Best track data from four agencies, the Joint Typhoon Warning Center (JTWC), Hong Kong Observatory (HKO), China Meteorological Administration/Shanghai Typhoon Institute (CMA/STI), and the Regional Specialized Meteorological Center (RSMC) Tokyo are used in this study. TC data in the eastern North Pacific (EP), North Atlantic (NA), North Indian Ocean (NIO), South Indian Ocean (SIO), and South Pacific (SP) are provided by HURDAT, HURDAT, JTWC, Bureau of Meteorology (BOM), and JTWC, respectively, which are used for discussing the global change in TCs. All these data are obtained from International Best Track Archive for Climate Stewardship (IBTrACS) version 3 (Knapp et al., 2010). The Japanese 55-year Reanalysis (JRA-55; Kobayashi et al., 2015) is also used when considering atmospheric circulation changes.

Future characteristics of WNP landfalling TCs are projected using the Meteorological Research Institute’s atmospheric general circulation model (MRI-AGCM3.2s), with ∼20-km (T959) resolution in horizontal and 64 unevenly distributed layers in vertical up to 0.01 hPa. The detailed dynamical and physical schemes settings used in the MRI-AGCM3.2s have been elaborated by Murakami et al. (2012b).

In order to evaluate the impacts of global warming on tropical cyclones and their inland influences, the so-called time slice experiments are applied (Bengtsson et al., 1996). In detail, the present-day experiments are forced by monthly-mean SST and sea ice concentration in the period between 1979 and 2003, which are taken from the Met Office Hadley Centre Sea Ice and Sea Surface Temperature version 1 (HadISST1) (Rayner et al., 2003). When forcing the 20 km MRI-AGCM, the monthly-mean SST and sea ice concentration from 1979 to 2003 are interpolated to daily values and prescribed. Two-member ensembles in the present-day experiments are generated by initiating from different initial time. In order to generate SST and sea ice under the global warming conditions from 2075 to 2099, the averaged SST and sea ice departures between 2075 and 2099 and 1979 and 2003 are calculated based on the 28 Coupled Model Intercomparison Project Phase 5 (CMIP5) models according to the Representative Concentration Pathway 8.5 (RCP8.5) scenario. Superimposing such perturbations onto the present-day SST and sea ice, the future prescribed SST and sea ice are built (Mizuta et al., 2014). Thus, the interannual variation of SST and sea ice at the present day is assumed to retain the same in the future. The global warming experiments consist of four ensemble members, each integrated based on different SST warming spatial patterns, with one from the averaged warming pattern of the whole CMIP5 model SST products and the other three different warming types (i.e., stronger warming signals in SP, Central Pacific (CP), and WNP, respectively) from the cluster analysis using 28 CMIP5 models, which were performed by Mizuta et al. (2014). The CP warming cluster will also be used to investigate the projections of TC inland impacts during the future “CP El Niño” like climate.

The TC tracking algorithm follows that of Murakami et al. (2012b), which has been successfully used in the 20-km MRI-AGCMs model. Using this algorithm, the detected global TC number in the present-day (1979–2003) was found and matched with that observed (Murakami et al., 2012b). In the TC tracking algorithm, a storm-like vortex is detected whenever 1) the maximum relative vorticity at 850 hPa exceeds 2.0 × 10^–4 s−1; 2) the maximum wind speed at 850 hPa exceeds 17.0 m/s; 3) an evident warm core aloft should be found, that is, the sum of the temperature deviations at 300, 500, and 700 hPa exceeds 2.0 K, which are calculated by subtracting the maximum temperature from the mean temperature over the 10° × 10° grid box at each level centered nearest to the location of maximum vorticity at 850 hPa; 4) the maximum wind speed at 850 hPa is stronger than that at 300 hPa; and 5) the time duration of each detected storm must exceed 36 h.

HKO and RSMC use 10-min (V₁₀) average, and CMA/STI the 2-min (V₂) mean, while JTWC uses a 1-min average wind speed at 10-m height above ground (V₁) (Knapp et al., 2010). This discrepancy can cause significant difference in categorizing TCs. In order to maintain inter-agency consistency, all surface wind measurements are converted to 1-min-based data using the relationships V₁₀ = 0.88 × V₁ (Knapp et al., 2010) and V₂ = 0.9V₁. The latter is based on the relation V_CMA/V_JTWC ≈ 0.9 at a stronger intensity level (Kruk et al., 2010). Using these converting algorithms, results for all metrics, that is, landfall TC number (Figure 1A), the annual landfall latitude change (Figure 1B), and the translation speed (Figure 1C), are shown to be consistent among all datasets.

