1 July, such elongated cloud patterns in the wake of Mindulle continued, with little deep convection around the eye (Figs. 4c and 4g). The cloud/rain bands became oriented more from north-northeast to south-southwest around 1800 UTC 2 July (Figs. 4d and 4h), when a series of MCSs developed near southern Taiwan to bring heavy rainfall (also Chien et al. 2008). All these above features are reproduced quite well, even at a range of 5–8 days (Fig. 4).
The rainfall observation in Taiwan from a dense network of some 400 gauges (Hsu 1998) provides an unequaled and unequivocal way to verify the CE results, particularly when the rainfall is our focus. In Figs. 5a–b, the modeled five-day total rainfall (from 29 June to 3 July) is seen to match the observation quite well, and even with some over-prediction despite the weaker intensity. As observed, the rainfall is mainly in the southern two thirds of Taiwan, with peak amount in the mountain interior in southern Taiwan. Thus, the extreme rainfall associated with TY Mindulle and the southwesterly flow at its wake is successfully reproduced realistically in CE by the CReSS model at a high resolution.
4.2. Sensitivity experiment and change in rainfall
In SE where the D values are subtracted from the IC/BCs and TY Mindulle is essentially placed in the background of past climate roughly 40 years ago, the modeled track and intensity are quite similar to those in CE without significant differences (Fig. 3). In Fig. 3a, the TC in SE moves slightly farther to the west, especially in early July when it tracks north. This is consistent with our earlier assessment that the long-term change in mean circulation would allow the modern TC to approach Taiwan from the east more slowly (Figs. 1a–b) and accompany a stronger southwesterly flow during departure (Fig. 2a). The TC intensity in SE is very similar to CE in minimum SLP, but somewhat weaker in terms of maximum wind speed (Figs. 3b–c). Overall, the track, intensity, and evolution (not shown) of the storms in CE and SE highly resemble each other. As pointed out earlier, since the D values are only a fraction of the total variables in magnitude (in IC/BCs), this overall similarity is expected but allows us to attribute the difference in rainfall in the two experiments to mainly the long-term climate change. Such an attribution would not be possible if bifurcation occurs and the pair of storms evolve very differently.
From hourly outputs of CE and SE, the time series of averaged rainfall inside radii of 200–500 km from the storm center can be constructed and compared (Fig. 6). These series show rapid temporal variations in response to the development and dissipation of TC rainbands and MCSs in individual runs, especially inside smaller radii, and thus need proper averaging in time to better deduce the result for discussion. Therefore, the mean daily rainfall (per 24 h) inside different radii (200–500 km) for the entire 9-day period of integration and those inside 400 km for each day in CE and SE for Mindulle are summarized in Table 2.
For the 9-day period, the averaged 24-h rainfall reduces when a larger radius is used (cf. Fig. 6), since the fraction of areas with significant rainfall decreases outward and farther away from the storm center (Table 2, left). Inside 400 km, the mean daily rainfall in CE (63.6 mm) exceeds that in SE (58.5 mm) by 5.1 mm and the highest percentage at 8.0%. If consider the first two days as model spin-up and recalculate for the remaining 7-day period, the percent increase in rainfall goes up to 10.2%. Here, the percent changes are computed as the change from SE to CE relative to CE [i.e. (CE - SE)/CE], since the modern-day case is our benchmark for comparison. From the third day of 27 June, the storm in CE also consistently produces more rain than its counterpart in SE, by 3.6–16.3%, including the most intense and rainy period of 28–30 June and the heavy rainfall period in Taiwan of 1–3 July (Fig. 6). The rainfall during 28–30 Jun occurs mainly over northern Luzon Strait and Bashi Channel (Figs. 3a, 4a–b, and 4e–f). The higher amount in CE is likely due to the stronger low-level confluence of the northerly flow component in D values (Figs. 1a and 2a) with TC circulation. The northerly component in D values is consistent with the results of Wang et al. (2013a), who suggest a strengthening of Northern Hemisphere summer monsoon during the past decades. For the heavy rainfall over Taiwan during 1–3 July, as mentioned previously, it rains continuously in central and southern Taiwan (Figs. 5a–b) even after TY Mindulle had weakened and moved to more than 500 km away (Fig. 3a). This rainfall at long distances is not accounted for in Table 2 (using 400-km radius), where a reduced TC rainfall is shown since about 1200 UTC 1 July (also Fig. 6). For the five days from 29 June to 3 July, Figs. 5b–d also show that Taiwan and much of its surrounding oceans receive more rainfall in CE compared to SE, particularly along the western plains (by ³400 mm in total amount). The result of excessive rainfall in CE is in line with the slightly eastward track of modern-day Mindulle compared to SE (Fig. 3a), indicating a weaker steering flow in CE as a result of mid-level westerly component in D values (Fig. 1b). Therefore, the SE produces more rain over the Taiwan Strait (blue area in Fig. 5d), a result apparently due to a track farther to the west (cf. Fig. 3a).
4.3. Water budget analysis for Mindulle
As described in Section 2.3, to investigate on the possible source of more rainfall in modern Mindulle compared to its counterpart in past climate, water budget analysis for a cylindrical volume that moves with the TC is performed. The results using a radius of 400 km and averaged over 6 days (28 June to 3 July) are presented in Table 3, where the first 3 days of model spin-up is excluded. In both runs, the total rainfall (P, 2.4–2.6 mm h-1) comes mainly from the inward convergence of vapor flux (CVF) through lateral boundary (60–66%), and to a lesser extent from the drying of the air column (TDC, ~12%) and local evaporation off the ocean underneath (E, 10–11%). In CE, the convergence of hydrometeor flux (CHF) also accounts for about 5%, while the residual term R is roughly 7% of P, which is considered reasonably small. For the CVF, the convergence of water vapor has large positive contribution (2.0–2.2 mm h-1) as expected, while the advection effect (ADV) is negative by bringing in generally drier air from the surroundings at low levels (Table 3).
Compared to the storm in SE, the modern-day Mindulle in CE has more precipitation by 9.6% over the 6-day period (Table 3), and this is contributed by a stronger convergence of vapor flux (by 16.7%), owing to a considerable increase in CONV of 11% and, as a minor factor, a reduction in the negative effect from ADV (by -8.1%). The increase of CONV in CE is mainly in low-level horizontal convergence (+15.7%), which implies a more active transverse circulation of the TC, and to a lesser degree in moisture amount (+4.6%). Thus, the above water budget analysis in Table 3 suggests a stronger secondary circulation in the present-day Mindulle, when its climatic background has become slightly warmer and wetter from about 4 decades ago. Consequently, more overall rainfall is resulted (Table 2).