In order to obtain robust TC-related inland impacts on the Asian region, five criteria are adopted in selecting landfalling TCs for both observations and model simulations. 1) Only TCs after 1979 are considered, when reliable satellite imageries were available (for observations only); 2) only TCs with peak intensities reaching the tropical storm class or above (i.e., lifetime peak intensity > 17 m/s) are considered, to exclude very weak depressions; 3) once a TC made landfall, the decay process is only considered when its intensity is greater than 17 m/s, to remove uncertainties related to weak intensity; 4) all TCs should make landfall in mainland East and Southeast Asia (TCs crossing Taiwan or the Philippines are not considered as landfall as they will go back to the ocean and may make second landfall); 5) only landfalling TCs that did not return to oceanic sites again are selected, to make sure the whole decay process occurred over land. Tracks of the selected TCs are shown in Figure 2A (from the JTWC best track data).

Note that criterion (1) is necessary as there are significant interagency discrepancies among various WNP best track datasets due to different practices among agencies and changes in assessment methods over time, especially in assessing weaker systems (below tropical storm strength) before the availability of reliable satellite imageries and related intensity assessment methods in 1980s (e.g., Dvorak analysis) and 1990s (e.g., the advanced objective Dvorak technique) (Knapp and Kruk, 2010; Kruk et al., 2010). Figure 2B shows a clear change of the “final intensity” when HKO issued the last TC warning for each individual storm; there was a discernable difference for the terminal intensity between before and after 1979, suggesting that the observed TC decay process may have larger uncertainty in the early years (before 1979). In fact, all the best track data, except RSMC (which roughly starts from 1975), showed stronger TC final intensity before 1979, as inferred from the probability density function (PDF) changes of issued TC final intensity before and after 1979 (see Figure 2C).

After 1979, final intensities for most TCs are around or below 17 m/s (Figures 2B,C). However, the TC terminal intensity distribution still has a spread in the range in 0–17 m/s. In order to reduce such uncertainty, the storm decay process is truncated when the surface wind speed is less than 17 m/s, hence criterion (3). This approach has been adopted in assessing long-term TC trends (see Lee et al., 2020). Note that RSMC estimated most of the TC final intensities greater than 17 m/s, so this dataset will not be used in the following analyses. After shortening of the TC decay process to <17 m/s, TC final stops issued by different best track datasets are consistent (Figure 2D).

The sustaining time and distance over the land are calculated as the time period and total distance travelled by a particular storm during the considered decay process, respectively. The power dissipation index (PDI) over land is calculated as the time integral of cube of the maximum wind speed during the considered decay period after landfall. Annual variations of indices are calculated by taking the average for all landfalling TCs in the same season.

TC frequency difference in Figures 9A,B is calculated using data aggregated in 2.5° × 2.5° grid boxes. Significance of storm frequency change is assessed by Monte Carlo tests with 100,000 random trials. At least 15 years are randomly chosen from a set of TC frequency data (1979–2003 and 2075–2099); frequency difference from two different sets is then calculated, and the process is repeated for 100,000 times. The increase (decrease) trends pass the 95% confidence level at one grid box if the positive (negative) sign of frequency differences on this grid point appear more than 95,000 times in the trials.

The entire trend analyses used in this study is based on the Mann–Kendall (MK) trend test. Annual variations of the indices are used when doing the tests.

In this study, TCs from 1979 to 2016 and their impacts on the whole East and Southeast Asian inland regions are analyzed based on outlined TC selection criteria. On an average, about 5–6 TCs made landfall in the studied region every year, which is 1–2 more landfalling TCs than that reported by Liu et al. (2020), who focused on the China region. There are marked decadal variations during the last five decades (Figure 1A). The landfalling location also shows secular changes, with a weak poleward shift during the last 50 years, in which the JTWC data passed the 90% confidence level MK test (Figure 1B), consistent with previous investigations (Park et al., 2014; Kossin et al., 2016; Li et al., 2017).

Previous studies have shown a remarkable increase of the TC landfall intensity over the China region (Mei and Xie, 2016; Li et al., 2017; Liu et al., 2020), which may be highly related to the rapid coastal ocean warming over the WNP (Mei and Xie, 2016; Liu et al., 2020). Here, the annual mean landfall intensity in the whole East/Southeast Asian region also shows a remarkable increasing trend of about 1.6 m/s per decade in the JTWC and HKO best track data above the 95% significance level (see the blue and red lines in Figure 3A). However, there is no significant trend of the landfall intensity in the CMA data (see the black line in Figure 3A). Interestingly, it is noteworthy that in addition to possible influence of coastal ocean warming (Mei and Xie, 2016; Liu et al., 2020), landfall intensity is also highly related to the TC lifetime peak intensity change (Figure 3B). All three datasets give a high positive correlation between the peak intensity and landfall intensity (JTWC: 0.56; HKO: 0.65, and CMA: 0.59; see Figure 3C). Moreover, it was found that landfall intensity also correlates negatively with the distance between the position where TC attains lifetime peak intensity and landfall location (referred to as D_pl). Based on multiple linear regression, 1 m/s increase in TC annual peak intensity is related to ∼0.5 m/s rise of landfall intensity on average (JTWC: 0.58 m/s; HKO: 0.53 m/s, and CMA/STI: 0.41 m/s). At the same time, 100 km decrease in D_pl corresponds to ∼0.53 m/s increase in the landfall wind speed (JTWC: 0.6 m/s; HKO: 0.5 m/s, and CMA/STI: 0.5 m/s). The correlation between fitted and observed landfall intensity is much improved when the D_pl factor is included (JTWC: 0.74; HKO: 0.74, and CMA/STI: 0.70; see the scattering plot in Figure 3D). We propose that for larger D_pl, TC will experience longer decay time after its peak and eventually a weaker wind speed at landfall.

In the relationship of the TC landfall intensity and the TC inland impacts, Figure 4A shows clear positive relationships between annual mean landfall intensity and sustaining time over land, with ∼0.77/0.51/0.47 h per 1 m/s increase within the 0.001/0.02/0.05 significance level based on JTWC/HKO/CMA best track data. All the best track datasets utilized in our study, that is, JTWC, HKO, and CMA/STI, display a striking extension of the life span of TCs after landfall, which has increased by 2–9 h since the 1979, above the 95% confidence level (see Figure 4B). Due to higher intensity and longer storm duration after landfall, the dissipated wind power over land, represented by PDI, has also doubled (Figure 4C). It is noteworthy that, despite a global slowdown of the TC translation speed reported by Kossin (2018), the robustness of such decreasing trends is still subject to further studies (Lanzante, 2019; Moon et al., 2019). In fact, if just the TCs in the satellite era are considered, no significant trend of these TCs’ translation speed was detected since 1979 (Figure 1C). If the storm translation speed after landfall remains unchanged, the higher intensity and longer sustaining time can substantially increase the travelling distance over land, meaning more inland penetration of storms. Indeed, on an average, there was an increase of ∼100 km (30–190 km) in distance travelled by TCs since 1979 (Figure 4D). These results are consistent with previous studies, which focused on the China region (Chan, 2008; Chen et al., 2011; Liu et al., 2020).

As the East and Southeast Asia coastline is long, the changes of the TC inland impacts may be spatially different. To further investigate the TC inland impacts on different areas in the Asian region, the Asia land area is divided into three parts, Southeast Asia (latitude in 10°N–20°N), Hanoi and South China (in 20°N–25°N), and East China (>25°N). TCs making landfall in the three regions are then analyzed. First, there was an obvious poleward shift of TCs. A dramatic decrease trend of TC landfall frequency over the Southeast Asian region was detected from all the three datasets in the 0.05 significance level (see Figure 5A1), while an increase of TC landfall frequency over East China and above was found in JTWC and CMA/STI datasets within the 0.05 significance level (Figure 5C1). Second, the TC inland impacts, that is, TC intensity at landfall, sustaining time, and travelling distance over the land, showed different changes in the regions. The Southeast Asia region experienced steady change of TC landfall intensity, sustaining time, and translating distance over land (Figures 5A2–A4). The most significant increasing trends were found over the Hanoi and South China region (Figures 5B2–B4), with all the above parameters passing the 0.05 significance level MK trend test, except the TC landfall intensity change in the CMA best track dataset. In the East China region, although the increasing trend of landfall intensity was detected within the 0.05 significance level in all the datasets (Figure 5C2), the sustaining time and travelling distance over land did not show significant trends except JTWC (Figures 5C3,C4).